Don’t talk to me, talk to my agent: Academics and the agentic revolution

Good afternoon, colleagues . Um,

. Um, Good afternoon, colleagues. Um,

t numbers are still rising, but w a have a full agenda this afternn Um, and I think Um, and I think we should actuay t and get ly start and get the proceedingg going. So, on behalf of the Acy

going. So, on behalf of the Acy h demy of Science of South Africa, Um good afternoon Um, good afternoon, and a very arm welcome to everyone joinings We sincerely We sincerely appreciate your ti , and your e, your interest, and your partn t cipation in this important conv.

So today 's webinar marks our So today's webinar marks our se a , as we m enth AI Hotla, as we name them, successful following six very successful e That have been brought together That have been brought togethery Um, different voices across Um, different voices across the and innovation scholarly policy and innovation reflect.

reflect.

On the evolving implications of On the evolving implications ofl So this webinar forms part of So this webinar forms part of t m ation Act e Digital Transformation Activin l arly Publish ies within ASF scholarly Publis.

And my name is then Susan Feltz And my name is then Susan FeltzI And my name is then Susan FeltzS ann, I manage of the Scholarly ublishing Program at the Academ Africa.

Africa.

So the Ayal So the Eyalchotra series So the Eyalchotra series aims t s trending and debate on trending AI topic While also informing While also informing the scholay t emerging ly community about emerging devs lopments, opportunities, and ch s very rapid y rapid and evolving field.

field.

As AI continues to reshape, um research research education publishing, And society more broadly, Plat forms of these are becoming Platforms of these are becoming t extremely important for collecte critical ve learning, critical engagemen e

exchange.

exchange.

So , we encourage So, we encourage your open part Today a thoughtful discussion Today, a thoughtful discussion And we hope today be both insightful a to all.

to all.

Um I would like to just um Um, I would like to just, um, w lcome then all our panel member forever grateful . We are forever grateful that u

. We are forever grateful that u s afternoon.

afternoon.

But before I do so But before I do so, I just wanto n the most t to mention the most important p e actical matters, like, please k s muted, Please use the chat function.

function.

And note that we are recording And note that we are recording .

And we do hope that you don't And we do hope that you don't hy against ve any objections against this So , this afternoon, we So, this afternoon, we actuallye have an incredible lineup of spk .

Wanyana, Mr. Herman Black ie Professor Mr. Herman Blackie, Professor S, Professor Aslam Fatahar Professor Aslam Fatahar, and th o hand them all n I am going to hand them all or capable er then to our very capable han .

Terence Von Sau , who will be Terence Von Sau, who will be our r um this facilitator for, um, this webi And Profs And Prof, I'm handing over to And Prof, I'm handing over to y And Prof, I'm handing over to yc u, and please introduce our spe kers to, um, the participants. u

kers to the participants. Thank.

Excellent .

Excellent excellent . Hope

. Hope Excellent, excellent. Hope you e

Excellent, excellent. Hope you e y . Very

. Very ll hear me clearly. Very excitet

Opportunity for us to Opportunity for us to actually s is iscuss what I think is probablyf topics one of the foremost topics in tf now This idea behind artificial This idea behind artificial inte lligence and how far we're goine . And I think

. And I think to take this. And I think sortf a moment is t December this year, which is December this year, which is sc ry to think that you're having d watershed moment every few mons When we really When we really were introduced a of t ics o this idea of a genetics agentI So I'm quite excited

about this.

about this.

So I'm quite excited about thise d to to I'm quite excited to to hear a s viewpoint f out the various viewpoints of os y r participants, see where they' I think the way I would like us I think the way I would like uss d , because we to take this forward, because w s 're going to have presentationsf by each of our participants whi

e they're busy loading up theirs l read their slides, I will read their bios, sort of introduce them during tt at point just before each one os So it's kind of fresh in your So it's kind of fresh in your md nd who the speakers are, but ju y we from across Individuals from across academim viewpoints , from across viewpoints on thic e 've got some topic. We've got some hype ment

topic. We've got some hype ment atists we've got some pragmatists, opm imists In the panel , and we're well, I don't o gett well, I don't know if ane of them are pessimists, but hoy t somee some efully we'll get some good view l of these s .

from all of these individuals.k

t

So I think just sort of setting e and my kind of feeling the stage and my kind of feelins The space Back in about 2003 Back in about 2003, well, preci y 200 into .

ely 2003, I walked into Prof. oe

She s then to be my fice. She was then to be my mas

fice. She was then to be my mas r And I said to her I went to study artific I went to study artificial inte e specifically And she said to me okay .

And she said to me, okay, that . And there was this

. And there was this ounds good. And there was this e

ounds good. And there was this e s called hing at the time, it was callede g .

At that time.

At that time, we thought the on y way this was ever going to ha we were going pen is we were going to have tos manufacturing use this additive manufacturing Nowadays.

Nowadays.

over the internet. Nowadays, evy

t hold of a copy rybody can get hold of a copy o n their this and put it in their room,r y But then it was really But then it was really cool, soe we were like, well, we're goings t to do this additive manufacturi e g, and we're going to introduces iating this resource negotiating agente t . And we were looking at platfo

. And we were looking at platfo ms like the Java Agent Developmt t .

nt Environment. There was a thi agent n g called agent communication la FIPA ACL Whole lot of pieces Whole lot of pieces that we werg .

In any case pulling together. In any case,

pulling together. In any case, f this is the story of all of this is that a half years about two and a half years lat.

I handed As A single agent that could A single agent that could active a print, a 3D r over te a print, a 3D printer over tt And I was a warded a master's And I was awarded a master's fo.

And I was awarded a master's fon that, a master's in computer s d the reason ience and AI, and the reason whm I'm smiling a little bit, becae like a few se I feel like a few of our pan people in this lists and many people in this rd e How the did somebody get and Masters for what I could and Masters for what I could efy right now ectively do right now over the d

Many of you would be able to Many of you would be able to ta a the current tooling that we've the current tooling that we've ing that we ot, the agentic tooling that wee t this app over have and built this app over thd at that time It was a it It was a different It was a different a different e And so As you can guess , I don't have

As you can guess, I don't have lot of thunder yet to steal fr k ers. I think

ers. I think m our speakers. I think our spes r with kers are very familiar with thef . And I think

. And I think some e're going to get some nice vie in terms t this points in terms of how fast thi has moved forward . I

. I want to, I has moved forward. I want to, Id want to read one little quote be And this is AI CEO Mustafa Suleiman .

AI CEO Mustafa Suleiman. I thin

or so He said this , and , he said He said this, and he said, I thk to have a nk that we're going to have a h e on man-level performance on most, l tasks So white collar work So white collar work where you' a e sitting down at a computer, er er or an

ther being a lawyer or an accout manager tant or a project manager or a arketing person, I think some p an Most of these tasks will Most of these tasks will be ful d by AI y automated by AI within the ne to 18 .

When I read a statement like When I read a statement like th t puts my I immediately want to deny that That's the reality But On another level.

level.

On another level, when I look On another level, when I look aw d ly transition .

happened from, you know, when 's to the fact did my master's to the fact th y t someone could effectively ach master's, I Within a day less than a day Within a day, less than a day, e e omeone could achieve way more tn over an what I did over two and a ha s In the current environment, we In the current environment, we .

ind ourselves in, there definity need for us to increase our need for us to increase our awa Increase our caution around Increase our caution around the space we're about to find ourse Do I believe Sule iman?

Do I believe, Suleiman? On somes

that levels, I think that this this n this levels, I think that this is myt a person is feeling on this that a person ih not a bunch of tasks, but we ha e do work And so you might be And so you might be able to aute e tasks mate some of the tasks but you e e won't be able to remove the pur l

I think at this point, what I think at this point, what I wr to hand uld prefer to do is to hand ove r panelists to some of our panelists, givem to them an opportunity to talk us h , and we're hrough this, and we're actuallyt e quite fortunate in the sense th t panelist Ms. Wanyana is

somewhat of an expert in the is somewhat of an expert in thed agents d background behind agents and ag e got to where we ntics, how we got to where we a ntics, how we got to where we ad e today, actually. And so it'lle t be quite exciting for us to sorl how technology has progressed technology has progressed over s he last X amount of years, and w

s in ow we find ourselves in this sp e where we are cific space where we are right w such ow, where we have such a tropic fold.

l discussion to unfold. So, may

This one, Yana .

This one, Yana, you can Take over the presentation Take over the presentation, and that while you're doing that, I can e you, if while you're doing that, I can t Okay .

Okay so our first speaker is Ms Okay, so our first speaker is M She's an assistant lecture r She's an assistant lecturer in e Department of Computer Sci Department of Computer Science t the Faculty of Computing and t ics r a University nformatics, Mambara University n

ology f Science and Technology must. e

PhD candid e he's also a PhD candidate in th e Department of Computer Sciencee f Cape Town at the University of Cape Town, where she recently submitted he He says, you're on the edge He says, you're on the edge, yon f Get ting that title change .

Her Getting that title change. Her

cognitive esearch centers around cognitiv esearch centers around cognitiv t esearch centers around cognitiv.

, agent architectures, ontologi very exciting s s. These are very exciting word

s. These are very exciting word to those of us that have been e to those of us that have been g n the space for a long time. On

And neurosy mbolic AI And neurosymbolic AI, with a st future of ong focus on the future of systs reason ms that can reason, interact, a g y.

d act with increasing autonomy.e

She's also the founder of Wanya c a, a company a.africa, a company focused on e

a.africa, a company focused on e f he custom development of intellt For real world use cases . And so

. And so For real world use cases. And s

yana I hand over to Miss Juanyana, nd she can take us through what s agentic t exactly is this agentic AI thate talking about Thank you so much Professor Thank you so much, Professor Te Um in the next round , about Um, in the next round, about 10 talking about minutes, I'll be talking about e

m What agents aree what agentic AI is , and I will what agentic AI is, and I will e s of agentic tate some patterns of agentic A I found.

found.

And then I'll conclude .

Um a good place to start is Um, a good place to start is in And we all know by now that AI And we all know by now that AI t .

It has gone through years of the It has gone through years of thl d technical oretical and technical advancemt And it has also seen And it has also seen some winte.

Uh, but with each surge Uh, but with each surge that we s little graph can see on this little graph thd Yogeshi's paper We see more machine and greater ambition dence and greater ambition has .

So if you look at the So, if you look at the foundati nal era, well, we're mainly tal ing about, or they were mainly t reasoning alking about symbolic reasoningr And maybe some abstraction And maybe some abstraction for s That era also gave us some That era also gave us some behat for machine intelligence um for machine intelligence, um, w

s ich we keep referring to as theg up to today.

And that error was And that era was followed, by export systems in the 60s by export systems in the 60s, ll Um these were Um, these were rule-based syste they were knowledge-based s, they were knowledge-based, td inference ey had some form of inference e.

And then we had some sort of AI And then we had some sort of AI the winter after that, and the wints havee Or the teams have over-promised Or the teams have over-promised o and under-delivered. And so, at

and under-delivered. And so, at e realized that hard-coded or hard -engineered um, knowledge knowledge is not what knowledge is not whatt what hel the s us realize the capabilities tt .

And so , at that point Um um, adapt ive systems um, adaptive systems started toe the conversation , and that gave the conversation, and that gave birth to what we havee we had, in the 90s r what was there in the 90s, whs Intelligent agents intelligent agents. So, the

intelligent agents. So, the intelligent agents. So, the int

intelligent agents. So, the int lligent agent paradigm at that t started in the late 80s , and started in the late 80s, and uhy early 90s, focusing on percept on Riesling on action, And these systems were built And these systems were built ar s like the und architectures like the beli t ions fs, desires, and intentions arc

Um the sole cognitive Um, the sole cognitive architecd ure, and we see these architect y because Um some agents are being built Um, some agents are being built, Um, some agents are being builtf on variations of these architec And then later , we see an And then later, we see an excit ical g

.

And not ably, the And notably the transformer And notably the transformer arc e birth to itectures that give birth to tha right now era that we are in right now, the agent ic AI era.

era.

So we've all witnessed this So, we've all witnessed this th.

where we are applying um where we are applying, um, main models, but I y large language models, but I hould say that it's not the onle But it's mainly large language But it's mainly large language flex odels because of their flexibil , In tool use, in um single and In tool use, in, um, single and

multi-agent environments, for ue e to be able to realize very, ve x objectives e y complex objectives that we th , Um if Um, a few years back.

back.

I would like to suggest that I would like to suggest that ma y say be we should actually say moderc e agentic AI, because, like we sm , and c e from this graph, and agentic since I systems have been around sincs And the capabilities of agent And the capabilities of agent-s c AI that we are talking about t

have been uh the same that have been, uh, the same that han t .

Um uh, a decade or more ago.

ago.

So at most, there is a liter So, at most, there is a liter t , Um in this tool Um, in this tool, prospective ss t students udents and current students, it k , too, so udents and current students, its s a reactive rank 2, so studentn find information t can, um, find information abou.

And it has significantly reduced And it has significantly reducee s and emailing phone calls and emailing inqui Especially during eras Especially during eras like now g to apply when people are trying to appls And um, it has a And, um, it has a knowledge man is constantly

gement system that is constantld designated updated by designated officers.

Um and And um its reliable to And, um, its reliable to a goodt Uh, because it bases all Uh, because it bases all its ant e wers on what is in the knowledg.

Otherwise , it will say Otherwise, it will say, does it no, and directs the person to tt .

But Imagine if Misa is able to Imagine if Misa is able to indey , um, which s endently tell, um, which course Apply for uh those courses Apply for, uh, those courses, o this, uh, go ahead and help this, uh, jul application t fill in the application form y through all the necessary admi .

With full autom ation , and maybe With full automation, and maybe m the minimal intervention from the s, Instead of merely answering Instead of merely answering que t what can I , tions about what can I do, wher d Um what is the Um, what is the procedure for ts

Imagine if is? Imagine if MISA can go ahea

is? Imagine if MISA can go ahea .

e are calling That is what we are calling age right now the tic AI right now, where the syse of autonomy or a good degree of of autonomy, or a good degree o.

And this, this And this, this does not just apy tools like , There's so many areas There's so many areas where peoe g for tools le are looking for tools that n answer questions But can independently plan But can independently plan and e tasks?

tasks?

Um for example, AI Um, for example, AI research as Um autonom ous customer support Um, autonomous customer supportr s .

Long ago , maybe if you Long ago, maybe if you've stumbd r science class ed into a computer science clas as a that had AI as a course, you m n this.

this.

Uh, but long ago, Stu Uh, but long ago, Stuart and Nok s a simple ik gave us a simple definition t s .

Anything that interacts with Anything that interacts with th h sensors s environment through sensors ad reacts or act s towards reacts or acts towards this env t And there's , of course, And there's, of course, some pr e in between, so cessing there in between, so it just s not mainly about just taking d n something and bringing out sog

And if you read And if you read this classic, y they also u realize that they also propos agent programs d some agent programs. Agent prt e The interesting one, which The interesting one, which is t d goal-based That shows us that goal-directed That shows us that goal-directe t behavior did not start yesterd, t ed behavior

y. So, goal-directed behavior i

y. So, goal-directed behavior i t one of the things that we talk e say we talk .

And it's been around for um, decades .

They also talk about what kinds They also talk about what kindsf agents, uh, of environments agents, uh, par of environments agents particip And an interesting one And an interesting one that we g y is re talking about today is the se the multi ngle agent versus the multi-age m task environment , where we see task environment, where we see g competitive

gents being competitive or colle to achieve um certain tasks.

certain tasks.

Um I just single Um, I just singled out that one hear about because it's what we hear about t ic AI .

era .

So what exactly do people So, what exactly do people mean y agentic when they say agentic AI system These are systems that are These are systems that are fullm , Um they operate independently.

independently.

Over extend ed periods of time .

And they use various tools And they use various tools to ah r tasks.

tasks.

And right now, if you look at And right now, if you look at l c AI, terature on agentic AI, they're drawing a line between traditiol , And agent ic AI systems, and And agentic AI systems, and they How much inter vention the human How much intervention the humang . So

. So is putting in. So, with traditi agents, , There are some human-defined There are some human-defined pa m may be well-scop ed tasks , but with Agentic ngs like that, but with Agentic, there is a lot of independence, Deliber ative planning

dynamic Deliberative planning, dynamic , Um environmental Um, continuous environmental int , the complex ut, and of course, the complexi.

of the objectives and goals is of the objectives and goals is .

high up, and what, um, high up, and what, um, we saw be I come across this um , paper I come across this, um, paper ts g at was showing a distinction ben l AI and c ween traditional AI and agentict A distinction between classical A distinction between classicals reinforcement

agents, uh, reinforcement learnd Agent TK ng agents, and also Agent TKI s earlier, The major distinction is around The major distinction is aroundy d goal-directed autonomy and goal-directed beha.

high Hi, well, agent uh systems Hi, well, agent, uh, systems ha m very low human inter actually very low human inter actually l .

High adapt ability , and they work High adaptability, and they wor in complex environments that a c and very a e dynamic and very context-awar And all these other character And all these other characteris.

l The characteristics and The characteristics and capabil k AI that ties of Agent Ink AI that we thk e say nk about when we say agentic AI t are Auton omy and goal complexity, Uh the complexity of the

Uh, the complexity of the envirs is nments that the agent is deliben independent decision-making independent decision-making and with very adaptability with very minimal n .

Continuous learning Continuous learning as the agens interacts with this environmen Plan ning acting , and Planning, acting, and carrying t ut multi-step acts. Now, if youk these characteristics Um, these characteristics have d een around before. People have

t een talking about these same th What is making this What is making this a little lo e is f der than before is the degree oe f independence, the degree of au Um the complexity of the goals, butt or objectives that agents butt or objectives that agents g

w re achieving right now, and of level of ourse, the level of planning ann that we are seeing at the moment that we are seeing at the momen There are some generic architect There are some generic architec Um a common one is the Um, a common one is the multi-a e , where ent system architecture, where y collaborating uh

too achieve a certain achieve a certain task. Sometimt

achieve a certain task, sometima s it's a competitive or adversa t .

And then we have a hier archical And then we have a hierarchicale a high case where there is a high-levee defines Some sub-go als, and then Some sub-goals, and then gives o l agents to hem to low-level agents to exec And of course , the goal-directed And of course, the goal-directer , where the modular architecture, where th

t agent function is just dividedo s , and each into modules, and each module ia task.

task.

Um, that the Um, that the agent is supposed While I was looking through While I was looking through lite this is a very rature, well, this is a very ne o people are area, so people are still thin m is still ing the ecosystem is still deve.

Uh, but I saw that there were no Uh, but I saw that there were ns ed design patterns , and also design patterns, and also this r e thing aper said the same thing. And sI

o the Anthropic page?

page?

And I found something .

Um The major building block of The major building block of age , or what I tic AI systems, or what I prefel agentic to call modern agentic AI systs ms at the moment, because, agai e agentic AI , we have agentic AI systems th before, The major building block is The major building block is the M. The LL

M. The LL augmented LLM. The LLM has augms

augmented LLM. The LLM has augms retrieval Tool tools Tools, um tools umm And memory , and different And memory, and different, uh, , And memory, and different protoh ols that help with implementings d Communication between agents Communication between agents ano r communication between agents ans

access to other tools, and thi So the day, when So, at the end of the day, when I you see modern agentic AI syste, the major building the major building block is goio um LLMs , and they are going to have at least one of these patterns at least one of these patterns, will which, if you look closely, wil

c map into the generic aperturest I that I introduced in one or twos o .

So um in the interest of time, I will just uh draw I will just, uh, draw your atte the orchest tion to just the orchestrator w.

um work workflow, so if you look at the workflow, so if you look at thee Anthropic page, I went to the Ae e in Lucia thropic page because in Lucia to there were not so many univers al there were not so many universa what design , um, agreement on what design e d atterns we have, so I found the h e design patterns on the Anthro.

And they try to um And they try to, um, put a distn n workflows d and agents um right now and agents, um, but right now, k hat I want to talk about is an , where you Some sort of pattern where Some sort of pattern where thera That breaks down tasks for That breaks down tasks for the ,

And then , for the worker agents And then, for the worker agents k so there are some other worker e environment, And then later, And then later, the synthesized and results are put together and gin I found this interesting because I found this interesting becausI at

Um this example from the Stan Um, this example from the Stanf Oh, Virtual AI Lab.

AI Lab.

Where researchers allowed the Where researchers allowed the ee be handled by tire floor to be handled by ageo PIAI And then And then there are research scis are handling ntists that are handling all th And so, literally, humans are And so, literally, humans are s anding on the side and just bei s g observers in the whole proces Now, this is

um , a research Now, this is, um, a research prs d cess, and they had some major o s . They

. They to Um generating ideas for COVID Um, generating ideas for COVID- .

And it was able to give good results in a very good results in a very short ti Until , at the end of the Until, at the end of the day, y e Professor Ter u realize, like Professor Terre, The work that we were taking so The work that we were taking soo AI agents are being able

to do AI agents are being able to do .

Um very short time .

And um if you see And, um, if you see the pattern I provided before t that I provided before, most o AI the things, most of the AI age s t, um, environments are going t in at least one of these um patterns um patterns. It could be a

patterns. It could be a patterns. It could be a chain w

patterns. It could be a chain w g output ere one agent is giving output The next , uh, agent The next, uh, agent, and that we t to that agent uld be the input to that agent.d

, it could It could be routing, it could b n working in parallelization, why reby there's some sort of secti r voting.

voting.

At the end of the day, we At the end of the day, we want o apply, one agent Or it could be one agent providg s and another t ng answers and another agent ev s before luating those answers before th in conclusion The story is just continuing The story is just continuing, i start yesterday.

start yesterday.

didn't just start yesterday. A

extends

entic AI extends traditional AI s .

agents used to have goals before agents used to have goals befor d area, uh, , goal-directed area goal-direc d around before But agents now, have more complex goals have more complex goals and objs Um they have more automony Um, they have more automon, a hf gher degree of autonomy, they a

e planning e doing all the planning indepey Um they have access to memory Um, they have access to memory s of course, they're nd tools, and of course, they'rg using a shorter time than we a Uh modern Modern agent systems are Modern agent systems are combing models.

models.

with workflows and, um, with workflows and, um, environ to ent feedback to accomplish comps And we have different architect And we have different architects s ral patterns. It's not a one-si

ral patterns. It's not a one-si .

And it depends on the problem And it depends on the problem dn f main and the kind of tasks thate o follow.

follow.

or to solve .

Thank you .

Over to you, Thank you. Over to you, Profess.

Thank you. Over to you, Profess.

Swany ana, thank you so much for Swanyana, thank you so much forg that interesting insights over And certainly gives us very academic view on the space very academic view on the spacet o .

that we're moving into. It kind

of reminds me a little bit of, .

of reminds me a little bit of, s e of reminds me a little bit of, t ike, jeans like over time, you .

r now, like fashion never changesy ow like fashion never changes r ally it just goes through theses to me cycles and it feels to me like e cycles and it feels to me like i But on another level, things are But on another level, things ar There's certainly changing. There's certainly fuy

changing. There's certainly fuy Increased auton omy.

Increased autonomy, increased Increased autonomy, increased ey r ontology pressibility, better ontology, s etter knowledge bases, these tys es of things that have come inte e the person to really sort of the person to really sort of tas difference is s What that difference is, what's so exciting about the space tha w is Mr. we're in right now is Mr. Blace o

if I can ask you ie. So if I can just ask you to

ie. So if I can just ask you to while I bring up your slides while I int d be Really great.

great.

Okay so our next speaker is Okay, so our next speaker is Mrk e received a Black. He received a Beng and

Black. He received a Beng and t eng degrees in Computer and Elen eng degrees in computer and eleg tronic engineering. So he's com

tronic engineering. So he's com .

As scientists , he's one of tho As scientists, he's one of thos As scientists is one of those fo om the other side. From Northwe

Pochastr t University, Pochastrum, Southa University, Pochastrum, South . And he

. And he frica, 2015 and 2018. And he's u urrently pursuing his PhD in co g at North west University .

at Northwest University, where d r e's involved in both undergradu te teaching and postgraduate sur n . He also

. He also ervision. He also runs his own t

ervision. He also runs his own t business. And as

business. And as

onsulting business. And as you w now, nowadays in in the modern t 's almost essential I space, it's almost essential hat you be an entrepreneur at te Gotta be trying Gotta be trying to become a fou want We also And this business of these And this business of these is a agentic s ound AI-driven agentic solutionp streamline to help organize streamline bu t

NWU, iness processes. And at NWU, he

iness processes. And at NWU, he t program s involved in a pilot program i y , an AI pow troducing Mindjoy, an AI powere STEM learning platform to enhae and learning for ce teaching and learning for une .

And so , Mr. Black ie, can I hand And so, Mr. Blackie, can I hand u and you can take over to you and you can take ity Awesome, thank you Prof essor Awesome, thank you, Professor C And thank you for the And thank you for the introductn um s on, and, um, to Zira, thanks fo .

So I think I 'm gonna take So I think I'm gonna take whatet let's hype er you presented, and let's hyp Professor Theran it up. I know Professor Theran

it up. I know Professor Theran e said that there are some hype m In the panel, and I think I In the panel, and I think I am .

So I would like to take you So, I would like to take you tot r this year And this was On the evening of the On the evening of the 5th of Mah n Stalin Bow I ch, in Stalin Bowers, where I ad This seminar presented by

the This seminar presented by the o.

And the person that looks almost And the person that looks almos like Rasi Erasmus there, but it His name is A.D. Pin

A.D. Pin

ar, and he His name is A.D. Pinar, and he t ure Studio urrently runs AI Venture Studio d Ubundi g Let's say Agent Let's say agentic systems Let's say Agentic systems to bet .

And at that specific evening And at that specific evening, hd a that dropped this idea that really e And he said How about wee we have a How about we we have a world wh no longer ere humans are no longer the ma systems n consumers of digital systems But we have personal agents th But we have personal agents that

And this really triggered And this really triggered a tho , ght where I said, okay, but let d where we have s imagine a world where we have A world where we can just A world where we can just chille , and on the beach, maybe, and you ha hand it over That AI agent has access to That AI agent has access to you, At the meetings

, review reports At the meetings, review reports check dashboards, and this doe like, nothing could go hey, imagine t hey, imagine a world like tha And where we no longer have apps And where we no longer have apps us or websites or systems for us this whole s people to use, but this whole And we're building tools And we're building tools and so

s for AI tware and systems for AI agentse And that sounds crazy.

crazy.

But some of the But some of the ideas is, imagi like, let's e you have something like, let'p s say, an app like this that loos .

Where you have people and agen Where you have people and agentn .

Like in the screen in the Like, in the screen in the middu a chat interface e, you have a chat interface wir can do h all your AI agents that can d anything for you, and on the f e something like r right, you have something likI agent Agent Discovery App Agent Discovery App Store, whery you can literally go, and you n agents an discover agents. You can disr over something like a career see e this

p track I TrackIQ track IQ, where you can just track IQ, where you can just chh r service about t with your career service aboul g where your parcel, scheduling k pickup, something like that, iu o attend an event you need to attend an event, h o w about you chat to an event pry And they send you your ticket to And they send you your ticket t

And some of these things And some of these things might d f , but e ound far off, but there aree the se tools are actually being shi.

And I want to show you this And I want to show you this ver cool use case of one of my fri l ingboards where nds in Stalingboards, where he his company uilt an AI agent for his compans . So, this is a screenshot fromn

. So, this is a screenshot fromn e from And he built an AI agent called And he built an AI agent calledy appropriate. But

appropriate. But

Rasi, okay, how appropriate. Bu

messaged Ras anyway, so he messaged Rasi ond WhatsApp and said, hey, Rasi, ts e team has been working very ha So let's order some beers for So let's order some beers for t ats.

e theme and some Kit Kats. He gm

e to go and log ve him the website to go and los d buy, And he said, this is the And he said, this is the order.

The AI agent comes back and say news, he's , hey, Chris, good news, he's bt en working on it, logged into Py ck and Pay, added it to the che Sorry, we lost We Yeah saw someone Sorry, we lost your sound there Yeah, um, I saw someone accidend u ally muted me. Can you still hee We lost you.

Yes.

Okay so then basically Okay, so then basically this agt Yes Okay, so then basically this agd e nt went, and he placed the orde t an update , and he sent an update. And the

o of Pick re you can see the photo of Pic, Arriving at his office Arriving at his office with hisy And now this whole world And now, this whole world start by someone very recently we unleashed onto very recently, we unleashed ont This chaotic

agent called This chaotic agent called OpenC Where you basically give Where you basically give this ts , and n ant, and the agent can decide w And just imagine what that can And just imagine what that can o for you, and even though it c t n be very helpful, but it can a so be very destructive. But nowe

people started to build on top f these, and they've launched se mething like the Hermes Agent, e t that learns How you interact with it, it How you interact with it, it le ds your perspectives, it lends our ideas, what you like, what n .

And it grows with you. And what

you. And what

And it grows with you. And what

e they even do is they give theses o agents the capability to, let's, at so say, to dream at evenings, so te ey process the daily interactio g s to store it in long-term and .

And even Google started with And even Google started with a e agent for us to emini Workspace agent for us tog start building these agents on e And just earlier this And just earlier this week, wha , if I we saw is, if I play some of tn have a ese snippets, that we have a trr t p planner, for example, that Go d , where you take gle presented, where you take a

g image of, let's say, something you like, and you send this thi , I want to go g off and say, hey, I want to gr on a tour, and prepare a tour n or me that's based on, let's sa e shop idea , this coffee shop idea that I And it would go and finish up And it would go and finish up a for d complete that plan for a trip Or the other

for you. Or the other agent tha

for you. Or the other agent tha Google demonstrated was, let's people in say, for the people in Stalenboh n ch and Cape Town who can't everd What if you're on your way What if you're on your way to at just event, and you can just ask yoI r AI agent to reserve a parkingt your way to spot for you on your way to you And then the last demonstration And then the last demonstration e

'd like to was, let's say we'd like to boo of a flight, and instead of, let' n your passport say, typing in your passport e You have an AI agent You have an AI agent that alrea and looked at y went and looked at your passpt file system, wherever rt in a file system, wherever t d at is, it retrieved the informa And it can fill it in for you on

And it can fill it in for you o s the spot. And this is some of e

the spot. And this is some of e the people he tools that the people are bu .

And with this massive explosion And with this massive explosions d with this agentic world, and td open claws, and is world of open claws, and thes e agents, people started to flock towards people started to flock towardsy buying let's say, buying mini-compute s to host all of this infrastrue ture in, because we're not livin e

Where you buy a PC for Where you buy a PC for yourselft g PCs r your but you're buying PCs for youro n AI agents to live in, so that tn o ey can go and do whatever it nes .

And you can just And you can just release them oo Then, also in this week, and Then, also in this week, and sp, What Anthropic What Anthropic launched was a bf s for financial nch of AI agents for financial teams ervices, where teams, let's sayw or KPM , can I have access to some of can I have access to some of the that can go se financial agents that can go

these financial agents that can go and run independently, and l terally last night, Google laun ally last night, Google launchew CoScient a new agent called CoScientist ent AI which is a multi-agent AI part p er tool that can help researche s, specifically in life science Go and explore the vast amounts Go and explore the vast amountsr p of literature, come up with newh and then hypotheses, and then giving thes

y to work through all of to work through all of this lite d new solutions rature to find new solutions to And the scariest thing, if I put And the scariest thing, if I puh up this graph that's been circg , a year lating, let's say, a year in th this year 25 of February of this year, i that this graph try to demonst If we took, let's say

If we took, let's say, more or f ess the population of this worl And they looked at And they looked at who has been these AI systems. And each one of those And each one of those dots repr n people.

people.

And as you would see, I think And as you would see, I think t t small. I

small. I

e legend is a bit small. I triet

.

But the grey blocks is basically But the grey blocks is basicalle o 's never used 84% of people who's never used Like, out of 8.1 billion, 84 % has never used an AI tool or 84% has never used an AI tool o.

16 %, which is the 16%, which is the green blocks , they actually f people, they actually use thee like ChatGPT free chatbots like ChatGPT and Then we get up to a 0.3 Then we get up to a 0.3% of peoe y , that actually y le, mostly, that actually pay f plan r a $20 subscription plan to uss And

then , the most shocking And then, the most shocking parl the f of it all is the builders of ts d is new agentic world that we'reg .

is 0.

0.

04 % of the population is 0.04% of the population. Andt

t a crazy statistic that is just a crazy statistic We're starting to live in We're starting to live in a wort o far divided d that is so far divided betwee the people who've never even s en or used AI or who've been exd And then the builders of And then the builders of this n 0.04 w world is 0.04%, and that's ju And we're so used to chat bots

And we're so used to chatbots wd e s the GPTs, the claws, and the Gemini But as the Zira also mentioned But as the Zira also mentioned,t our behalf.

our behalf.

that agents act on our behalf. n

, hey can reason, they plan, they y delegate, and use tools, they delegate, and t And in a nut shell And in a nutshell, from an, let s say, a software development p n agent is nothing rspective, an agent is nothing t ther than just an application iu models, With tools , some sophisticated With tools, some sophisticated m To achieve a desired

To achieve a desired goal. And

pull back n a bit f we pull back the curtain a bit Behind behind the technology behind the technology, we see tt basically at the agent is basically just e the LL the tools the tools, and they run in a lo.

And the loop keeps running And the loop keeps running unti .

So we can define it in terms of So we can define it in terms of the , and then the hands and feet, and then th Where the brains Where the brains are the foundas s ion models we have that examine the goals, breaks it down, dec t des the steps that it needs to e ake in order to achieve that go.

Then it can decide, okay Then it can decide, okay, whichP or o API tools or integrations do I And and connections that I can use and connections that I can use, h estration layer and then the orchestration layes n a loop.

a loop.

And this loop goes from And this loop goes from a thinkg , bubble ng, let's say, bubble, where the s agent perceives and reasons, wt s context, from the From the goal , the tools, the From the goal, the tools, the mt mory that it have, and then it s ecides, do I have enough informn based on y tion just based on the query, o

else?

else?

And then it goes into the acting And then it goes into the actine stage, where we output a strucd t . It's

. It's ured request. It's just basicale

ured request. It's just basicale u use.

use.

Or you can kick off parallel Or you can kick off parallel agt nts that can do something for yd e u, and then we observe the resu.

We put that result back into We put that result back into tht context, and now the context h n again.

again.

So these tools that we mention So, these tools that we mention t is none other than just code th we define t executes, where we define a t We give it We give it a name, something li , we give it e Web Search, we give it a prop And we define what the input is And we define what the input is what do we need, what is the v riables in the software develop And

basically, what the AI model And basically, what the AI modes does is it creates what we cala response this JSON a JSON response, and this JSONe response, we have in something N ike a decoder of the JSON paylo, Then we can extract what Then we can extract what the tos used, we l is that needs to be used, we e an see what the variables are, t to normal

nd we just pause it to normal s And then the functions And then the functions and the , and the ode executes, and the model see the results, and the loop runs.

And all of this happens And all of this happens in eith of the AI r working memory of the AI agen t that we call the context windo Which is the fast, let's say Which is the fast, let's say, b e agent has unded memory that the agent has t access to now, and it fills up that we use ith each interaction that we us And this is how

And this is how, uh, yeah, whate window, and we call the context window, ande e then the new thing that we're l s is giving oking into these days is givingg , like agents something like, uh, likee like long-term we humans have, like long-term .

Where in a specific moment Where in a specific moment in t I agent e day, the AI agent or the soft e have are tools that we have process e he daily interactions, store itn e And this gives the agent let's say , long-term memory of let's say, long-term memory of e

ow we've interacted with it ove But anyway, as we build all of But anyway, as we build all of hese software tools, we always eed to design for failure in mi these AIs d, since these AIs can fall apa apart in t, and they fall apart in predis .

And one of the things is uh it And one of the things is, it mir ht never converge onto a specif so p c answer, so it can loop foreve can uh can put a cap on the iterations can put a cap on the iterations s hallucinate it can sometimes hallucinate ad s tool and think that there is a e e ool and come up with someone thl tt some tool that's not there,

n catch that.

catch that.

context blurred context blurred if the brain ge then s too full, then it sort of losf s track of what's happening, an t cost surprises then you can get cost surpriseh , because each loop costs a loty .

set budgets or alert s on your set budgets or alerts on your u.

And then what we can do And then, what we can do to mak l it impactful and for us to bui s d some of these systems for the e e AI tools to use, we have somes foundations that needs to be in well-doc umented processes and well-documented processes and sg andard operating procedures. We

weld , need to have weld, let's say, d data and centralized clean data and centralized clean datae n a very cur source, and then a very curated d crisp descriptions tools and crisp descriptions of We may need to put in some We may need to put in some guar e rails, like identity and accesst management, with giving the age c

scope t only a specific scope, and whd t not.

not.

Keep a human in the loop , and as Keep a human in the loop, and a , I've seen on social media, peo , le give their AI tools, like, do databases lete rights to wipe databases, And then yet again always And then yet again, always add s d ackups, Oded logs, and stuff li And then And then, what we see in indust

y these days, some of the proto ols and interfaces that are sta Uh, so if you would like to Uh, so if you would like to buin d some of your own tools, is lot s the k at MCP, which is the Model Co text protocol, which basically your tools, into an AI system into an AI system. The other on - agent protocol g where you have inter-agent com .

A new development is something A new development is something e an agent ic naming service an agentic naming service, for hose of you who are in computer t S science, would know about DNS, r computer networks, where we he name servers For websites to look up, For websites to look up, now wek can look up agent capabilities m n also nd discover them, and then also ability to we give agents the ability to he

s ndle payments for us, and that e s where we have the Agent Payme.

So the last question that So the last question that I juse wanted to leave with you as wee uh u How will you build for a world e agents are your y here AI agents are your primaryr ?

And most probably not And most probably not humans ane So that's something to ponder on So that's something to ponder o , so that's , and then, yeah, so that's thef y my discussion .

Mr. Blackie, un believable , unbel Mr. Blackie, unbelievable, unbe As you were talking.

ievable. As you were talking, I t this was just thinking about this, t was just thinking about this ide individuals digitally native , digital native digitally native, digital nativs o up in a individuals who grew up in a tn l ogue t wn post analogue. And I just ge this feeling like we're going e f o have a generation of individuo

native They will grow up in an era They will grow up in an era wheu They will grow up in an era whet e the internet is almost irrele 's an interaction ant. It's an interaction betweeu

ant. It's an interaction betweeu and agents you and agents and agents orcht e That world . Just before

. Just before we That world. Just before we head

That world. Just before we head l discussion towards our panel discussion, w o ere we really get a chance to u pack some of the questions on m mind, and I think there's a lo s on of questions on other people's pop the Q A minds, you can pop the Q&A if yy s . I think

. I think This is a good time for us This is a good time for us mayb allow Kenna just to allow Prophet McKenna r reflect a little bit for us ande e bit of additional things to think about . So.

. So.

things to think about. So, maybe

while you're just bringing up s just our slides, I can just introducy So If you see , so Prof McKenna If you see, so Prof McKenna runl several international projectsn education As a public good , she has long As a public good, she has long What it does and which knowledge What it does and which knowledg

e lined are silenced or sidelined, thi lens means that she sees the w as both rlds of Gen AI as both fascinatg . I like

. I like ng and terrifying. I like I hav terrifying feelings sometimes terrifying feelings sometimes a n But also.

But also, I see the But also, I see the festivatione for her She has won awards for her teah She has won awards for her tean hing research internationalizn a co-auth tion, and for a co-authored boo g on understanding higher educatn So if I can hand over to you So if I can hand over to you, Pn a , and you can s AI agents e through AI agents and the Knowl

Great , thank you so much Great, thank you so much, Teren e whole SF Well um, you started Well, um, you started, Terence, by saying that the panel's madep perhaps d up of, perhaps, halfmen and pess Um Herman Kemp, the harp man, Stat

us , I am going to probably Status, I am going to probably,y the role of very sadly, claim the role of t t .

Um the previous Um, the previous speakers have y what lready outlined what agents actd t , so I don't ally are and are not, so I don' Um as I indicated, you put in a Um, as I indicated, you put in n task. You don

task. You don

concern or a task. You don't eo en need to phrase it as a reque en need to phrase it as a reque, t, um, into, uh, your agents, ay language d they will use large language s to determine the next they actually name Agent y Thus the name Agent. And they t h oop that. They do that through f

oop that. They do that through f s um Um they do that loops . So at its ce

. So at its ce s of loops. So at its core, an t um gent is, um, as Herman just sai a series of s , a kind of a series of loops, s .

And that 's that's And that's that's already prette But I 'm not ready to call that But I'm not ready to call that sense.

sense.

Um Hanna h Frair, and she's a Um, Hannah Frair, and she's a m thematician, and if you don't fw r um You have that great um thing to look thing to look forward to, becau s e I think she's pretty phenomen.

d she l. Um, and she has shown that t

l. Um, and she has shown that t s can't even t ese systems can't even outwit ay c kind u very basic kind of, are you rob And that has led to this very And that has led to this very b y that zarre reality that agents are o r tsourcing some of their activits s who have

es to humans who have to, uh, nd embodied ed an embodied reality to enactm , and them, including, and it costs ad a human ound 2 rand, asking a human bei fill in the g to fill in the capture, to alw agent to h ow the AI agent to continue witr .

Um I mean, they don'tt Um, I mean, they don't they don 't have the human failings of f n tigue and frustration. Um, theyt

e up.

up.

So, you know, if Home Affairs So, you know, if Home Affairs p d , they don ts them on hold, they don't han r 25 If your municipality ref uses to If your municipality refuses to they don't ve up. They keep sending the em

ve up. They keep sending the em g other ways of ils and finding other ways of g an absence of e and But an absence of fatigue and f believe, is ustration, I don't believe, is,m necessarily uh intelligence.

intelligence.

So what can they do for So what can they do for researc we know ers? Well, we know already from

ers? Well, we know already from s the large language models, but g ncreasingly using agents, they liter an do these enormous literature.

They can retrieve , read, synth They can retrieve, read, synthe y flag contrad ize, um, they flag contradictio c itation s, they map citation networks, I mean, it really is if you're I mean, it really is if you're working in a huge interdiscipli u ary field where you're not an et They really are They really are an incredible bn

n test on. They can test hypotheses, t

on. They can test hypotheses, t ey can generate hypotheses, the g can do virtual testing of varis Um and we saw that Um, and we saw that recently wi 's Alpha Ev e h DeepMind's Alpha Evolve, um, g t hematical ort of solving mathematical pro t lems that haven't been able to e .

Last week, I was very lucky Last week, I was very lucky. A

s at Freyer eek before, I was at Freya Unive ky. A week before, I was at Fre

ky. A week before, I was at Fre g ersit in the meeting with some D who AI hD scholars who were using AI a hD scholars who were using AI as ents to hypothesize and figure t How to break down the How to break down the long prots in strands that cause Alzheimer o , really, I mean s. So, really, I mean, mind-blo

s. So, really, I mean, mind-blo Um, and then for the ordinary Um, and then for the ordinary s f cial scientists like myself who y are not working in a laboratory they're still pretty cool, bec um can use they, um, you know, they cam data transcrib , , um, analyze data, transcribe,e n write

track literature, they can write your entire document if you wa grant t them to. They can do grant ap lications, you know, you can lo t application d a grant application, tell it w r o draw on all your prior resear n fill in the entire application fill in the entire application,n potential it can identify potential, um, olleagues that meet the criteri e of the funders, it can write u f

wonderful things But uh my own concerns , which But, uh, my own concerns, which come from my own research, maine , um non-Pilot Schumer, Non-Pilot Schumer, Neil Cram Non-Pilot Schumer, Neil Cram, u l a Pal , Nicola Palatz, and various ot about the ers, is really about the Knowlet ge project. So that concern lie

ge project. So that concern lie mean for . What does all this mean for o

. What does all this mean for o f knowledge?

knowledge?

Um and I'm gonna identify Um, and I'm gonna identify fived interconnected threats that I s g from AI .

Um, and I'm going to have to Um, and I'm going to have to ta Um, and I'm gonna have to talk r k very quickly to get through t ery quickly to get through them in time, but before I jump into o raise that, I want to raise a really t e that mportant issue, that as we movem o co-pilots e arge language models that assis to lab pilots , the large

the to lab pilots, the large the AI y do the agents that actually do the re.

We need to consider where these We need to consider where these Because a lot can go wrong, and Because a lot can go wrong, and, already, um, we heard from Herm t Peter Steinber , n about Peter Steinberger, his e who used OpenClaw who used OpenClaw to develop the who used OpenClaw to develop th who used OpenClaw to develop th who used OpenClaw to develop th se AI agents and made it readilo

e big available. So these big, um, bh

available. So these big, um, bh g tech companies that had alrea d modern AI and refused to make and refused to make it yet publ e cly available, because they wer so aware of the enormous threa moment Peter s. The moment Peter made it ava

s. The moment Peter made it ava y couldn't .

The enormous funds , and they The enormous funds, and they we .

But we need But we need to think very caref y this is being lly about why this is being giv entirely by n. It's being driven entirely b

n. It's being driven entirely b profit, and any of us who thin e big tech t that the big tech is driving i for um scientific for um scientific um advanc s, for the goodwill of people o y the planet. You simply haven't

the planet. You simply haven't watched enough interviews with r berg, Sam he likes of Mark Zuckerberg, Sa, , Elon Musk Altman, Jensen Huang, Elon Mus r People who Or, um, what can I Or, um, what can I say? A littlt

a little bit crazy , a little a little bit crazy, a little bi what are , uh, psychopathic. So, what ar my five big, um, concerns abou ?

Well, the first Well, the firstt I won't just j mp straight into the first one.e

The first one is about the resp.

So when an AI agent produces So when an AI agent produces a g lawed finding, a fabricated ref, analysis rence, a biased analysis, a spus n ious correlation, who's respons ?

When it gets out there When it gets out there in the w t rld. And we know that this is n

rld. And we know that this is n e t hypothetical, because we alrey dy, um, and the previous spoke s e as alluded to this, we are alre e thousands of emails thousands of emails. We're seeis

e owners' g agents use owners' signatureso contracts to sign legal contracts, to rac credit card debt Who's responsible for Who's responsible for the fallo Who's responsible for the fallo t? And what does that mean when

t? And what does that mean when ethics our research ethics and integri y frameworks are all built on t f responsibility g ? And I don

? And I don n a human being? And I don't th s nk we are thinking these things .

Hassani Murad and Reznik Hassani, Murad, and Reznik, in heir report just a couple of mo d this e ths ago, called this the respon y ibility gap. And they proposed

ibility gap. And they proposed team hat any research team needs to I validator validator ave an AI validator or validatoI r who is responsible for who is responsible for overseei s of the AI g all the actions of the AI age l their t and double-checking all theirs s requires a

outcomes. But this requires a k

outcomes. But this requires a k nd of institutional will and fu I'm not even hearing I'm not even hearing these convs universities yet rsations in our universities ye .

Another big concern that I Another big concern that I havet k illing.

And I know that this has And I know that this has been p , in some of the o-pooed, perhaps, in some of th istic more techno-optimistic literat re, but I think it's a real ins . And that is

. And that is dious risk. And that is that enl

dious risk. And that is that enl s ry-level researchers, I'm talki , Under take very and to take very boring and to take very boring, basic s , like cod esearch tasks, like coding tran Conducting literature Conducting literature reviews, g t s, managing riting first drafts, managing da process of ta, but in that process of doine d

ane tasks those somewhat mundane tasks, e hey developed the tacit knowled of their disciplines of their disciplines, they figu t what s knowledge e out what counts as knowledge,t as truth uh, what counts as truth, what need evidence Um, and all of that kind of Um, and all of that kind of thi I worry that g. And I worry that when agentse

g. And I worry that when agentse tasks, take over these tasks, early ca t skip eer researchers might skip a ve n into y important induction into the e .

that we hold so dear . And again,

. And again, that we hold so dear. And again

n i d drawing on Hussaini, Murad, an They say that elimin ating these They say that eliminating thesem have roles from humans could have wi m e-ranging and long-term adverses scientific implications on the scientific e Now, of course, depends Now, of course, depends whethert you care about the scientific wk

.

Uh, GreenGuard found Green Guard found that knowledge Green Guard found that knowledgo y use workers who regularly use, rel s on generative AI, this is befo , have become e agents, have become increasiny t , which is ly confident, which is the anti t a scientist .

Hed hedges and says , in light hedges and says, in light of cut n , we tend rent information, we tentativeln , but those can conclude, but those who arg AI relying on AI are already beind r more confident found to be far more confidentr And far more willing to And far more willing to hand ove f r more and more of problem solvg AI.

AI.

, ng to AI. So, I worry that thisl humanity out of will take the humanity out of s consequences for ience with consequences for hum Third big concern , epistemic Third big concern, epistemic ine ustice. Whose knowledge gets am

ustice. Whose knowledge gets am ? We know

? We know lified? We know that they are t

lified? We know that they are t existing lified? Uh, we know that they a

lified? Uh, we know that they a e trained on existing corpus of.

With a heavy with a heavy privileging of with a heavy privileging of a pr , a rticular form, a Global North, the Big nglish, published by the Big Fi following e, um, following particular the retical traditions. This is whey

retical traditions. This is whey actually sits e my own research actually sitsn c injustice It's on epistemic injustice inf .

Um and Kay uh I'm saying Um, and Kay, uh, I'm saying hise Kasir Z adeh, and Muhamm ad have Kasir Zadeh, and Muhammad have r epistemic great paper on epistemic injus.

And they identify And they identify these four me s you can see on the slide hanisms you can see on the slidh e um , by which these, um, these lar s and e language models and increasiny c AI is systematically under is systematically underrepresen of the Global South ing scholars of the Global Soutt . Now, you might say our studen

. Now, you might say our studen g biased Um knowledge when they read any Um, knowledge when they read ane articles, because that is alre e Scholar But um this is at a scale that But, um, this is at a scale tha r . Oh, I

. Oh, I is far greater. Oh, I can heary up, so let my time is almost up, so let meg t one.

one.

Knowledge homogen ization that Knowledge homogenization, that n problem eally draws on the same problemt to all t that we are going to all start ncreasingly framing problems ine y synthesizing the same way, synthesizing litee ature in the same way, intellec s Um and that 's nothing to

Um, and that's nothing to say a l out the environmental cost. Youw

large language models know, large language models uses enormous amount more than a Goo h le search did, although now, ofe course, Google searches use it t .

But our agents use that by a But our agents use that by a hu even more.

even more.

e amount even more. So we reall risk homogenizing knowledge at f the planet y the cost of the planet. My finat

verifiability point to the verifiability pro ed to.

Um how do we know how the Um, how do we know how the ageno s got to their answers? A lot o f it's black box kind of stuff, e actual nd a lot of the actual develope agents admit s of these agents admit that th how the s y don't know how the results ge .

Um, and we Um, and we're already seeing hu academia ans in academia abusing this te using ribly with, um, with using a us ribly with, um, with using a, us , to do their , using agents to, to do their g their eviewing, to do their writing, n So sorry in closing, I am over So, sorry, in closing, I am ove.

time. I don't think we can rest AI , I don't think we can resist AI, I don't think we can resists e to stay agents. They're here to stay. U

agents. They're here to stay. U

, I think that ship has sailed.m

But I really think we But I really think we should be calling for something that's ac ually much harder than resistan.

I think we need to ask I think we need to ask the very s about h important questions about whichh research tasks should and shoult not be automated. We need to a k questions about what happens hen our students automate everyg o do hing that they need to do. We n

t ed to decide what the Knowledge y for, who Project is actually for, who doe Um what is the point of Um, what is the point of all ofs us academics doing research in t And this is not just a technical And this is not just a technica, question, this is deeply polit , and cal and ethical question, and i worries me that universities a s to

e leaving these questions to bih tech with their profit interes And that they are not And that they are not actively ethics behind ngaging in the ethics behind AIs .

Over time thank you.

thank you.

Don't worry , we lack a little Don't worry, we lack a little b e here Thank you very much for that Thank you very much for that di t 's kind of like cussion. It's kind of like we wh

cussion. It's kind of like we wh of the nt through this phase of the en n of of truth at the dawn of generas And I fear that now human sort of even being human sort of even being replac the the e world d as the the actors in the worl where we're going even beyond it's almost he end of truth it's it's almos you know that the items you'reg

are going to sign al an end are going to signal an end of s , and ience in some ways, and we're g d Who did the science, anybody Who did the science, anybody ev n was real and what was the mote t So certainly some of these So certainly some of these thin g . But

. But I think this brings us I think this brings us to our nt have our panel xt part. We have our panel disc.

xt part. We have our panel disc.

ssion. Do you mind if all of yot r Your Your screens, I think Your screens, I think that woule On our panel so we can see all On our panel, so we can see all .

of you. Yes, excellent. Now, I

Herman is probably thinking Herman is probably thinking tha .

you might have gone too far. Ao

I

would maybe d so I would maybe like to give maybe over to him to maybe just refle bit the The items that have been The items that have been raisedr f Tirens.

Tirens.

Yeah, so I really enjoyed your Yeah, so I really enjoyed your iscussion now on highlighting t , or highlighting e concerns, or highlighting all .

And I think one of And I think one of the one of te I I try to ask I try to ask myself these days , AI really came and disor iented.

my so I felt totally disorient my so I felt totally disoriente feel d in this new world. I feel lik You know when you're standing i a pool on something like thosel floating, and pool stuff that's floating, ande you try to balance on top of th t thing? And I think this entirs

t thing? And I think this entirs e these days world feels to me these days l.

Feels like I'm constantly Feels like I'm constantly in thl feels like I'm constantly in th s pool, there's waves, and I'm balance rying to find my balance on the strange , floaty thing , and I strange, floaty thing, and I thk the AI world created that for me the AI world created that for m s How can we start to

engineer How can we start to engineer an systems for systems for this new world thate n ? How can we

? How can we we live in? How can we acknowlee s and the ge the risks and the challengesd , but as well, and the concerns, but e s hen how can we develop the tools that we can teach that we can teach the new studes s ts, or the new employees, or the the upcoming generation, and the upcoming generation, and ho Oppenheimers

to not be the Oppenheimers of go and use his world that go and use this d bs, but ech to build atomic bombs, but ither how can we harness the otr s , like, let's er possibilities, like, let's s, try to build fusion try to build fusion reactors toe generate energy. And I think th

generate energy. And I think th s the tension that I s is always the tension that I p .

And , um, I think there 's always And, um, I think there's alwayse benefit, and a an advantage, a benefit, and a t f rade-off when it comes to all o I'm I might be also an IPMAN, but I might be also an IPMAN, but I really much m And let's say balanced And let's say, balanced in my vI

, ew, that I know, yeah, we reallk w we need to think hard of how we a d when do we unleash these tooly And um and I really And, um, and I really like one e t Google De of the things that Google DeepMd s Dennis Alsabi, Well, I saw this what he said i wished that y he wished that AI stayed a bit n So that

they could build So that they could build more t AlphaFold, But Sam Altman 's greediness But Sam Altman's greediness rea d unleashed a new ly went out and unleashed a new And I kind of agree with Dennis And I kind of agree with Dennisr really on this matter, but I really alg s o like using AI coding tools th e

se days. They really bring me j.

se days. They really bring me j.

I hear you .

Before I allow you I hear you. Before I allow you I would just like to I would just like to ask Ms. Sw a , the feeling I would just like to ask Ms. Tht we're feeling I get is we're hearings e that This is a new world.

world.

This is a new world, but This is a new world, but somethI from you was ng I kind of felt from you was t such a new hat maybe this is not such a ne world, this iss we were on thee l multi-ag same exponential multi-agent sys , they're now tems existed before, they're no . Like.

. Like.

calling it agentics. Like, is

g here really something new? Shoue

n be scared Or is Or is this just the path we've n ?

Maybe you can t been on? Maybe you can reflect

been on? Maybe you can reflect to n the other two speakers to sets optimistic us should we be optimistic or pc ?

Where are we?

You're muted.

muted.

Thank you, Professor Terrance . I

. I Thank you, Professor Terrance.

t um have been thinking about, um, e ome of these things for a while And personally , I am not scared .

Uh, maybe one thing that I just Uh, maybe one thing that I just t wanted to know, which I haven't t or looked into um, read about or looked into,r Maybe uh umm We shall hear more about it We shall hear more about it, but I don't know

whether expert I don't know whether expert gros all l ps and, um, all these ethical as e actually g encies are actually providing a , because It seems like agent ic AI has It seems like agentic AI has br of night ught some form of nightmare at e , that

we probably that we probably haven't seen be this um people , you know, jump people, you know, jumped onto t quickly e wagon so quickly, um, like itn y said.

said.

Until personally I Until, personally, I am not sca I see the benefits of what I see the benefits of what agen especially serious domains like healthcare serious domains like healthcaree Um, maybe what I would want tos e retain the say is that we retain the levelf t we've had of responsibility that we've ha

n other l AI Um when , when traditional Um, when, when traditional AI ts t So, I think that is where we So I think that is where we areo .

Keep keep, um, staying responsible keep, um, staying responsible, g , elegate, um, some of the tasks , and e new tasks o the AI, and maybe new tasks w that ll actually even come up that, f e h, maybe some of the domains ar in the way going to evolve in the way, ifg e we are talking science, in the y .

So yeah.

yeah.

I think there is, at the moment I think there is, at the momentt to not really a lot to be scared t So I got two people who don't So I got two people who don't td d ink that we should be scared riw y excited. I don

excited. I don

ht now. Pretty excited. I don't

know, I think I'm with them on y excited.

excited.

ome level. I'm pretty excited. r

l , but feel the fear as well, but may na, you Reflect a little bit on that Yeah I 'm I'm very scared um I am very scared iff universities , and I want to specify

universities and specify universities and organis n ations, scientific organisation As opposed to big tech .

If universities and other If universities and other intell s step up ectual organisations step up an good role play their common good role, I y .

g . I'm

. I'm That's not what I'm seeing. I'mg

um seeing big tech, um, driving th .

Um they have very little Um, they have very little incen guardrails ive to put guardrails in place, y seen e and we've already seen the futif t ity of most of the current guar I'm sure you're following I'm sure you're following the sf t cases around ory of the court cases around ao far, least two court cases so far, e

s Um very clearly gave Um, very clearly gave instructi r user to ns to their user to commit suice e are seeing e de. We are seeing, um, we are s

de. We are seeing, um, we are s that all you have to do is that all you have to do is thret ten an agent that you'll turn i off, and there's some great st dies that have shown what happe n agent is s when an agent is threatened wg th being turned off, with havine d .

And how it will then And how it will then completelyr e its a ils it'll override its guardrails, it'll and over your details to hacker h .

So all of this is not So all of this is not, is not s worry about a future me kind of worry about a future g This is this is happening herew I am deeply and now. So I am deeply worrie

and now. So I am deeply worrie I wouldn't be worried if I wouldn't be worried if univer p ityy if universities stepped up e And I also wouldn't be worriedf advances if all of these amazing advance e our , which can drive our understan t ing of people on the planet, of of Scient ific breakthroughs in

Scientific breakthroughs in unb wouldn't be as lievable ways. I wouldn't be as

lievable ways. I wouldn't be as worried if it were if these wer t into e being brought into a society valid human flourish hat really valid human flourish e ng, that took care of the vulne ourselves able, that saw ourselves in coml unal, understanding ways. But tt

t The world in which this is being The world in which this is bein k advanced. So I think we also nd

advanced. So I think we also nd link this to r ed to link this to our economic s that and political systems, that gen e agents rative AI and these agents are g Um handed out Um, handed out in a very indivic o -sum ualistic zero-sum hyper-capitt Um and so, you know, Um, and so, you know, you lose

cares?

cares?

our job, who cares? That's not

Um, oh, y problem. Um, oh, you don't ha

y problem. Um, oh, you don't ha buy food e any money to buy food? You're

starving on the street? Who car

problem.

s? That's not my problem. In a

ifferent kind of political envit onment, a different kind of ecoc it would be omic environment, it would be a y ed.

azing if we were all unemployed write poet We could sing and write poetry Uh walk by rivers , and do Uh, walk by rivers, and do all e walk by rivers, and do all thos that incredible human things that wy very rarely can do, because we But re too busy working. But I don' y .

So uh yeah so to end So, uh, yeah, so to end off my d about talk ather roundabout talk, uh, or c ather roundabout talk or commens t Our intellectuals stepping up Our intellectuals stepping up tr around e around rdrails, around care, around co d holding big tech

to account for the big tech to account for the tere some of these ible effects that some of thesee agents are already bringing to Maybe , I think we have two Maybe, I think we have two acads . Both

. Both mics right here with us. Both om

s them from universities. What'sr

t

? I

your feeling on that? I think tt point.

point.

at's a good point. Are we respoy ?

We would like to take that Jackie Yeah , go for it . Please

. Please Yeah, go for it, please.

Yeah thanks . So, from

. So, from Our engineering faculty's Our engineering faculty's perspt ctive at Northwest University, e spend a lot of effort into thin e empower ing how do we empower the next ?

And to prepare them for And to prepare them for this ne world, because AI will not go .

There's no way that a Pandor There's no way that a Pandoraa s open, and now andora's box is open, and now w ers have to deal with the monstersf .

Another question is , me as an Another question is, me as an er y responsibility now ucator, my responsibility now is , in all my lectures and all my classes from the engineering si e, I need to teach my students I ow to use it ethically. I need

o , because Indust industry is asking for industry is asking for these sk companies lls. The companies at the confe

lls. The companies at the confe ences and the places I go to, tg o ey're asking, okay, who do I em s the new student loy? Who is the new student thao

loy? Who is the new student thao t to join my ?

Um and the question Um, and the question is not, ok e less students y, I need to hire less students t , but because I can do it, but they a I hire that k, okay, who can I hire that co s e and help me build these tools let's say let's say, solutions to our pros w this is e lems. So now this is a massive s

s hange for us, but that's why we t we e hange for us, but that's why wee don't shy away, but we're rathel don't shy away, but we rather m ask , uh, use these tools ask, uh, use these tools proper , and w y, effectively, and how, from ag engineering perspective, you nd a professional ed it's a professional, uh, cars o eer as well, so we need to be rd

s well, so we need to be regist at Exile's at Exile's Professional Enginee s professional engineers , and as professional engineerso a specific you need to have a specific st , you need to have a specific s ndard unto which you need to co With with, in terms of with, in terms of your what you , and I think do as an engineer, and I think responsibility, that is just our responsibility

is to teach that, to model thao s , and to show the students how hat works without shying away, t very .

As to the dangers and the As to the dangers and the challs we can't run m nges, but we can't run away froe if we don't at us up alive if we don't face And I think the best way for And I think the best way for uss world, to is to jump into the world, to dr scover. And the more we know, te

scover. And the more we know, te more e more we discover, the more I see I get my hands dirty in the, the more I can be an the more I can be an advocate f bad as r either the good and the bad aI well, because I can see what's .

And therefore , me as a research And therefore, me as a research proper work r can do the proper work to edu at conferences ate the people at conferences o t at our student gatherings, or Now, to think through Now, to think through these chas , lenges and this new world, we fn I think und ourselves in, but I think t e at that's our perspective a bitd w

we approach that.

Ms. Swan yana I agree entirely , but Yeah.

I agree entirely, but also Yeah, agree. With

Yeah, agree. With who? I don

who? I don

Yeah, agree. With who? I don't

But also Yeah, no, I I Yeah, no, I, I agree with Yeah, no, I agree with Hamil.

But also, I just wanted to uh Say say something about um say something about, um, what e reflects.

reflects.

So when we say things like So, when we say things like goa, s are definitely s, the goals are definitely coms The biases that are reflected The biases that are reflected ie m us.

us.

Everything that is bad in there Everything that is bad in there o , I don comes from us. So, I don't wanto a conversation about to start a conversation about u Before we let AI out there .

Because the agents that you Because the agents that you giv t is entirely a goal, that is entirely your l of the oal. It is not the goal of the t

oal. It is not the goal of the t the day.

the day.

And so, whatever domain you And so, whatever domain you tryo e agents in, It's definitely going to be It's definitely going to be enty is e rely you that is responsible fo the agent is doing .

I hear you , I hear you.

you.

It's I hear you, I hear you. It's th

't kill t guns don't kill people, peopl don't know if that do, but I don't know if that'sy We've entirely certain. We've got a sf

entirely certain. We've got a sf our t of questions from our audienct n and I do want to touch on theme think Prof McQueen a, you'll be able to Prof McQueena, you'll be able td panelists respond to the panelists and ae swer the questions at the same f the questions ime. One of the questions is be

ime. One of the questions is be ime. One of the questions is bee

ime. One of the questions is bee ng, how does this change the ac s demic landscape? How does this

demic landscape? How does this change our academic landscape change our academic landscape. u

o aybe you can just respond to th Yeah that 's a great Yeah, that's a great question, I m nd I'm gonna I'm gonna respond byy by to that question by totally agrg s eing with Herman's point about literacy.

literacy.

And that has to be a And that has to be a fundamentar part of our jobs, all of our js , I see bs. At the moment, I see univer

bs. At the moment, I see univer in bs. At the moment, I see univerg

bs. At the moment, I see univerg in ities replying inn responding i two ways, and I found both of somewhat problematic somewhat problematic, or at lea . The first

. The first t narrow. The first way, which

t narrow. The first way, which way, is s the dominant way, is really ad . So

. So ound academic integrity. So we a y saying, if pent a lot of energy saying, if you use AI, we're going to catcu o punish you you, we're going to punish you m It didn't work with plagiarism I don't know why we think it'sk AI gonna work with AI, that respon the pedag e. It totally destroys the pedal

e. It totally destroys the pedal .

But , um, but that seems to be But, um, but that seems to be w is being ere most of our energy is beingt of our energy is being expend, h, is being spent. Wee I keep h I aring academics saying, I've gon I -proof to design an AI-proof assignmem , dude t. I'm like, dude, that ship ha

t. I'm like, dude, that ship ha So it really worries me So it really worries me that th g t's where we are spending, that s where we're spending our ener that doesn't help y. Um, I think that doesn't hell

y. Um, I think that doesn't hell Um, the other response that I Um, the other response that I ss ation which e is domestication, which is inu dustry's going to require you te to train you use AI, so we need to train yo . And I think

. And I think to use AI. And I think that's response, but bit of a better response, but k a partial one still think it's a partial one and that's why, as I say, I reh lly agree with Herman's critica AI literacy. We can't run awaym

AI literacy. We can't run awaym .

It's here to stay Um, and the more we know Um, and the more we know, the m for l re we can advocate for ethical s se. But it's got to go beyond jt

se. But it's got to go beyond jt how to use it in cost-eff ective how to use it in cost-effective n hancing ways s and profit-enhancing ways. It's

gonna go it's gotta really grap Um what does it mean to use Um, what does it mean to use AIn t are cognitive in ways that are cognitive or og to floading? What does it mean to n

floading? What does it mean to n that may ely on AI that may introduce ha d botch it?

botch it?

lucinations and botch it? What

um What does it mean ? And

? And What does it mean? And to pick n p on Tezera's point about the bs it's not that the ases, it's not just that the bi ses are inherent in the trainin g data and now beingg they are b d , but ing escalated at scale, but it' more than that. It's the syste prompts, and I think that's imt It's way beyond the It's way beyond the training da The system s that

a. The system prompts that reque

a. The system prompts that reque you are really re, or unless you are really go many of us d, and many of us have done thi , to try and tweak the way in w or others Your system prompt is one Your system prompt is one of ab olute confidence in doubling do s . Um, that's

. Um, that's n on errors. Um, that's the def system prompting ult in the system prompting, and , we need to we need to train, we need to gl g t critical AI literacy amongst , To help them to put in place the To help them to put in place th l kinds of prompting that will p kind of stuff event that kind of stuff from eg um I think that's what

I think that's what critical AIy .

It's literacy looks like. It's going

back to fundamental questions.

o hat is knowledge? Who's it for?

e Who does it serve? And what doe how you use AI So I'm hearing Ost rich approach is not the Ostrich approach is not the apph got to get the oach. Yeah, we've got to get th

oach. Yeah, we've got to get th head out the sand, and we've g s academia t to face this as academia. We'o

our research e got to face it in our researc s me . It just frightens me this ide

. It just frightens me this ide the actual research the actual research we're produ tered with ing is being lettered with refe even when the ences, and even when the refere e , it could ce isn't fake, it could actuall e to the wrong be a reference to the wrong are Yeah .

be a reference to the wrong ar e Yeah.

The whole network is corrupt The whole network is corrupt oft w , and what we know, and the network oe .

knowledge that we have. It reay

that ly it's something that does doe bit e frighten me a bit. Anyone else

t or I want to respond to that or I've actually got some other little uestions here that people are id maybe Mr. Blackie , someone Mr. Blackie, someone is asked a are you using out what tools are you using sog Ah ething that's a little bit lighr h .

s ething that's a little bit lighl er, yeah. It's a little bit lig

er, yeah. It's a little bit lig Oh nice .

er, yeah. It's a little bit lig So basically my workflow is So, basically, my workflow is, e either in hese days, I live either in Opes , or Claude C ode . Those are

. Those are my two or Claude Code. Those are my twt limits by . When I hit my usage limits by

. When I hit my usage limits by a Wednesday on Claude, I jump o x to work for er to Codex to work for the res .

Um, and anyway, so that's And anyway, so that's that.

As also being an admin, I'm also As also being an admin, I'm also for using something Paranoid for using something li Co-work e Claude Co-work, or some of the t use my desktop se agents that use my desktop a and do stuff like d click for me and do stuff likI d click for me and do stuff lik, It fre aks me so out when I do It freaks me so out when I do se t

, and the AI ftware development, and the AI oding agent would actually opena r , it would up a web browser, it would go a the app that d click and test the app that y o u're building. And I'm like, no

u're building. And I'm like, no You freaked me out, You freaked me out, don't do th a hacker t. It feels like a hacker takiny

t. It feels like a hacker takiny , so that's over my computer, so that's whI m always just leaning into always just leaning into, I wan to build the code, give me thee e me codec code agent, give me codecs, giv code.

code.

And that's .

24 7.

Onyana , what you got there?

got there?

What you got there? What you gor What you got in your toolbox?

Uh, Major League Cloud C odes Uh, Major League Cloud Codes, a lately I have d, um, lately I have been prepa material using ing my material using Docysauruo using , so I'm actually using Cloud Ce make all my docus eries.

Um it has actually Um, it has actually, um, saved e e a lot of time, and the materi for all My students , yeah.

Excellent excellent .

Excellent, excellent. Personall

Excellent, excellent. Personall'

t , I'm a traditionalist. I'm tryg

using agentic AI to do my work for me agentic AI to do my work for me.

agentic AI to do my work for me Can I Cannot dump Yeah Can I jump in with Yeah.

Can I jump in with a with a poll Yeah, please.

Can I jump in with a with a polt Yeah, please. Yes.

Yeah, please. Yes.

Can I jump in with a with a pol itical comment. Yeah, what well

itical comment. Yeah, what well f 3 for , with 3 out of 3 for Claude, u various tools , and Claude's various tools, u , and totally agree with, with, not uh, Herman's point about not le .

And it's so easy to allow that And it's so easy to allow that n 'm gonna make a o happen. But I'm gonna make a t

o happen. But I'm gonna make a t h olitical statement here, which l ou may not all agree with, but , , lease please please please e G lease, if you're using ChatGPT,t .

Just stop.

Please go on to Quit Just stop. Please go on to Quitd

Just stop. Please go on to Quitd it is PT and have a look at how it iss g Uh it is funding the use of Uh, it is funding the, the, the Uh, it is funding the use of au omated machinery, uh, weaponry,t is yeah, I it is, it is, um, yeah, I, I, Ie It is, um, yeah, I, I, I will se op here, because not everyone'st going to agree with me, but ChaP

have a look at Palant elationship to Palantir, and ju t it'ss what t have a look at what it's what Okay.

Okay, I hear you.

We do need to Okay, I hear you. We do need tos e that e always make sure that those tec n 't no bros don't they, you're righ.

, they're a little bit insane.

y I agree with that effect . We've

. We've I agree with that effect. We've

got a few other little questione m 's I here. One of them's I don't knw

here. One of them's I don't knw ow, who's willing to take this e it out there ne? I'm gonna put it out there.

ne? I'm gonna put it out there.

us?

Anyone want to put put it on the Anyone want to put put it on th e . Any

. Any line, put a date there. Any ta

I think line, put a date there. Any ta?

I think, I think we I think, I think we're not yet e We're not yet Yeah.

We're not yet there. It's

not We're not yet there. It's not a .

Even when it looks very Even when it looks very promisi t um if g at the moment, um, if you loo e at things like thinking fast a , System 1 and System 2, Um you find that things like dynamic attention um, weigh weighing trade

-offs, Uh that . Yeah,

. Yeah, And things like that. Yeah, mayg

e anticipating some consequence Um we're not like Um, we're not, like, yeah, thos achieved things have not yet achieved, d Right now we Right now, we're still, um, focg t even if it looks broad right now even if it looks broad right no Truth be told, there's still

Truth be told, there's still na.

So Yeah So, we are not we are not Yeah, I hear So, we are not we are not there Yeah, I hear you.

Yeah .

It's actually sometimes It's actually sometimes quite d as humans fficult for us as humans to asss w difficult it ss how difficult it is to do a But and that that that But, and that that that is somes overestimate imes how we overestimate capabi things We've got one other question

We've got one other question. Y

We've got one other question hen e. It's an interesting question

e. It's an interesting question because we've touched on this ia w , but ea now, but I'm interested to s e who's willing to take this on.

what does s on. It's what does human oversy

on. It's what does human oversy What are we When we're talking about guard When we're talking about guardr t ils, when we talk about these t ils, when we talk about these t I 've got the wrong word I've got the wrong word there. e

es is there lacate the masses or is there r t ally a true notion that we haven What does human over sight really What does human oversight reall?

mean? Is anyone willing to tak?

Yeah I willl I will do my Okay good .

Let's say, take my Let's say, take my chances on ts I know one of the ways I know one of the ways that I t is ink about this is I see an AI a y an ent in, let's say, an enterpris system or in a software tool tt g , or whatever, as a user or a consumer of as a user or a consumer of the So for me So, for me, is I

So for me, is I would define So, for me, is I would define t So for me, is I would define ths scope roles, the access, the scope tr to at this user has access to and o . The same way

. The same way an do or can't do. The same wayd t that I would approach it for a e ormal human or employee that yo And therefore , on a system level And therefore, on a system leven a code l , I , on a code writing level, I usy I try to point it as the I try to point it as the code iw .

So, the AI agent cannot override So, the AI agent cannot overrid e code that you've what the code that you've writn the actual en with the actual programmatic u build into guardrails that you build into Can't override that.

So that is how I think of So that is how I think of keepi e human in control g the human in control and the n p , and I know uman in the loop, and I know it kind of defeats this whole notia m ous agent that you just agent that you just totally rel 't think we are ase, but I don't think we are r t

ady nearly as well for that, eve n though the capabilities are te willl ere, the adoption will is laggi .

And I know for me specifically And I know for me specifically,s s and see the is build the systems and see thr t agent as a user of it, and tha Wanyana , you've got any Wanyana, you've got any feeling does this human over on what does this human oversin ht really mean that we keep tal Um , when I say human Um, when I say human oversight,e

responsibility.

responsibility.

I see gate keeping.

Um for KISS, where we have Um, for KISS, where we have aget And maybe um continued, um, And maybe, um, continued, um, sl , whereby quential processing, whereby th t is supposed to be t same output is supposed to bey r processed by another agent and t . I see

. I see hings like that. I see humans cg And uh um accepting or agreeing to what accepting or agreeing to what i like going on, and like Harmon says e I indeed see the human in the loop.

loop.

And the human everywhere in And the human everywhere in the Yeah .

Brooke Mc Kenna Yeah I think I think I would Yeah, I thinkk I think I would gree with everything that's beed g said around around having somef I do think we need stronger I do think we need stronger gual s drail legislations, which inclus l legislation, And I think that legislation And I think that legislation nee

specific about ds to be very specific about wh e big tech itself n the big tech itself is respon t ible, because at the moment, lio other e so many other forms of techno user ogy, it's always the end user w e , and Many end users are ignorant, are Many end users are ignorant, ar e

hap that comes naive, buy the hap that comes Um and they need Um, and they need to be protect I do think e d. I do think we need to have a

d. I do think we need to have a clear legislation very, very clear legislation ar Um where responsibility lies Um, where responsibility lies wo g , and I think en things go wrong, and I thinks for these big companies to for these big companies to alwa know, s say, you know, you signed thes So yeah , I would like to So, yeah, I would like to see mh ch stronger legislation around And maybe

one more thing Certainly And maybe one more thing, maybe Certainly.

And maybe one more thing, maybeg one more thing, Professor Terra Um while we are working Um, while we are working with as systems, we o ents and AI systems, we need to realize that we need to have a g common ommon understanding, common sem, Oh, we need to have the same Oh, we need to have the same ung t the agent erstanding of what the agent is understanding the situation to .

If it's data , we have too Make sure that we see Make sure that we see it the sa y the agent is seeing it e way the agent is seeing it. O,

o herwise, the agent is going to n , and we ake its own direction, and we s all be thinking it has all the t And so this these And so thiss this need for us And so thiss this need for us t s always make sure the agent hasd That we are seeing at the That we are seeing at the momena t, we have a lot of tacit knowl We have a lot of things

We have a lot of things that wep and expect keep to ourselves and expect th r accordingly.

accordingly.

And then when the And then when the agent starts the other way o go the way the other way, we o , start to worry and we say, oh, n , and it's Um executing things that I Um, executing things that I havd o , but it n't told it to, but it's becaus of knowledge we have a lot of knowledge, bo and that we

need to give to that we need to give to the age o be able to r Accordingly accordingly.

Yeah .

Although, if I may, that Yeah. Although, if I may, that s

Yeah. Although, if I may, that s the user ssumes that the user has that t that is I mean, I is I mean, I totally agree withu you to zero. I'm not disagreeiu t g with you here, but when you he vice users, and ve novice users, and here, agai s , I'm thinking about students, , g agents They are unable know They are unable to know what th where, you know

And that's where, you know, thet caught get themselves caught in getti s that are g responses that are at odds wi y , or that can lead h our reality, or that can leadm c ways them in really problematic ways I I mean, it's I mean, it's, again, goes back I literacy o critical AI literacy, that if Yeah

o critical AI literacy, that if, Yeah.

you are using, uh, whether it's Yeah.

you are using, uh, whether it'sl a large language model or whethg r you're using an agent to do ak n task that you yourself don't un then you'ree then you're in then you're in trouble, and we ble, and we need to make sure tt then you're in trouble, and we Hmm.

eed to make sure that ourr that Hmm.

eed to make sure that ourr thate Yeah.

we need to make sure that critil , i.e. our

, i.e. our

, al citizens, i.e. our students,

That uh that you shouldn't That, uh, that you shouldn't be .

to do stuff that you don't to do stuff that you don't havee o yourself.

yourself.

Yeah that's Interest ing Yeah, that's true.

Thank you .

I havee I have Thank you. I havee I have one qn

Thank you. I havee I have one qn n go there, this fore I go there, this is quite we n interesting question we have t rom in the chat. I kind of can e t rame it a little bit for you gu It's I guess it's talking to a I guess it's talking to a littlt post-truth bit about this, this post-trute world that we find ourselves i s , how do

, but the question is, how do wt Human -generated digital human generated digital content d ? Like, how do

? Like, how do Is that something we need Is that something we need to be something that protecting, like something thats by a was exclusively written by a hun That we know this, thiss this is That we know this, thiss this it o our product as opposed to someg an agent hing generated by an agent. Anye

es to ne have any responses to ever t Sort of this notion of Sort of this notion of knowledg knowledge that is human knowledge as opp of that was sed to the proof that was gener a computer or Maybe Hermann code that Mm.

Maybe Hermann code that was gen human versus rated by human versus code gened An agent. Should

agent. Should

And you see And I seee An agent. Should we be protectid

An agent. Should we be protectid g that? Should we concern ourse

g that? Should we concern ourse It's a bad It's strange.

strange.

way to think of it, because way to think of it, because whae s What the agents are What the agents are sort of pro t they got s ucing is what they got from us t And now they're just reiterating And now they're just reiteratint e initially what they got from me initiall

justt represented in a new way and mix represented in a new way and mi.

Now, the thing is I me Now, the thing iss I I me, persI Now, the thing iss I me, person ally, I'm so sentimental with ro g between gards to differentiating betweee gards to differentiating betweee whether I write an article or t whether I write an article or I nother article, when I write, l t's say, a blog or a pipe, uh, e parag versus the AI tool paragraph versus the AI tool hee

For example, to phrase what For example, to phrase what I w e a bit nted to communicate a bit bette Because whether I go too to Grammar ly to start off with to Grammarly to start off with l d he initial idea, and the AI hel e it.

it.

Okay so I see it as thiss collaborative .

let's say artifact let's say, artifact that has be And I'm and I And I think inn maybe in And I think in maybe in the art r or music specifically or music, specifically, that or music specifically, thatt and That can be a other type That can be a other type of quen there's a tion, where I think there's a bt t

t more to it than just writing And whatever , but at the And whatever, but at the end of, the day, I really don't know ho . I

. I we should think of this. I stio

h that notion l also struggle with that notio AI . I don't enjoy listening to AI

. I don't enjoy listening to AI r generated music. I would rathero

generated music. I would rathero And listen, something else.

Um, but um yeah, I reallyy Don't know .

Maybe Um Maybe Um But maybe while you're But maybe while you're respondiI the audience g, can I just ask the audience who has f there's anyone who has a ques ion they would like to pose thes d be very excited selves, we would be very excite . I think the panelists would le

. I think the panelists would le you ve to hear from you, hear your d uestions directly and respond. e

o please just put up your hand e f you've got a question there a We'll prioritize you We'll prioritize you. You can bd e list moved to the top of the list.

moved to the top of the list. t

id someone else want to respondo Here Yeah Here we go, we've got Yeah, I was I Here we go, we've got a mut a So Muta and then ZCore So Muta and then ZCore. So we he s and but ve two questions and but maybe s n ve two questions and but maybe r ve two questions and but maybe a

It was as it Makuena or Manyano who wad It was me, it was as it Makuena or Manyano who waf It was me, it was me.

Yes.

Yeah Yes please Vany Yeah.

and then we have nd and then we have two questioe .

Okay so I was saying that the Okay, so I was saying that the I if you write ast time I checked, if you writ , Um and they Um, and they stand to be correcd You do not put AI ass one of the authors , you can one of the authors, you can ack AI

w owledge that he used AI and how you used AI. Now I'm talking abt s ut academic articles, because tt e a place that I am comfortable in a place that I am comfortable i And so , you will not And so, you will not say, I wro with some um agents, and maybe give um, agents, and maybe give the t

the authors You will just say , I used AI You will just say, I used AI, m f the paper ybe at the end of the paper, ane w you used AI So at the moment, we have not So, at the moment, we have not t d full et transferred full responsibil there ty of what we are seeing there Um

agents um AI agents . We

. We Um, agents, um, AI agents. We an

a e still in an era where a humano for what has to be responsible for what u wrote the prompt heyy so if you wrote the prompt Um it 's still your work.

Until maybe one day , we think Until maybe one day, we think, , , or AI e kay, agency, or AI agents have Um part of us in a Um, part of us in a different s should accord yle, so we should accord them s me, um, individual credit. But r

you're still responsible or now, you're still responsibln or now, you're still responsibla . So, you can't generate conten

. So, you can't generate conten Put it out there . It's

. It's Put it out there. It's offensivn

u distance , and then you distance yourselt Yeah.

I hear you .

Okay.

I hear you. Okay, Adams

please I hear you. Okay, Adams, please l Alright , thank you so Alright, thank you so much for e s .

m he conversations. I'm really in

he conversations. I'm really in And I have been for some time now I have been for some time now i a of Okay.

Okay, I need to start my video Okay, I need to start my video, Okay , so I have been Okay, so I have been for some te d in me really interested in these d AI scussions of AI, and especiallyn And what to me to be an And what appears to me to be an even listening issue now is even listening to n We

all agree that AI isn't doing We all agree that AI isn't doin k great, but I think it is optim r other roles n zed for other roles rather than it where we are trying to apply itn And the reason I say this is And the reason I say this is bee it ause when it replaces our thinkg is clearly s

ng, which is clearly it does th t replaces t when it replaces our thinkinge claim to Actually .

Actually, and advancing Actually, and advancing intelley .

Actually, and advancing intellen I help students I have students postgrad I have students, postgraduate ss Some of them I've tried to use Some of them I've tried to use will o ChatGP , I, someone will go to ChatGPT, a topic ut a prompt, get a topic, a pose from there Then they can use the topic Then they can use the topic, ta

p ilot e it to Microsoft Copilot. Theyn

e say, can you give me a structur n for this? Then you can get a s

for this? Then you can get a s Microsoft Cop You will use that structure You will use that structure, pun r AI do it in another AI platform, do And so my worry is always that And so my worry is always that h e a ven though I can produce a papet y successfully using

in that way successfully usings and series of plums and also editine t come up What has just happened to me What has just happened to me? Ie

e of thinking have lost my sense of thinking I not have ecause I do not have originalit I only rely on creating a prompt I only rely on creating a promps g . It's like asking your directis

. It's like asking your directis we're using ns the same way we're using Gooe r example o le Maps, for example. I can go , and o a place I don't know, and it t on't mean I master the route. Is

m using the means that I am using the info me And Let's see if we And so Let's see if we can get Let's see if we can get that pa r question el to respond to your question e Yeah.

Yeah, thanks.

Anyone want to respond ?

I can jump in as a start.

Yeah please .

Um yeah, I think I think what Um, yeah, I thinkk I think whatt o s really important for us to tak What universities value , and I What universities value, and I d perhaps a little controvers perhaps a little controversiallt universities suggests that most universitiee m value performativity rather th e .

So to get promoted, you need So to get promoted, you need pu t matters less what lications. It matters less whats

lications. It matters less whats t those publications are about, w at impact they have in the worl And what meaning they bring to And what meaning they bring to r t ur understanding, what matters ore is the number of publicatio y , depending on s, and possibly, depending on yn you know, what ur institution, you know, what r l has mpact factor the journal has, ot What matters

in terms of What matters in terms of our st they perform in dents is did they perform in ths ir assessments? Not have they l

ir assessments? Not have they l ?

arnt, have they interrogated, ao , e they able to reproduce? Now, d

s Check check for all those things , but check for all those things, butt most of our assessments are acty for e of ally checking for performance o of memorization possibly of memorization, possibly under generally memor tanding, um, but generally memof ization of a particular, uh, ca So the way ons of knowledge. So the way in which the university is increasy Um

means that anyone Um, means that anyone, any studt o doesn't use AI nt who doesn't use generative A to assist them with their assis academic who doesn nments, any academic who doesn'e I to write their articles for them to write their articles for the d , has basically misunderstood te , because

e rules of the game, because an am being controvers d I know I am being controversi y l here, but just to make my poi hyper , Um if the university is Um, if the university is rewardg e and not giving Much care to substance . Anyone

. Anyone Much care to substance. Anyone t

m properly s missed the rules of the game.

the game.

missed the rules of the game. Sd

, perhaps, before we we need to, perhaps, before weh , whether worry too much about AI, whether We need to be going back to the We need to be going back to theg t very big questions about what ds get es it mean to get a higher educn tion? What does it mean to have

tion? What does it mean to have transformational a personal transformational relp ?

What do we do as educators instill the value of lectual cognitive work? What do

we do as researchers to genuine Make an impact in the world.

world.

And those conversations are not And those conversations are not evidence.

evidence.

in our meetings , in our faculty in our meetings, in our faculty s board meetings, our senates, orn r reward So in many ways So, I think that in many ways, I enerative AI is actually exposi a kind of problem that a kind of problem that pre-exisd .

And so yeah, of course our And so, yeah, of course our stu ents are using it. That's what

.

Awesome.

Yeah so true.

true.

Could I ask Brandt belong brand or Zanette Jansen.

Jansen.

Zanette Jansen, would you like Zanette Jansen, would you like k a question I see you have some questions I see you have some questions ie Hi There we go Hi.

There we go. Milan

There we go. Milan, how are you Good off doing. Doing

Good off doing. Doing well, andf Please fire away, fire Please fire away, fire away at Lovely, thank you . I just have

. I just have Lovely, thank you. I just have g question regarding who essenti e Uh in our education Uh, in our education institutio Um since they are not usually Um, since they are not usually f e tools and he developers of these tools an But they are the deployers.

of it . So there's

. So there's of it. So there's multiple peop

of it. So there's multiple peop inst uh brings these applications brings these applications into e explores the use of the explores the use of the applicas n ions, and then ultimately decido t .

s to deploy it. What if uninten comes to the student ional harm comes to the studentf or the staff who take responsib?

Thank you .

An Thank you.

Anjana Any thoughts I am thinking about it.

about it.

Who's ultimately respons ible Yeah I I don 't think at the end of the day, at the end of the day, the desie iss who is going to be who is going to be responsible r that or the output that you generate the user has to be responsible

. And so

. And so has to be responsible. And so,

that hatever output that you generat these tools You have to be able to You have to be able to go back as an academic run through it , and indeed run through it, and indeed belis what you want ve that this is what you want t And if that's not the case, And if that's not the case, the g e is no point you putting it oue The

moment it goes out there Regardless of the tools you use Regardless of the tools you uset p um you can't call up, um, OpenAI d Um guys , you're responsible for Um, guys, you're responsible fo.

I hope I did not misunderstand I hope I did not misunderstand No, I think you got it right No, I think you've got it right.

No, I think you've got it rightk say I think some of us might say t should be a shared responsibily Certain ly as Certainly as the user of a tool Certainly as the user of a tooll with some responsibility must e .

Okay.

with some responsibility must n ie with you. Okay, Zanette, I t correct?

correct?

Yes.

Yes, thank you so much Yes, thank you so much. Thank y

panel. It's

u to the panel. It's such an ing y interesting eresting, absolutely interestind presentations and discussion.

y to repeat the 'll try to repeat the question section So we do live in a financially So we do live in a financially,n d you know, driven world and cryp making a ocurrency is making a, you knowI I don't understand cryptocurre l , but my question cy at all, but my question had o t 80% world

o do with that 80% of the world population that are not even usg Will there be any kind of Will there be any kind of ethic counter-te y l or counter technology, you kn see that being w, do you foresee that being pu , at the moment in place by, at the moment, wod s or l ld governments or ethical agencs that % that are es to protect that 80% that arey

n possibly, well, they don't have l or the knowledge prowess or the knowledge prowess to parp in developing world icipate in this developing worl Well, I don't Well, I don't know who wants toe .

I can say one or two things um The first is The first is that earlier, I red e ommended that everyone follow H , the mathematician nnah Fryer, the mathematician. o

'd now like to recommend that y , um, Stephen Sidley u follow, um, Stephen Sidley, wa u follow, um, Steven Sidley, wh Um, and he has a fantastic Um, and he has a fantastic blog He's got a book that's just co cryptocurrency e out on cryptocurrency, um, bua he's really looking at the int and And society generally but he And society, generally, but he, certainly looks at he, uh, he certainly looks at ht

ertainly looks at how that inte way above my e . But,

. But, is way above my pay grade. Butm

But I definitely think that uh We need we need to think about that 84 %.

And so , of course , the user And so, of course, the user is e 's idea d sponsibility, and I think that It can't just rest with the It can't just rest with the indl . It can't

. It can't vidual user. It can't rest enti

vidual user. It can't rest enti university ely with the university who selo cts to make available certain t s , but I do y ols or others, but I do worry tn at South African universities j 't necessarily e on't necessarily bring the crity cality that they would bring to s .

to decisions about which to decisions about which, um, w ich kinds of software should bee available to made, um, available to their st Um including now, um, AI Um, including now, um, AI agent make sure that eded to make sure that their stf k ff use it. So I think that univ rsities do bear some responsibi niversities do bear some respon.

responsibility , but I responsibility, but I definitel e designers think that the designers of th do need to be, um, held do need to be, um, held respons and ble too, because and that's whee l ation re national legislation, and let n the e 's not even open the door to th conversation about the embarra

South Africa's sment of South Africa's draft A u policy, but those of you who wh rked through that, you'll see t very lacking at it really was very lacking i d terms of conversation around gs Um it just said that, you know, Um, it just said that, you know Stellenbosch University's gonn , I don't know, have a center f d weaponry r automated weaponry, and who k t doesn't

ows what else. But it doesn'tt idn't really speak about how we re going to protect the 84% of . And if we

. And if we eople in the world. And if we a t an individualistic e not an individualistic hyperc world , but actually a caring world, but actually a caring soy d take iety, we would take concern for 8 4 y .

Thank you.

you.

I think we've sort of Thank you. I think we've sort od

Thank you. I think we've sort od reached the end of our questioI s. I know there's other questio

s. I know there's other questio s there, and I'm sorry that I d get us to dn't manage to get us to there.

g We have a conclusion coming up,o and so we've just got a few mors .

I really just minutes here. I really just wa

minutes here. I really just wa t to give our panelists an oppo tunity to maybe just say one or close out You know, it's kind of You know, it's kind of a rest o increase our fears f years or increase our fears, r if you want to be the hype mam m ist or the pessimist, the optimist in your last sentence, maybe wet n can just start with Herman Blace

ie there. You can just give us t

ie there. You can just give us t s to leave h our last words to leave us with Yeah thanks, everyone . This was

. This was Yeah, thanks, everyone. This wag

y enjoyed.

enjoyed.

I am having this discussion, a I am having this discussion, an words would bee approach it with confidence approach it with confidence, th s se AI tools and with curiosity, h a sense of but also with a sense of respon And uh And uh and yeah, so that's And uh and yeah, so that's that Thank you so much . And

. And Tazir a.

Thank you so much. And Tazira,

s .

Um feel free to explore.

explore.

there are a lot of benefits .

Um um, very many enterprises Um, um, very many enterprises, y s are real ery many domains are realizing Um just do it responsibly Thank you.

you.

Prof McKen Thank you. Prof McKenna, last w

Thank you. Prof McKenna, last w .

Uh yeah, I'd echo the previous Uh, yeah, I'd echo the previouss Uh, yeah, I'd echo the previous speaker's curiosity, I'd like ts .

Always asking questions about Always asking questions about w ose interests are being served.

ose interests are being served.

ose interests are being served.o

ose interests are being served.r

And of course, to everyy all, u You know, we have academic You know, we have academic freem c freedom t Just a right. It's

a right. It's

Just a right. It's a responsibi e getting involved ity. So, are we getting involve

ity. So, are we getting involve s ?

Um I think , as Taryn said Um, I think, as Taryn said earl er, sticking our head in the sa.

It's it's too late It's it's too late for that, I' .

Thank you so much.

Thank you so much, guys . So

. So Thank you so much, guys. So thi

d of our brings us to the end of our pad I know it was el and Q&A. I know it was all je g mesh , but I thought it was very mesh, but I thought it was very exciting, and I really enjoyed e h he discussion with all of you.

d over ives me the honor to hand over Prof As lam Fat tah to Prof Aslam Fattah to just concle , if I de for us, if I can just quickle introduce him before I hand ovo m prophet ize research prophetize research and develop r ent professor in higher educati d transformation n and transformation at Stellen

d osch University and a member of the Academy of Science of Southa Africa. He's a former editor-in

Africa. He's a former editor-in chief of the South African Revi c ation He has authored w of Education. He has authoredd books or co-edited 14 books, publishe 0 academic over 150 academic articles ands and super vise 25 PhD and supervise 25 PhD students, students hat's a lot of PhD students. Hi

e the recent books include the Educah s and Experiences ional Pathways and Experiences f Stellenbosch f Black Students of Stellenbosc And cultiv ating an ethics of And cultivating an ethics of be in planet uty and excellency in planetary He currently leads times. He currently leads the dl

times. He currently leads the dl e , democracy and alogue, democracy and Developme cross-c t Project, promoting cross-cult educational ral and interfaith educational t ork for inclusive development a Over to you, Pro fita.

Thank you very much I look forward Thank you very much, Chem.

I look forward to your conclus I look forward to your conclusi Thank you , what an amazing Thank you, what an amazing, uh, conversation set of conversation. A mind-blo

ing, actually. I haven't made a have many notes as I have in a longe the live y time. Thank you for the lively

time. Thank you for the lively provocative onversation, the provocative tie got us going.

It is a bit charming It is a bit charming, but of co unsettling rse it's unsettling. Uh, I wanto 3 kinds of to make 3 kinds of comments, co l ceptual, methodological, and mae d inal.

If I may.

may.

Umm The discussion today made one The discussion today made one t We have moved beyond AI We have moved beyond AI as a to what is called agent ic AI.

that they wait for a prompt, uh, that they wait for a prompt, uh that they wait for a prompt, uh sort of, um, gene AI, generati sorry, Jin AI, generative AI, agentic AI h, we are now into agentic AI t terrain Where systems can coordinate Where systems can coordinate tas e our actions d ks and execute our actions and .

Um and as Tazira Wany Um, and as Tazira Wanyana explas these systems e ns, these systems integrate gene what rative capabilities into what s e calls, and what the literatur architect calls, cloud loop architecture The goal-direct ed adaptive The goal-directed, adaptive, mu Adam ti-step systems built on Adam S LL M sorry

, and connected LLM, sorry, and connected to tos act ls that allow them to act. And e

think we've got to just kind o what way we are in regard to what way we are in regard to ths But I think that this should But I think that this should ner push us into panic .

No naive celebr ation No naive celebration. This is te No naive celebration. This is t e attitude and all point I'm ma.

Our first scholarly Our first scholarly responsibily , is to e ty, I think, is to just a littls d premature t bit suspend premature judgemen o just long enough for us to see clearly.

clearly.

You know, if you go too hard You know, if you go too hard wi l foreclose what we may see .

What is this thing ?

How What is this thing? How does itk Where is d ?

work? Where is it embedded? Wha?

What forms of dependency does What forms of dependency does i Because it is, uh, Because it is, uh, technical cly the rity helps us begin the basic q Hermann Black ie helped Herman Blackie helped us see Herman Blackie helped us see the systems As ians can work across Asians can work across applicat data and personal ons, the data and personal works e

lows, and many of us are doing lows, and many of us are doing o t. I prepared these notes basedn

t. I prepared these notes basedn on working across these applicas They can assist with They can assist with literatures searches, organize tasks, solves .

And for universities And for universities, there maye improved be real benefit, improved produy better student support tivity, better student support,e new forms f Academic academic assistants, and so on.

so on.

So we just should not discount So we just should not discount d technoo Um, they just do they just do a techno-optim they just do a techno-optimistit ic AI insight. Agendic AI can indeede

insight. Agendic AI can indeede .

It can extend human capability.

capability.

I'd let her work with I led to work with this human cy pability a bit more, uh, concep g ually centering and more concep f a way.

a way.

Because Herman indeed, um, Because Herman indeed reminded I is changing the s that AI is changing the acades e .

t ic landscape. His point about ac

ic landscape. His point about ac c ademic critical academic literay cy is then underscored by the t s .

Uh we can't simply teach Uh, we can't simply teach peopl s .

We must We must teach them how to underd s .

Prompt responsibility, as prompt responsibility, as Profe s sor McKena argues, verify outpu ure of systems, And judge what kind of And judge what kind of knowledg .

System prompting, that's System prompting, that's what It m calling it. I'm running off P m calling it. I'm running off, a System prompting must become our academic literacy .

It is mere ly it is not merely It is merely it is not merely al technical skill. Um, and we hak

technical skill. Um, and we hak e to ask our universities even g eginning to think about doing t Um, so we have to ask questions Um, so we have to ask questions that what knowledge is being ca From which sources ?

with and within With what assumptions and withis whose interests? These are ther

whose interests? These are ther kinds of questions that our cri should be able to surface in should be able to surface should be able to surface and t Sue McKenna pressed us into Sue McKenna pressed us into tha n .

A warning is not that we should A warning is not that we should .

Well, of course , we cannot . A

. A Well, of course, we cannot. A ce

ncern is that we must not allow cern is that we must not allow for what she calls cognitive for what she calls cognitive ofd .

Univers ities must learn to use Universities must learn to use cost-effective I in cost-effective critical anl But the danger is not But the danger is not simply th tasks t agents will do tasks faster, . The danger

. The danger is that they may Which which academic uh uh which academic, uh, through whi capability is formed.

h academic capability is formede That is a key issue that we hao that you have to take on board.

academic How do how does academic capabiy t formed?

formed?

Because academic judgment is Because academic judgment is buh g , searching failing, Patience time , not Patience, time, not compressingf , all of these things, interpreti, c . And

. And g, revising, etc. And as she ar, Learning the desert norms of Right, so that is what we Right, so that is what we give o quickly p when we move to quickly into n If agents re, um, remove If agents remove too much of If agents remove too much of the , this s struggle, right, this real im ortant academic science reason .

capacities They may weaken the capacities e p possible hat make scholarship possible.

n nd here, Hermann's point about g n eeping the human in the loop bes But the human cannot merely be a But the human cannot merely be .

The human must remain intellect The human must remain intellectp judgment ally capable of judgment, and mg o deeply ving too quickly and too deeplyn e and too unquestioning into the means that we gendic AI domain means that we d e Suspending the processes by Suspending the processes by whie e intellectual h we acquire intellectual judgm

That is why Sue's concern That is why Sue's concern about .

is so important .

When an agentic system produces knowledge thatl Who is eligible for it?

it?

The user or the developer ?

or the institution or the funder , or the platform.

platform.

So as she says, God reveals So, as she says, God reveals ma.

But what happens , in fact But what happens, in fact, my qn when o says, estion, when what she also says s or under e or operate beyond or operate beyond these guardra Okay with the God Realty Okay, with the God Realty in pr e are smart ise that people are smartly usi e

agents in order to g the agents in order to pay pa Govern ance becomes Governance becomes an afterthout that.

that.

ht, doesn't think about that. I

o the has to be built into the instin ?

Do we have these Do we have these guardrails bui university t into our university's institun In the ional design? In the research en

ional design? In the research en hics, in our assessment systems s in our procurement decisions, n ? What

? What s has been argued? What are we s uying? What kind of tools are w?

uying? What kind of tools are w?

How are we setting up How are we setting up our elemes f ts? What kind of academic critil

ts? What kind of academic critil g al literacy training are we doi The ethical question is, sharpened in our unequal world.

unequal world.

sharpened in our unequal world.n

This is an important point thats e .

c aya will gets made here. Agendic aya wil d much arrive on a playing fieldt that has not arrived on a playis g g field. It is not functioning

g field. It is not functioning d urrently on a playing field, on Some institutions will Some institutions will indeed he , clean data ve powerful tools, clean data, l strong AI echnical support, and strong AIy t of r literacy. But most of our instie

literacy. But most of our instie utions in this country are faci e Expensive access platform Expensive access, platform depe expensive access, platform deped .

And if universities then go on And if universities then go on, t in that context, to adopt agend n ting these ng AI without confronting thesee structural inequalities that Prd y indeed sera did, they may indeedd they deepen what I call epistemic strat what I call epistemic stratific

.

In my own work , I described In my own work, I described thin c as an epistemic reconfiguratio of the universe. That's what's

Universities are going on here. Universities areg

.

It's losing its intellectual It's losing its intellectual prt from s to platforms , to agents very to platforms, to agents, very uy I critically. AI is becoming partf

critically. AI is becoming partf e infrastructure through of the infrastructure through ws ich knowledge is produced, circ , and acted lated, interpreted, and acted u So the deeper So the deeper question is wheth will their r universities will retain theic Not handed over to AI tools.

AI tools.

Use it critically . By

. By Use it critically. By epistemico

, you can sovereiginity, you can guess th t the human t I mean that the human and ins o interpret itutional capacity to interpreto and to contest and to judge and to take responsibility for knows n edge. That's what's on the tabl

edge. That's what's on the tabl Universities can use AI Universities can use AI. They c

n with AI n n learn with AI. They can benefm , but they cannot t from agents, but they cannot e ethics utsource judgment, ethics, plurr .

lity, or responsibility. And I o t if we ant to plea that if we run unqu n to plea that if we run unquestii ningly into this domain, we areo sourcing e But there are a process by whi But there are a process by whic a university we remain a university, and the se are the processes that are i o We shouldd Rochelle, should We shouldd Rochelle, should we d

c cared, optimistic, or pessimist?

I would say we need a kind of disciplined hope.

hope.

Not fear ? Not

? Not hype Not fearfulness? Not hype, not l

Not fearfulness? Not hype, not l surrender Disciplined hope means Technical technical understanding , ethic technical understanding, ethica l governance, critical academic And a commitment to the public And a commitment to the public s ood, and the university's commi Let me conclude by Let me conclude by saying that where Asaf

his is where Asaf has an importt .

needs nt role. South Africa needs sob.

nt role. South Africa needs sob.

.

Public publicly trust ed spaces publicly trusted spaces like th.

Had to exist there Had to exist at, where had to exist at, where these qut stions can be addressed without t na m panic and without naive optimis that enge , and with that engenders a spaf A space of convergence, a space A space of convergence, a spaceh of synthesis, because if we're wide varieties of peaking across wide varieties o Disciplines

have certain kinds Disciplines have certain kinds e f behaviors. Some run in this d

f behaviors. Some run in this d s rection, others are running in e he ethics direction, others area Knowledge direction . We have to

. We have to Knowledge direction. We have toe

Knowledge direction. We have toe a have build a kind of a trust whe ere we are going to suspend a sf bravado own Our own bravado, and our own laf k of ability because we've got f find . Because

. Because Because with South African, Because with South African, it' Multid isciplinary , interdiscip Multidisciplinary interdisciply n ary, nary, transdisciplinary, public d spaces where these y trusted spaces where these qu d stions can be addressed without panic and without naive optimis So in conclusion, we need So, in conclusion, we need to ay

sure's clarity about the archit sess clarity about the architec.

We need Her mann's insistence We need Hermann's insistence ony and design literacy, guardrails, and desig We need Sue 's warning about We need Sue's warning about resy illing, and onsibility, de-skilling, and thf e future of the Knowledge Projec The task before us is not simply The task before us is not simpl .

It is the It is the shape it democratical Educationally educationally and ethically .

And our responsibility is to And our responsibility is to pr , concept serve human judgment, conceptua, Epist emic plur Epistemic plurality, very impor um ant point that, um, Sue made abt , or how much n ut losing, or how much homogeni Preser ving public trust and the Preserving public trust and the y

capacity to know wisely and eth colleagues Thank you so much for that.

for that.

Cool .

If I may hand over to our Host To close out for us Um, thank you very much , Terence Um, thank you very much, Terency Um, thank you very much, Terenc . I must say, I'm feeling like

. I must say, I'm feeling like t hat actor that has just receive an Oscar, and I'm walking up t And I'm taking the mic, um, And I'm taking the mic, um, in,m receiving um, and I'm receiving the mic, m g , sort of, you know, inwards and outwards.

outwards.

as to what has just happened .

Um the last two hours Um, the last two hours. So, so

eally, there are a lot of peopl to thank, and I really want to for taking us through this whole for taking us through this whol So lighthearted ly, but very So lightheartedly, but very conr between iderate at navigating between q s , and estions, panels, and, um, reall as such.

as such.

Thank you very much for being Thank you very much for being p discussion.

discussion.

And then I really would like And then I really would like to l ists. Panel

ists. Panel thank our panelists. Panelists,t

t yourselves you cannot doubt yourselves, um after this. You can see all th

after this. You can see all th That is that is following in the chat that is following in the chat b.

Um thank you very much for for Um, thank you very much forr fo But also bringing us back But also bringing us back to wh we But also bringing us back to wh In this very difficult In this very difficult, sort of, agentic AI um, theme of agentic AI and wht g about , especially in research,

and in services, Um, and in services, etc. Askin the real questions. Thank you h .

Professor Pattal Professor Pattal, I couldn't dot u very much it better. Thank you very much r

it better. Thank you very much r everything together or bringing everything together a incredible sort of, umm ceiling and the discussions, um, ceiling and the discussions, ume had in that we actually had in this t.

h o hours. Thank you very much fo

o hours. Thank you very much fo .

I'm just wrapping up I'm just wrapping up and bringi r .

And then , of course, to our And then, of course, to our par h icipants, thank you very much f c questions r the fantastic questions and k . It was

. It was eping the debate alive. It was

And to see , um, what is And to see, um, what is being sd And to see what is being said ae d what's being answered, and wh r mind, I t goes through your mind, I muso h say. So thank you very much fo.

say. So thank you very much fo.

And then, of course, this cannot And then, of course, this canno t what I now happen without what I now, fort r content the moment, call our content te.

He did not feature today Martin He did not feature today Martinu He did not feature today Martinh Thank you very much for alwaysg h these nice, Um, areas and themes Um, areas and themes for us to s Please keep your, your Please keep your, your, um, cree Please keep your creative juiceg e not flowing, please not, never clot

And then thank you also to And then thank you also to Teree um Tommy Meyer , that also out and ads you from nd out and advises you from the h side. Thank you very much. Last

side. Thank you very much. Last

but not least, thank you very mh y team.

team.

, For all the effort that For all the effort that you've n , of course, u ut in, and then, of course, you um Renewal um, renewal way, um, in our, um, um, renew way, um, in our, um, , Nad a Um everybody from my team that Um, everybody from my team that m

supports us here from this side k I do want to thank you very mu, And , um, I 'll close the meeting And, um, I'll close the meeting.

But I do have one request , and But I do have one request, and s e any hat is please, if you have any s r hemes for us to consider, becau w already The work has been done for this The work has been done for thist webinar, but now we have to thi webinar.

webinar.

And we 're really moving from And we're really moving from st ength to strength. So please not us a mile, And you give us some interesting And you give us some interesting , exciting, and thought-provokim o consider.

consider.

Uh, thank you very much Uh, thank you very much, and had Thank you!

Thank you!

Yes.

Goodbye .

Bye