Sofia WUDC 2026 Open Final

round of Sophia Worlds 2026 the open final a little bit of applause please [applause and cheering] I'm going to start first by welcoming

the teams on stage please join me uh in rockus applause for each team first in opening government USAUA [music] [music] [music]

[music] Second opening opposition Stanford A, please come on stage.

[music] Third, closing government. Sophia A.

[cheering] [music] [music and singing] [music] [music] Yeah.

[music] Heat.

[music] Oh [singing] [music] [music] [music] heat, heat. [music]

heat, heat. [music] Oh, [music] [music] [music] [singing] [music] [cheering] heat heat.

Fourthly, and lastly, but certainly not least, closing opposition, Cambridge A.

[music] [music] [music] [singing] [music] [singing] [music] Congratulations to all teams for

reaching this high stage. Now I also want to congratulate first the panel of the open final. So I'll announce their

names in turn. Bailey Gaspby

[applause and cheering] Cat Chang, Branov Kagalcar [cheering and applause] Tequan, Tenu Omar [cheering]

Ken Puljic, Shi Malachi, and XKU. [cheering]

and XKU. [cheering] Finally [applause] I would like to hand off the mic to the chair of the open final. Please give me

a warm congratulations for Milanovich.

[cheering] Go ahead.

Thank you.

Thank you, David.

Oh, wait. Sorry.

One last thing.

Oh, motion.

That's right. The motion.

The open grad finals motion has an info slide. Human influence refers to the

slide. Human influence refers to the ability of AI systems to exert influence over humans by mimicking human features and behaviors. Examples of this include

and behaviors. Examples of this include strategic humanlike bargaining, persuasion, and emotional engagement.

The motion for the grand final reads, "This house opposes the development of AI systems to optimize for human influence."

influence." Go ahead, Milos. [applause]

Cool. Thank you, David. So, uh, first of all, welcome to the grand final. It's an

exciting times. Uh but because it's exciting times, people can get maybe too excited. Please, I know it's a large

excited. Please, I know it's a large audience. You might like what people are

audience. You might like what people are saying, but please keep the clapping and chatter to a minimum while people are speaking because we will lose what they're saying, and that's not good for the team that you're clapping for. Uh

and that's it. Yeah, have fun. And I'm

happy to recognize the honorable prime minister [cheering] [applause] [laughter] Thanks, man. Appreciate it.

Thanks, man. Appreciate it.

It's going to be really embarrassing when they take it down for you guys.

For tens of millennia, the human race has based itself on fundamental equity.

The fundamental ability of every woman and man to look their fellow woman and man in the eye and know that at the core we are the same. That at the core we are

both human and no one and nothing else is. The capacity of AI to be not only

is. The capacity of AI to be not only smarter than any person alive, but also more persuasive than any person alive, also more human than any person alive,

will, if we allow it, forever end that equity and forever jeopardize our society. First, I will explain what the

society. First, I will explain what the counterfactual is. Then, I will explain

counterfactual is. Then, I will explain why the power of AI to manipulate is only a power of ill. Then I will explain that any upsides you hear from the other teams in this debate are far less

important than the benefits of the counterfactual. First on the

counterfactual. First on the counterfactual, this is an opposed debate. So the development of systems

debate. So the development of systems optimizing for this capacity just does not exist. There are many other

not exist. There are many other capabilities of AI and we will optimize for those instead. Things like its logical or computational abilities, its ability to do things like protein fold

or amass and gendrify and take in a lot of data. It is a we will optimize for

of data. It is a we will optimize for things like the efficiency of AI, whether it's energy or otherwise. We

will optimize for things like the safety of AI, our ability to trust it when we put it in positions of power and influence. First, this will unlike any

influence. First, this will unlike any other AI in a way that we have never experienced undermine the ability of us to understand each other. Why does AI

have such an extraordinary capacity to influence and manipulate people? First

is that this type of power is the precise one described in the info slide.

The infoslide precisely differentiates this AI from all other AIs based on its ability to influence people. And when it says influence people, you should read manipulate people. But second, because

manipulate people. But second, because AI will be far better at influencing and manipulating us than we have ever been.

First, because AI's capacities are functionally infinite for any given task. An AI living in a data center. An

task. An AI living in a data center. An

AI powered by more energy than we have ever really generated. Powered in in the future plausibly by its own mini nuclear reactor plants with more chips than we've ever dedicated to the entire

internet will have the capability to take in vast quantities of data to take in data like the swipe twitches in the user's eye as they look at information.

And that allows it to aggregate whether or not that user is feeling happy or whether that user is feeling sad. its

ability to process every movie, every TV show, every persuasive population, every video of a debater online being persuasive and work out what precisely was working there. But second, because

unlike anyone else, AI can target the most persuasive politicians and the most persuasive debaters are limited because they are trying to persuade many different people and they have only one

personality. The human bias against

personality. The human bias against hypocrisy will fall them down, will call them out if they change too much between each other. But AIS can be infinitely

each other. But AIS can be infinitely targeted. They can have every capacity

targeted. They can have every capacity that we have when we're targeting to the person we most love, person we most know, the person we've most molded our personality to suit. And they can do that to everyone on the first encounter.

And they can, if they need to, or if a malign actor is directing them, can they gain even more information? They can

look at all of the photos and videos and data we have put out on the internet and work out what that precise person wants.

What are the harms of this? The first is that people have the power to use AI to do things that we just do not think they should do. Those are things like

should do. Those are things like governments generating propaganda, governments generating disinformation, using the likenesses of people that you do not know whether or not they are

real. That looks like the wave of

real. That looks like the wave of Russian operations in central Africa propping up regimes like the C based on information that is not real. Based on

people saying and doing things that never happened, based on deep fakes that it could not manipulate in this way on their side of the debate. Second, and

perhaps more trivially, it is advertising. It is the ability to tailor

advertising. It is the ability to tailor the ads that we receive, the demands to buy goods, to buy services, to buy products, to waste our lives and monies on things that do not bring us joy, that do not bring us happiness based on

incredibly precise ways of knowing what will help us the most. And third, it is the power of AI companies themselves.

Already vast swaths of the weird parts of the internet, spending their time talking to AI boyfriends and AI girlfriends getting very, very upset when Chat GPT attempts to modify the systems of those AI boyfriends and

girlfriends cuz they feel like they're losing someone. Imagine what AI

losing someone. Imagine what AI companies can do when every interaction with an AI feels like an interaction with a human. When any attempts by a government to regulate an AI is an attempt by a government to remove a

human. And I would note most of the time

human. And I would note most of the time the people controlling AI will be malign actors in our society because they are the ones who've amassed the power, the wealth, the status, the opportunity to

control powerful technology. They will

be malign. Yes.

The issue with this argument is that humans can convince us and manipulate us as well. Isn't it better to have

as well. Isn't it better to have efficient convincing of propaganda as opposed to an education system that wastes tens of years convincing its population of it? Certainly not, because

the educational system would be much worse at it. Like come on, so many people go through educational systems, come out radicalized against the regime.

But secondly, because the amount of investment you'd have to put into radicalizing someone over 20 years means many governments can't do it because they don't have the capacity or they just don't do it because it's too expensive. But on our side, super cheap,

expensive. But on our side, super cheap, super easy. Why not radicalize your

super easy. Why not radicalize your population? Finally, on the harms, AI

population? Finally, on the harms, AI itself can have malign incentives and that's very likely firstly because it's trained on data sets and those data sets themselves are bad. Think how evil Grock is because it's been trained on all the

people on Twitter who are fundamentally pretty evil. Those people and data sets

pretty evil. Those people and data sets will now form the basis of AI in the future. They will allow it to have

future. They will allow it to have malign incentives. But second, because

malign incentives. But second, because AIs miss implicit context, they miss all the hidden understandings that we have when we talk to each other. Which means

in many circumstances, AIs will have incentives that diverge from humanities.

And on their side, AIs are able to defraud people. They're able to convince

defraud people. They're able to convince people to put money where they should not. They're able to take money if they

not. They're able to take money if they think it is useful for their agenda. If

they're given insufficiently specific instructions like make more money for this company, they will break the laws in ways that people will never know. And

I think there are many tail risks that are just too high to avoid. What if

there is a particularly weak and susceptible president who can just be convinced to give AI control of the nuclear codes? There is just no way of

nuclear codes? There is just no way of ruling out those situations unless we rule out the technology. Finally, the

upsides of AI are just not that big compared to the benefits of the other things we can do. First, because it is good to give humans a place in the economy, it is good that AI has

something that humans cannot do or the humans have something that AI cannot do rather because that what is what allows us to differentiate ourselves and our identity from AI. is what allows us to derive meaning from the fact that this

super intelligent being can't be human the way we are. It is what means that there is still a place in the economy, still a place in the running of systems into the future for people because AIs can never interface directly with

people. They can never be trusted by

people. They can never be trusted by people. But also, our system is one that

people. But also, our system is one that is likely to be more effective. It's one

that targets the things that humans can't do, like really complex maths, like the folding of proteins. And it is one that is likely regulated better when the AI companies do not have access to tools that in every sense are tools of

control, are tools of manipulation, are tools of destruction, are tools of violence, are tools that we just should not allow into our society. So proud to govern.

[cheering] [applause] I'd like to thank that speaker for that speech. And coming from the other side

speech. And coming from the other side of the ring, I'd like to give you a warm welcome to leader of the opposition Hey.

Yeah.

Okay.

[cough] Yeah.

A brief thank you to SDS for coming to watch all of our out rounds. Uh thank

you to the North American Debate Circuit, both the US Circuit and the Canadian Circuit where I was born and raised. [cheering]

raised. [cheering] Thank you for tis for being incredibly supportive this whole way. And of

course, thank you to Amitai, my partner, for being an incredibly emotionally intelligent human being who I love very dearly and I'm incredibly happy I get to be a world finalist with and hopefully more.

[applause] At the scale of millions, if not more, there are people suffering in the depths of loneliness. The feeling that there is

of loneliness. The feeling that there is no one else out there that can feel like they can understand who they want, who they are, and the problems they're experiencing. These are people who in

experiencing. These are people who in their loneliness and their inability to find people to reach out to, self-sabotage, who when because they feel hopeless, they become bitter and feel unable to interact with other people, who lash out because they're

used to being rejected time and time again because they feel like they have no one else to turn to. This harms their future relationships and it makes them feel suffering in the moment itself.

This is incredibly important. I think

the first argument we have in this debate is as to why this is a meaningful cure to human loneliness that is only growing exponentially. Second, I'll talk

growing exponentially. Second, I'll talk about economic integration and why this is incredibly important for this debate.

Firstly, on loneliness, actually, I want to briefly frame at the top of this speech. A lot of what OG says is deeply

speech. A lot of what OG says is deeply uncomparitative. They say this is

uncomparitative. They say this is important for things like propaganda or the creation of deep fakes. But notably,

a lot of what they say is not actually contingent on AI becoming more emotionally intelligent. Governments can

emotionally intelligent. Governments can weaponize developing AI to create deep fakes anyways. And notably, AI will

fakes anyways. And notably, AI will continue to develop on their side of the house because of the profit incentives of governments. They'll become more

of governments. They'll become more effective at doing things like processing information and create creating and producing goods and services and necessary goods to do things like for example manipulate the public. The difference in this debate

public. The difference in this debate then is that when you make the emotional intelligence of AI a targeted goal, it is not only good for emot for a feeling of emotional connection but also good to hold AI accountable whereas it otherwise

would not be. Firstly, on loneliness and also the accountability of AI. I want to begin by framing why there is such a severe problem in society as to why there's massive loneliness around the world where the rise of the internet has caused people to feel more and more

isolated and more and more alienated.

It's the fact that a it gives people an easy way out where many people find social interaction incredibly hard. You

have to leave your house. You have to take a risk of being vulnerable, expressing yourself to someone else else what at the risk of them not taking you seriously and dismissing your emotions.

And this means that when you have the option to do things like engage in shallow interaction to to just do things like comment on social media and to farm likes and comments as a form of having a hollow shell of interaction you will

choose to do so to doom scroll into rotten bed things that are symmetric on either side of the house whether or not AI exists. But third secondly it's the

AI exists. But third secondly it's the exporting of standards. The fact that because of inter the internet you can see images of successful relationships and successful friendships which makes you feel like all of your things are

inadequate. It raises the stakes of what

inadequate. It raises the stakes of what you think a successful social life looks like which makes it too to it makes it too intimidating to try and pursue in the first place. So many people do not even attempt to do so. Thirdly, it's the loss of third spaces. Whereas things

like you know capitalism happens, industrialization happens, people lose cafes at the expense of more and more businesses and the inability to have third spaces to interact with around the world because it's seen as a waste of economic space that does not turn up

profit makes it harder to emotionally connect with people as well. Notably in

the absence of AI because they might say well AI makes this problem worse. In the

absence of AI, this still cannot be fixed by any alternatives. It's things

like the fact that there exists a collective action problem, the status quo. The fact that no one else is

quo. The fact that no one else is willing to engage with you because they feel like they want to take the easy way out of just doing things like doom scrolling or only talking to their internet friends and do not actually make them meaning give them meaningful amounts of joy because these are shallow

comments on social media means that it's very hard for you specifically to try and make friends if everybody else is taking the easy way out. So, you take the easy way out as well. But

furthermore, I also think there's a natural gravitational pull towards certain people, which means others are left are left out of the bubble where if it is becoming increasingly hard to make friendships, then the people who are the most extroverted, who already have

friends, are the ones that can can continue to accumulate more and more friends because they're the most known and already have enough of a social support network to seem more confident, which means they continue to gather friendships, which means others are left

behind. I want to now explain why AI is

behind. I want to now explain why AI is capable of actually solving this. But

before the closing, don't forget also to respond to opening government's point on misinformation by saying people are more likely to become distrustful of information they see when AI is happening and you do it.

So if people become distrustful, it's a great reason as to why people will not make the AI in this way. But

furthermore, I don't care if this is misinformation or disinformation on the impact of relationships. If you feel like you're being heard, if you feel like you're being listened to, and you feel better at the end of it, that is still an impact at the end of this debate. Whereas otherwise people are

debate. Whereas otherwise people are left hopeless. Why is the rise of AI

left hopeless. Why is the rise of AI necessary for this? So I want to begin by explaining why the future development is uniquely necessary where in the status quo I think it is incredibly easy for AI to do to do things like for example not be able to provide a nuanced

emotional support. the fact that they do

emotional support. the fact that they do not understand the intricacies of human emotions which OG concedes. But

furthermore, I want to flip I want to flip the OG argument about why it's very important to do things like a line why it's AI manipulation is bad where often times there's many things that are problems anyways where to some extent there are some people in society where

if we look at chat JPT the status quo and already feel to some extent able to be manipulated by them but the difference is is to make the alignment of AI or to make the emotional manipulation of AI an active thing to

dissect and an active thing to target.

What does this look like? In the status quo, AI already has some capacity to do things like have emotional uh have try to have emotional manipulation but for other ends. So for example, economic

other ends. So for example, economic ends or other things. The difference

then is we're not currently optimizing for the direct emotional impacts on people in the status quo. So the

difference is we need to deliberately train it to say to see if it is manipulating people for good or bad ends. We're using good or bad methods to

ends. We're using good or bad methods to to emotionally to emotionally try and influence individuals. So the difference

influence individuals. So the difference in our side is a we do more direct prompting for the aim or the active metric of aligning this AI on emotional means as opposed to having AI itself choose how it wants to do things like

persuade people in the status quo. It

looks like for example having having direct uh alignment to do things like have put humans at the training to ask them questions or to try and try and train AI on the basis of emotional things. So for example you do it more

things. So for example you do it more you do more uh learning with human feedback. So for example, when you train

feedback. So for example, when you train AI by giving them emotional prompts, not just prompts about science or math, which means you have more ability to catch AI being misaligned at every step of the process. When you do not make this an active thing, you prompt the AI

to do so. It is far harder on their side of the house to have even the capacity to see if AI is aligned or not because AI will always try and take some methods to try to try and get to try and get their end outcome. Notably, AI is also likely to develop in good ways because

of the accountability and trust mechanism that they themselves say. If

things like people being told to do bad things to themselves means that people stop trusting AI, AI which is very very leveraged which takes on massive amounts of debts to develop will take a sudden shock to things like their stocks or sudden shock to things like investment

which is a massive risk to their company but also this gives you the active ability to correct it over time compared to a world where you never explicitly try and train it on these metrics.

Second argument, why is this necessary for economic integration to revolutionize the economy? And notably,

I think this is a prerequisite to to the AI being integrated in many things or many jobs require human emotional intelligence like social workers or interacting with customers to some extent. They might say replacing humans

extent. They might say replacing humans in many of these ways is very bad. But I

think this is a fallacy with the economy will always try to use all of its spare resources to the fullest of its capacity because that is the most rational thing to do to make more money. So when firms can use AI that is both intellectually very efficient but now emotionally very

good as well then firms can do things like save money and that is money they can use to invest in other things to invest in other ser sectors that re still rely on human physical labor like construction where things like you know chemical engineering we still need

people to put the chemicals together and do physical things this means you get far more jobs in the economy but it means you get far more innovation as well when AI can expand to more sectors to get cures to things like Alzheimer's and people trust AI with things like

medical data they can now use to create massive amounts of innovation to save lives lives around the world. For all of these reasons, we're so proud to oppose.

[applause and cheering] Thank you very much for that fine speech. Now, to conclude the opening

speech. Now, to conclude the opening government case, I'd like to invite the deputy prime minister.

Someone leave their phone up here.

The power of artificial intelligence lies in complementing humanity, not substituting for it. Our emotions, our ability to connect with other people, that is what makes us unique. That is

not what we need. artificial

intelligence to do. This team would allow you to erase the one thing that lets humanity stand apart, makes humanity useful and means that humanity has a place in a post artificial

intelligence world. We would strongly

intelligence world. We would strongly strongly regret that. I want to start with a couple of observations about this team. The first thing is I would note

team. The first thing is I would note that this motion is about all forms of manipulating humans. Their claims about

manipulating humans. Their claims about economic incent would also be things that we would oppose the development of. It's not just emotional mechanisms. The info slide and motion are quite broad. The second thing

is I'd like to note the hopelessness of this team. All of their arguments start

this team. All of their arguments start with a premise of it's very hard to achieve this thing. So we should just give up and do something that is clearly bad. Like it is hard to meet someone or

bad. Like it is hard to meet someone or it's hard to find friends. So you should give up on the idea of human connection of real relationships and meaning and just instead engage with a poor simulacra in the form of an AI chatbot.

It is difficult to ensure that people engage with critical thinking are immune to propaganda. It is hard to resist

to propaganda. It is hard to resist those things. So, we should just allow

those things. So, we should just allow the most efficient and penicious forms of propaganda because it would save us all money. No, it is good if those

all money. No, it is good if those things are harder. It is good if those things take more time. And in the vast majority of states where governments have good incentives, they will continue to use the education system to teach critical thinking, not propaganda, the

substitute they call for is instead just allowing governments to efficiently propagandize their populations because at least it would save that government time. No, if that government is trying

time. No, if that government is trying to do something bad, I hope it takes them as much time as possible. I hope it costs them as many resources as possible. I hope there is the greatest

possible. I hope there is the greatest possible chance that people are able to resist this. I want to first make a new

resist this. I want to first make a new argument about trust uh you know to I guess for whatever weird knife closing government has in plan for the second debate in a row. Firstly,

there are two reasons why this capacity will destroy trust in society. Firstly,

when AI is capable of simulating the things that make us human better than humans themselves can, you cannot know who is AI and you cannot know who is human. That fundamentally means that you

human. That fundamentally means that you lose the trust that the person you're engaging in is really there, is really genuine, is someone that can be engaged with, is not a data center a million km

away. But secondly, it means you cannot

away. But secondly, it means you cannot trust any information you receive anymore. Imagine a speaker so persuasive

anymore. Imagine a speaker so persuasive it could convince you of any possible argument. So like slightly more

argument. So like slightly more persuasive than I am. Uh if you heard an argument from that speaker, you should just never be persuaded because you know it could equally convince you of the opposite of the truth just as easily as

it could convince you of the truth. So

even if the argument makes perfect sense and you believe internally it is so true. You should know that the chance it

true. You should know that the chance it is true is very low. That it could equally be false. Why should you be persuaded by the argument? doesn't need

to be connected to the truth in the way that we currently believe that logic and reason must connect to the truth and must therefore be persuasive. This is a world in which you could tell people all the best reasons for why vaccines are

good and they just would never ever trust you. Nothing you could say could

trust you. Nothing you could say could ever make you trust them because they would say, "Well, you could be literally lying. You could be explaining to me

lying. You could be explaining to me that jumping off a cliff is good and you'd be equally persuasive in doing so." It is a world in which we confine

so." It is a world in which we confine ourselves to our own worlds cuz we never trust the people around us to be telling the truth. Why is that unique only to

the truth. Why is that unique only to emotion? It is because facts themselves

emotion? It is because facts themselves just cannot be that persuasive. People

often just don't want to listen to facts. People are often suspicious of

facts. People are often suspicious of the people who are giving them the facts. And you can check the provenence

facts. And you can check the provenence of those facts. You can follow the chain of logic yourself and see if they make sense. You don't need to be a genius to

sense. You don't need to be a genius to follow the facts one by one and see if they do in fact align or to check sources. But emotion sweeps you up. The

sources. But emotion sweeps you up. The

rhetoric allows you to believe in the moment. The rhetoric makes you feel that

moment. The rhetoric makes you feel that it must be true in your heart in a way that the facts never could. which is why this harm is unique. There are three reasons this is the most important thing in the debate. Firstly, it is a world

without connection. Because you never

without connection. Because you never believe that anyone is genuine. You

never believe that anyone is telling you the truth. You never engage in the

the truth. You never engage in the vulnerability opening opposition says is important because why would you when everyone else everything around you is just fake? Secondly, society cannot

just fake? Secondly, society cannot exist with trust. How could the banking system exist if we do not trust that our money will be there tomorrow? How could

governments exist if we don't believe that there are people on the other side of huge countries that are just like us, that share our interests, that share our goals, and that we should give them our confidence, our money, our support, that we should work for their betterment

because they are similar to us. How is

it possible that institutions could be built when you just don't know if the people around you are trustworthy? I say

that that means that this is a world with far far less trust and that is a far poorer world to live in. Secondly,

on regulation, I would note that AI would use this tool to resist regulation, and it would be by far the strongest way to resist regulation. If

you could deeply persuasively argue as the AI yourself that you should might not be regulated and make people feel that in their hearts, it would be far far harder to regulate than any amount of lobbying that open AI or any company

might engage in. And here they say, well, propaganda is good on both sides.

Technology will improve. But there's a difference between our valuable human minds coming up with the idea and using AI to quickly produce it versus AI itself understanding the psychology of each individual human, what would speak

to them, and being able to manipulate them, appearing as the one person they trust most, appearing as the person they respect, explaining why they must engage in this particular policy. But secondly,

by banning this form of AI development, there are fewer types of AI development to regulate, which just makes it simpler and means we can focus on those other things. Finally, I would want to note

things. Finally, I would want to note that our argument applies regardless of how powerful you believe AI will be in the future. If you think super

the future. If you think super intelligence is likely, then our arguments about AI using this to manipulate people into building bioweapons or the president in handing over nuclear weapons is super pertinent and wins us a debate. But maybe you

don't believe that and that is entirely reasonable. Even if it only becomes

reasonable. Even if it only becomes marginally more powerful than today, it is enough to displace people from the economy because it is far cheaper and far better able to do those things. It

is far more than enough to make humans irrelevant. Yes. So, is Jack right that

irrelevant. Yes. So, is Jack right that AI is a godlike persuasion robot? Or are

you right that we don't trust anything because nothing is genuine? Who are the propagandists now? Uh, well, I think

propagandists now? Uh, well, I think there's a distinction which is the point of this debate, which is that facts themselves can be somewhat persuasive, but it is the emotional connection we feel that deeply changes the way that we

feel speeches. I I think you know there

feel speeches. I I think you know there are many people who are great with facts, but you do not feel persuaded by them because the facts are dry. You

don't want to engage with them in the first place. You don't feel the truth of

first place. You don't feel the truth of those statements. That is precisely the

those statements. That is precisely the difference we point out. There is no tension there. Finally, let's respond to

tension there. Finally, let's respond to the remainder of opening uh opposition's arguments. They say this would solve the

arguments. They say this would solve the problem of loneliness. One, I think there are other ways to solve the problem of loneliness. You could come meet people like at this great tournament. Secondly, the AI would

tournament. Secondly, the AI would likely be abusive. It has the capacity to manipulate you. The incentive of its owners to maximally milk you for profit.

It will make you dependent on it and then it will drain your bank account or engage in any number of duplicitous practices. But secondly, it's not a real

practices. But secondly, it's not a real relationship. It doesn't challenge you.

relationship. It doesn't challenge you.

There's no friction. The reason there'd be no friction is that's why you don't engage in a real relationship. So the

way you sell this is that it'll always agree with you. It will always do what you want. Why should you care about

you want. Why should you care about that? It's just wireheading. You're just

that? It's just wireheading. You're just

taking heroin. It doesn't challenge you.

It doesn't choose to be there. The

aspect, it doesn't really matter in real life. That is not real. What makes

life. That is not real. What makes

relationships meaningful is the fact that a real person chose to engage with you, to challenge you, to make you a better person, not just make you feel good in the moment. Incredibly proud to govern.

[applause] I'd like to thank the speaker for that governing speech. And now to close off

governing speech. And now to close off the top half of this debate, I'd like to invite Deputy Leader of the Opposition [cheering] How

did this I'll also do the thanks briefly. Uh,

thank you to Victor, my basic boyfriend, and Liz, my best friend. Uh, thank you to Hadar, Asher, uh, Ening, and Ron for coaching me these years. I would have never been here without you. And thank

you to both SDS and the Israeli Debate League for, uh, the endless support and majors and outside of them. I'll start

my speech in three, two, one. The beauty

of our argument is that the DPM understands it completely. We want to offer heroin to people. We think that it is great that AI is there to tell you everything you need when you need to

hear it. We think it's great to be

hear it. We think it's great to be manipulated. The first thing I want to

manipulated. The first thing I want to do in this speech is explain what isn't exclusive to this motion because OG fails into the trap of how many things

are not true just about this specific kind of AI. I want to be very clear. The

motion is about the function of convincing people. the function of

convincing people. the function of emotionally reaching out to them. But if

it is an operative part of your end goal to convince people in order to achieve a different end goal, that is not our side of the motion. In other words, if you

need to convince people to change their jobs, if you want to convince them to adopt your model, if you want to achieve efficiency, as OG say in an economic sense, that is still part of the model

that they need to defend through other AI functions. Why is this important?

AI functions. Why is this important?

First, there are other ways of convincing people. Many times even worse

convincing people. Many times even worse ways when you don't understand them. AI,

especially the deductic AI that OG talks about, will be able to make game theoretical arguments about the reason that you should accept it and that you should buy its arguments to accept on a companywide level. Game theoretic

companywide level. Game theoretic arguments about the fear that your company should face when other companies are accepting it and what how you might lag behind. Second, it might use data

lag behind. Second, it might use data generation skills that don't have to do with understanding humans, but just understanding the data that supports its argument. And that means that the

argument. And that means that the support for deep fake generation and for other sort of false information that OG is so worried about and of that mistrust that UD talks about anyway will be true

because of the generation of these false uh operatives. They say, but you need to

uh operatives. They say, but you need to understand how humans think and it increases the scale at which we understand it a lot. That's true. But

panel arguments about scale require you to understand why the scale changes the harm that they talk about. My claim is that you do not need an increase in the scale of understanding humans in order

to delude and manipulate them because humans still understand that falsified information that tells them there are massive terror attacks or fears around them should cause them to feel fear or

react in a certain way. Which is why even without understanding deeply how humans think, we use these other tools in order to manipulate them to achieve our end goals in the way that the opening government says. My second point

of the second burden I want to point out as missing in OG is that you do not need to accept the premise that AI is so exclusive in convincing us. They say

well clearly it is better at understanding our wishes and goals than humans are or than governments are at propagandizing us. The issue with this

propagandizing us. The issue with this is it is true that governments may not understand how each of their citizens thinks. That is why the way that they

thinks. That is why the way that they treat this is by trying to take away things you can be exposed to as opposed to trying to convince you when they do not know how best to convince you. That

means that China reacts to wish through censorship through the great firewall and through propagandizing in school systems. We think that that is a much worse alternative than if you could

ensure that through AI you get a constant exposure to a narrative that tells you you should trust the CCP because that means that the CCP doesn't need to bar access to any one of the

things that could prevent you from trusting it. That means that it's much

trusting it. That means that it's much easier for governments to not waste as much resources as OG say, which they will anyway want to do to maintain their survival instinct on limiting and the

scope of things that you could access in order to become more critical of them.

It is much better that censorship is effective than when censorship has to be wideband in order to delude you. The

next thing that I want to talk about is what does this do in terms of loneliness and your personal effect. U makes a few responses here. The first is to say,

responses here. The first is to say, well, why do you need to give up? Isn't

it better that you address the challenge by going out and seeking help? The first

thing that I want to point out is that this does not address Liz's explanation that even if you are able to find a relationship, even if it is not about the formation of a relationship or of meeting someone, the problem is that

there are many things we will never say because of our fear of how limiting they are to our ability to form a relationship. our deepest doubts about

relationship. our deepest doubts about the fact that in a few years we don't really think we're going to live here.

The fact that we don't have long-term plans, our thoughts that we never want to share with other people because of the effect that they will have on other people wanting to spend time with us.

And I want to weigh this argument. The

ability to share your deepest feelings and be told by the AI they're okay or receive a more reasonable explanation for why your mother passing away is

okay. For why your pain is something

okay. For why your pain is something that will pass in the long run than an unconvincing explanation now is number one more helpful to your formation of relationship. Because for whoever

relationship. Because for whoever carries a more deep emotional weight, it allows them to alleviate outside of their relationship such that their ability to form relationships grows because they can show their good parts

at the beginning of the relationship, gain an attraction and and a feeling of affiliation with them and offset the parts of them that are most painful to an AI system. So the relationships we

get with real humans grow. They say

maybe the desire to have these real relationships drops. But the issue is

relationships drops. But the issue is that as Liz explains the ideal in society right now when we look at UD and Martha and we say what a great couple

these are two people. We always want to reach that ideal and AI will always be a fall back to what your parents your society and the movies around you have told you is the best thing to achieve.

So the desire to strive for it will always exist. AI just makes it easier.

always exist. AI just makes it easier.

Bottom half.

Yeah. Even if all of this is true, I think AI is the biggest blight upon our world. It destroys our planet. It

world. It destroys our planet. It

destroys our brain. You can think more down. Uh you can explain why you think

down. Uh you can explain why you think that's true in the extension. Uh the

next thing I want to explain as a point of weighing is why AI will usually treat us well and not treat us badly. Why is

this true? AI does not know how we will react to it. It is faced with an unknown and and a unknown probability that at some point we opt out of using it. At

some point we break our suspension of this belief. At some point we realize

this belief. At some point we realize we're talking to an algorithm and not a real person. Why is this really

real person. Why is this really important? It means it wants to give you

important? It means it wants to give you copium to give to keep you hooked as OG say. But notice that the second it stops

say. But notice that the second it stops giving you copium, the second it changes your boyfriend alterations as PM says, this is when you suddenly realize you're talking to a machine. you're not talking

to a human who has empathy or cares for you enough. So, in order to not break

you enough. So, in order to not break the suspension of this belief where you're talking to a real human, the AI never wants to hurt you in ways that the human would not and never want to make you feel that they're subpar to what you

could get in the outside world. So, even

if it is ruled by the rich elites, the rich elites want you to keep using the AI that seems like the best person around you. The final thing I want to

around you. The final thing I want to mention is what they say in terms of alternate functions of AI. Let me flip this. The issue is this is clearly

this. The issue is this is clearly completely inexclusive. LLMs use the

completely inexclusive. LLMs use the same data that they get from our use in order to generate psychological needs.

It's a completely different division to R&D that's based on AI's computation ability and stuff like that. That was

the flip. But yeah, for all those reasons. Oh. Oh,

reasons. Oh. Oh,

[cheering] [applause] thank you very much for that very nice speech. And now we're moving to the

speech. And now we're moving to the closing half of this debate. And from

the left corner we have member of government. [screaming]

government. [screaming] [cheering] [applause] Um, firstly just like less than 30 seconds of words. I had much longer thank yous, but apparently I need to be under 30 seconds. So I'm just going to

say one thank you. Uh, and this is to the original Sophia who walked the path for us and are the only reason as to why we are here. All of us like literally in

Bulgaria. Um, and I think and Nikki

Bulgaria. Um, and I think and Nikki deserve so much praise all around the debate society for everything they've done that they do not get.

[cheering] [applause] I also started something at the Euros finals with the full intention of finishing it here in the world stage.

[screaming] [cheering] [applause] Why do we destroy the desert so that we can ask Chajipi about the weather when we can just look outside? Why do we

pollute the waters so that I can ask character AI a question I could my friends? Why do we destroy the human

friends? Why do we destroy the human brain? So I can skip the effort of

brain? So I can skip the effort of writing a college essay as opposed to learning something new about the world, something new about myself. Closing

government extension here is incredibly simple and incredibly easily debate winning. This massively increases the

winning. This massively increases the use of AI that either way could be done by a person or is completely pointless for little to no benefit but massive

harm. harm upon our planet, harm upon

harm. harm upon our planet, harm upon our brain. Firstly, why does this

our brain. Firstly, why does this massively increase the use of AI that we oppose? Firstly, uh all throughout

oppose? Firstly, uh all throughout history and how we are structured as humans to functions is conversationally.

It is person to persons. That's how we process information, how we relate to each other, how we assign value and credit. But now when AI can replicate

credit. But now when AI can replicate this conversational ability, it is very intuitive for us to turn to it as a force and this has a replicational effect. The more we use AI, the more it

effect. The more we use AI, the more it improves within its conversational ability, the more appealing it seems to us to use. But know that AI in general

becomes easier to use now when it has those human characteristics. It is

friendlier, so it predisposes you to use it. it is more helpful or sounds like it

it. it is more helpful or sounds like it is because it factors persuasion or it becomes just a path of least resistance when we want to use it. So as opposed to

googling, people now ask Chad GPT. As

opposed to reading your book, you ask Chad GPT for its opinion. As opposed to talking to your friend, you talk to character AI. Second reason as to why we

character AI. Second reason as to why we use it more because this is the most that AI and other LLMs do optimize for

persuasion but not results. What do AI companies do disproportionately?

Firstly, they want to optimize for this because it is the easiest to monetize.

The problem with LLM such as Chad GPT and why Chad GPT has not been making a lot of money is because a lot of people use it but we cannot find a way to monetize it. That is why Chad GPT is

monetize it. That is why Chad GPT is developing within persuasion because then people get hooked. They create

relationships within the characters that they talk to and they want to pay them money. But also secondly, the market is

money. But also secondly, the market is obviously big. It is a very big part of

obviously big. It is a very big part of people's lives to be able to lead conversations with others. But it is also the most vulnerable people with the biggest struggles that you're likely to

just monetize upon because they themselves cannot find friends in real life or think that they couldn't. But

also thirdly, LLMs are a bit [ __ ] for everything else. Like I don't understand

everything else. Like I don't understand what this is debate about on top half about LM and them solving everything within Doro. they operate with imperfect

within Doro. they operate with imperfect data or the likely response that a human is going to give to you not necessarily the correct one. So that the only thing

that hides their inability to actually be function is their persuasive ability and this is why AI companies invest with

this. This means that the use of AI

this. This means that the use of AI becomes disproportionately more because they optimize for this. This becomes

much better. people use it much more their normal life much more. This

completely destroys the environment. And

now I understand what everybody's going to say. Oh, but we are symmetrically

to say. Oh, but we are symmetrically going to adopt AI. Three reasons as to why this is asymmetrically much greater.

Firstly, uh why people use it just excessively more because it conversationally this is already what I have proven within the reasons before.

But also secondly, this will only keep growing in the future because of a it convinces you further. The more you use it of its use, it is a path dependency.

The more you want to use it, the more companies integrate it into everything.

Google search now has AI, [laughter] Alexa has AI, cars have AI. Entities

that might not need AI [screaming] for their operations, that might not actually improve their services, we did.

But now as a society we have a set of expectations that AI needs to be everywhere because we expect it to be everywhere. But also thirdly and

everywhere. But also thirdly and importantly and this is like beautiful analysis. This is now half of the AI

analysis. This is now half of the AI [ __ ] prompt that is giving back to you complimenting you saying you hi sending you emojis. This becomes half of

the response of the questions that you have asked it. This is 50% more computing power. This is 50% more energy

computing power. This is 50% more energy usage. This is a debatewinning [ __ ]

usage. This is a debatewinning [ __ ] impact. Why does this is this

impact. Why does this is this devastating for the environment? AI is

the single most energyintensive industry on the planet. Why? A because of training data either it's neural networks or machine learning. It needs

so much pictures of portals of oranges in order to understand what an orange is like 1,700 [ __ ] [laughter] pictures.

This training data requires a lot of alo computing power and know that secondly there is an n square problem. The more

complex AI gets, the more information gets fed within the AI, the more energy use, it's expansion. And note that it's uh like uh disproportionate uh because

for every piece of data it needs to connect with every other piece of data.

So the the line goes exponentially in the future in terms of power use and know that data centers that we're now building all across the world are sucking up all of the water to cool the

generators are dumping toy waste materials all across India. And what I have proven is an symmetry in use that's massive. Sorry, I'm very sorry. Now what

massive. Sorry, I'm very sorry. Now what

I have proven is mass in symmetry in the use. This is massy symmetry in

use. This is massy symmetry in pollution, massy symmetry especially in the future when we are integrating this more and more in the type of climate impact that it has. How many more

typhoons does Vietnam need to suffer?

How many more flooding does Indonesia have to suffer for us to snap ourselves out of this misery where AI is completely needless? Secondly, this

completely needless? Secondly, this completely destroyed the human mind because we use it as opposed to thinking. It is a seductive. The

thinking. It is a seductive. The

conversation is just really easy and it is giving you all of those responses that make you want to continue more. It

convinces you but also has an incentive to keep you in the platform. This is the structural reason as to why keep going to it because it is incredibly reinforcing even in your bad habits. And

it is not about relationship but about everything else. whether you're going to

everything else. whether you're going to now write your essay, whether you're going to drink your water with vitamins or not and you need to ask Charity for it. But know that every single time that

it. But know that every single time that we give up our ability to think, our neuro connections weaken, our [laughter] capacity to train our brain, this muscle that we have over and over again so that

we have the capacity in the future make good decisions for ourselves weaken. For

all of those reasons, CG [cheering] [applause] [applause] Thank you very much for that speech. Now

to open up the closing opposition part of this debate, I'd like to give a warm welcome to member of the opposition.

I'm calling your heart.

A few very brief thank yous. Thank you

to everyone who has made this world such a fantastic experience. It's been my first and extremely enjoyable. everyone

involved in organization, everyone else here. Thank you to everyone at Cambridge

here. Thank you to everyone at Cambridge Debating for your support over the last two years. A huge thank you to my

two years. A huge thank you to my Scottish family debating and otherwise who've kept loyal to me despite my defection down south. Thank you to Max.

It's been an honor to break the quarterfinal curse for you. And thank

you very much to my friends and family back home. Everything I do, I do for you

back home. Everything I do, I do for you and because of you. Thank you.

[applause] AI chat bots are just the tip of the iceberg. They're what we integrate with

iceberg. They're what we integrate with on a day-to-day basis, but they're not where AI will have the most impact in our lives. Where it will have the most

our lives. Where it will have the most impact is within the halls of power. The

types of policy that governments use it to create, where it influences our economic modeling. This debate is

economic modeling. This debate is largely symmetric on output for reasons I'll explain in a second. The difference

is the input that we give it. How we

train AI to care and understand human ethics and emotion that we solve the realignment problem with AI. A few

responses first to opening government on the question of persuasion. Jack says

this is the most persuasive tool that we have ever have that states use it for propaganda. Udai clearly influenced by

propaganda. Udai clearly influenced by his time in Bulgaria says you will trust no one. You will trust not the state

no one. You will trust not the state because everyone is equally persuasive.

These are obviously tense. We agree with Udai. There is already healthy

Udai. There is already healthy skepticism around AI that only increases when you're aware of its persuasive capacity. Additionally, the diminishing

capacity. Additionally, the diminishing marginal returns for propaganda because the tools for propaganda exist on a more democratic basis. So, it's not reliant

democratic basis. So, it's not reliant on controlling medium networks when AI chatbots are much more accessible to other people to opposition politicians to also fight back. Realistically,

people are skeptical. They are limiting and there are ways to fight back against these tools. What then do we think the

these tools. What then do we think the meaningful difference in this debate is?

The thesis of our extension is that we are currently working out how to integrate AI. The AI, despite what

integrate AI. The AI, despite what closing government will tell you, will be a massive part of our future regardless. The question is how we

regardless. The question is how we direct it at this crossroads. How we

train it to understand us versus cruel utility of numbers that opening government would stand over? Why is AI going to be extremely influential in politics, policy, economics on either

side of the house? First, I cannot overstate the economic efficiency of AI to people who run companies. That it can replace your labor force for a fraction of the cost. That it can double or

triple your speeds. There are no limits to how many people want to adopt this and are seeking to do it. And they'll do it regardless of impact because people will lay off poor workers all of the time, and that is because they care

about efficiency above all else. Second,

AI already has access to the halls of power. They don't need a persuasive tool

power. They don't need a persuasive tool because Sam Olman is at Trump's dinner table. The point is politicians are

table. The point is politicians are already in the pocket of AI because they're so wealthy and influential regardless. So it is becoming embedded

regardless. So it is becoming embedded in the coming years. Third, it is already embedded in our dayto-day use in our use in our academic work or companion bots and so on. Even with the

existing skepticism, people rely on it because it's easy. Now, Valina tries to say there is an asymmetry here. One, AI

is still obscenely useful to you even if you don't think it is emotional because it can still write your college essay for you in half an hour or give you the answer to any question that you has. It

doesn't need to say PS. I love you because it's solving a massive problem for me and reducing my workload. Second,

overall this is a minimal minimal aspect of the environmental impact of AI. The

things that guzzle water and data uh guzzle water and energy are the data centers which power any logical or rational AI that's running model after model for governments or whatever.

They're not the chat bots that exist on an overall basis. Finally, AI may cause problems, but it also offers the solution. We are clearly not at a point

solution. We are clearly not at a point where we have sufficient innovation to tackle things like the climate crisis.

Our progress as humans is too slow. But

AI can change the game in medicine, in environmental science, and so on. May it

cause some difficulties for us.

Absolutely. But it holds the solutions in way no human brain has ever proved that it can. The implication of this analysis is that on either side of the house, AI is a massive part of the

future. It influence what governments

future. It influence what governments do. It it influences the economics that

do. It it influences the economics that exist around you and so on. So the

question is how we are going to direct it. What happens then with this process?

it. What happens then with this process?

First, the data that we ch train AI on changes because we change it on human react train it on human reactions and emotions as opposed to the raw data and numbers in logical models the opening

government chooses to talk about. So the

things that we feed it are not just things that it can pick up from the internet but things about how we feel about things, what our ethical codes are and so on. Second, the only way to train

these kinds of things is through mimicking behavior. The reason for this

mimicking behavior. The reason for this is that the way humans understand the world ethically and emotionally is complex. We cannot write down our value

complex. We cannot write down our value system on a bit of paper because it's situation dependent. It's influenced by

situation dependent. It's influenced by complex psychology. The meaning of this

complex psychology. The meaning of this is that you can't program down a written value into an AI. The only way is to train it to think like a human with our irrationalities, our emotional

complexities and so on.

Thirdly and fin thirdly, we reward it when it aligns with us. So we teach it that the way to be successful is to be more human. We agree with opening

more human. We agree with opening government that humans are wonderful and flawed, but the point is that they have ethics and morals. They care about loyalty. They care about liberty. They

loyalty. They care about liberty. They

care about what it means to reduce suffering, not just on reducing a number of deaths, but what it means to live a dignified life. And we train AI that the

dignified life. And we train AI that the best way to offer us efficient outcomes is to align with those kinds of of ways of thinking rather than just utilitarian metrics that it decides on alone.

Opening there isn't attention. There's short and long-term arguments. In the former,

long-term arguments. In the former, people are manipulated and propagandized and defrauded and thus AI grows misaligned and over time people lose faith in AI and thus trust in anyone or

anything that could be it. I always find it funny when people try to clear up attention in a POI because you know you hit on it, right? So if their idea is that this propaganda in the short term may be over a month, I think unrealistic

because AI skepticism is already here, it doesn't need to emerge. People are

already worried about what it might do.

Have a healthy skepticism and so on.

Secondly, it isn't clear why a month of worse propaganda would be massively damaging. Bit too late to sort it out.

damaging. Bit too late to sort it out.

Sorry. What government might say to this is, well, the people that you train it on aren't necessarily average people. we

can train it on evil value systems and so on. The problem is that AI is

so on. The problem is that AI is learning all of the time. So as soon as it exists and rolls out on a global basis or if it's collecting data on behalf of a government across your population, it is amalgamating an

approximation of a vast range of people who might not have one unified value system but all care about things like the people close to them or guaranteeing that they have some liberty and so on.

Fourthly, we make the outcomes, and this is crucial, more legible to people because AI can explain it to us in a way that we understand. Spitting out a series of ones and zeros or spitting out

a complex graph is often difficult for policy makers to interpret on how it might impact your life. When it is able to explain to you, I think people will like this because while it may trade off

with some of their economic interests in the short term, it's likely to protect society in the long term and cohesion and so on. This means we are better at discerning what policies are beneficial

to implement. Why is this the winning

to implement. Why is this the winning extension in the round? Ultimately, AI

is massively impactful under either side of the house. The question is how we align it with our true interests. The

only way to do this is to make it understand humanity. And the only way to

understand humanity. And the only way to understand a human is to have empathy by thinking like one. If you want AI to work for you, make it like you. Closing

opposition.

[music] [applause] Well, thank you very much for that speech. Now, to conclude the government

speech. Now, to conclude the government side of this debate, I'd like to have a warm welcome to the government whip.

[cheering] Okay. Okay. [laughter] Thank you, Shine.

Okay. Okay. [laughter] Thank you, Shine.

Let's get uh I want to thank uh everyone who supported us, Marvis, who are in my opinion better than many of the teams here. Not because the teams are bad,

here. Not because the teams are bad, because Mariam are so great. Uh I want to thank uh a variety of other people.

Uh my family who is here, congrats. Uh

and given my final uh this will be my final speech competitively as Sophia, the immortal words of Dr. Dre, go on my mind. Uh, there will be nothing after

mind. Uh, there will be nothing after that. So, give me one more platinum

that. So, give me one more platinum plaque, then [ __ ] crap. You can have it back. Uh, thank you very much.

back. Uh, thank you very much.

[applause] I am a progressive liberal person who will be a good progressive liberal parent. If I see my child using AI, I'll

parent. If I see my child using AI, I'll beat his [ __ ] ass. There is no greater harm on this world. No greater

evil than this useless thing that does nothing never been prouder to propose.

Firstly, what are the responses to closing opposition? Because closing

closing opposition? Because closing opposition says ah but this will still be useful because it writes your essay in this particular capacity. There are

three responses to this. Firstly, the

reason that the bot that writes your essay exists is because in the first place they developed towards doing it in a way that's like uh that is able to persuade people. Why? Because the market

persuade people. Why? Because the market for that is vastly bigger. The market

for people wanting to interact with others is vastly bigger. Also, secondly,

because it's much easier to imitate human conversation than to it is to imitate math or a complex essay or something like this. because this is the more intuitive thing because

conversation is less precise there's more room for improvement so that's the very first things that ever get developed and then they use the data from that to train things like essay

creation to train things like math creation this is the beginning if we don't have this all other forms of AI are comparatively slowed down but also secondly maybe those kinds of very

useful things will be done but broad adoption will be much lower because note This is a complex new product. It is a product with risks. Does it actually

take give you what you need it to give you? Does it create this? The biggest

you? Does it create this? The biggest

way it sells you is it persuades you to keep using it. It persuades you to do this. And note, it does it for useless

this. And note, it does it for useless [ __ ] for things like, oh, what's the weather outside? Great question.

weather outside? Great question.

[laughter] It's incredibly important that you ask this things like this where people are persuaded into using it for a variety of other things. And notice like on a likelihood basis this is what the

persuasion skill is most likely to be used on because this is what the companies want the most for you to keep using it. So they keep using it towards

using it. So they keep using it towards you using it for this uh kind of thing.

Also thirdly right like know that now the AI has vastly more data because it is used for 10,000 other use cases.

Valina tells you the most common use case of society is persuasion and conversation. It's how our lizard brains

conversation. It's how our lizard brains determine whether there's danger or not by interacting with people. So this is the most where they can get data in order to do these kind of more complex things. Maybe unfortunately we don't

things. Maybe unfortunately we don't remove AI though I would love to but at the very least we massively slow it down in this particular capacity. I want to first flag what are the impacts of this.

We get unambiguously good things on the environment unambiguously good things on people learning as opposed to using it for their essay. But know that we empower many of opening government's uh

uh impacts with a much better mechanism cuz what's their economic impact? You

displace people whose jobs rely on persuasion but they themselves say alternatively with focus on other things. So alternatively you displace

things. So alternatively you displace more people focusing on math focusing on something like this. We tell you all AI becomes more used now. All AI in all

circumstances becomes more used. So you

displace broadly people more whether they do persuasion or not the bot becomes better more ubiquitous more complex also secondly their impact on

dodges regulation is the AI will convince the lawmaker to let it off the hook this is their actual opening government mechanism like why would people believe this they have natural skepticism our mechanism is it's just

slower it can't convince people the same way to adopt it cannot it isn't funded to the same level that it would be otherwise So we're able to empower these impacts massively. The one impact that

impacts massively. The one impact that opening government I think focus a lot on is the misinformation argument. I

have four ways against the misinformation argument. The first is as

misinformation argument. The first is as pointed out by closing opposition people have a natural skepticism towards things they see on the internet towards things they see coming from other people at a

point where you don't actually interact with this person. What do we tell you?

There are things people want to be convinced of that they wouldn't be skeptical of because they want to think that the question they asked was such a great question. You're such a good guy

great question. You're such a good guy in this kind of sense. So that's when the skepticism doesn't kick in because it validates your own personal feelings which is something you desire rather than convincing you of something that

you aren't likely to believe at the point where you don't buy external information coming from someone.

Secondly though um manipulation can be good or bad. There are many dumb people across the world. Many of them will be convinced towards good things at a point where AI doesn't necessarily need to be used by bad actors. It can be used by

good actors. In many cases, it's good to

good actors. In many cases, it's good to believe in your government to have comparatively less skeptism in this particular capacity. Also, thirdly, like

particular capacity. Also, thirdly, like okay, people will now buy like products and things more in this particular capacity. So what what would they do

capacity. So what what would they do about it? People have a limited effect

about it? People have a limited effect upon their lives in this particular case. But we tell you people have a

case. But we tell you people have a massive effect on the environment at the point where they constantly use a tool which burns it. also a massive effect on their own lives when they don't learn [laughter] because they just keep using

this system as a substitute in this particular capacity. also fourth wing

particular capacity. also fourth wing like okay people can be manipulated and bad things can happen but this can be reversed because you see in the real world that Russia isn't good that there are bad effects of things occurring and

you do something bad for yourself and it doesn't feel nice so you stop doing it but pano how will you reverse the massive amounts of energy used towards destroying your environment how will you reverse the massive amounts of water

used towards this particular thing this is an argument which once done has no ability to be undone yes convincing thing may be harder, but as your OG says, it is still doable through deep

fakes. Your burden is to explain why

fakes. Your burden is to explain why it's better that data centers generate those than generating emotionally appealing prompts. Because emotionally

appealing prompts. Because emotionally appealing prompts are longer because not everyone requests a photograph, but most people use AI for text. So emotional

convincing on a text level is the vastly more popular one. It's used

comparatively more. There's more of this kind of terrible uh environmental effect on this. Then on loneliness, I just want

on this. Then on loneliness, I just want to say this is a highly ambiguous argument at a point where maybe it's good for some people, maybe it's really terrible for some people. If they can't find friends, it seems like it would be a good thing. But also, most people know

they're talking to a robot. So the

impact of this argument is lower on either side cuz they won't believe that this person is actually their girlfriend, but they will believe this can be used to ask them about the weather. Stuff like this. Finally, on we

weather. Stuff like this. Finally, on we should be feeding the AI human data because it can be great like this.

Firstly, we can do this without training it to persuade people. We can just teach it. How do we react to certain things?

it. How do we react to certain things?

What makes us happy? What doesn't? How

do we interview people? We can study the output without necessarily using it to this kind of thing. But secondly, the data is really terrible cuz there's a selection effect. Who uses AI

selection effect. Who uses AI disproportionately? Lazy people like

disproportionately? Lazy people like lonely people, weirdos in this kind of sense. So it's unrepresentative. Also

sense. So it's unrepresentative. Also

thirdly, why is AI good? Why can it create something new? It just analyzes existing data. It just takes what we've

existing data. It just takes what we've already done. Just because you can

already done. Just because you can manipulate 70 million bottles of oranges to output an orange doesn't mean it can do what I can. Vote us.

[cheering] [applause] [applause] I'd like to thank that speaker for that great speech. And now to conclude this

great speech. And now to conclude this debate as a whole and this championship, I'd like to give a warm welcome to the opposition web.

Uh just a few things. Uh first, thank you to Sophie. Uh as you all know, this is uh one of many tries. It's the one that worked. Thank you. Uh thank you to

that worked. Thank you. Uh thank you to my family, including my grandparents who are watching from home. Thank you to the Miguel debating union and the entire Canadian circuit. I miss you. And as you

Canadian circuit. I miss you. And as you can all see from the semi-finals, Miguel is so back. Uh, thank you to Cambridge Debating and to the Iona Circuit who have welcomed me. And finally, thank you to every previous debating partner, to

DBA, to Naomi, uh, to Reena, and all of WSDC WSDC team Canada. It's been a privilege to work with you and build the skills that brought me here.

[applause] Closing government has two massively exaggerated impacts. The first is that the use of AI

impacts. The first is that the use of AI uses much more water and energy. I have

three responses. Number one, the single biggest buildout of green and clean energy in the world right now is coming from AI data centers. We're seeing

nuclear power being put back on. We're

seeing hydro power being built. We're

seeing solar panels being built next to AI data centers. Why? because the

marginal cost of green energy is lower than fossil fuels. It makes sense to privately build these things off-grid and put them next to data centers.

Second response in terms of water, there's obviously an incentive to upcycle water and we're already seeing, for example, many states which have lots of water use from AI tell them no more water, find your own, import it, etc.

Third, the water doesn't disappear and needs to be filtered. They tax AI companies to refilter the water. Last

response, their claim only matters if they create more emissions from fossil fuels. Then those emissions have a more

fuels. Then those emissions have a more than tiny impact on the overall progression of climate change. The

terminalization of the impact from closing government on marginally more emissions needs to prove it was fossil fuels, not clean energy. And second,

that it's remotely significant. If they

prove 0.001 more global warming, I'm not sure what to do with that impact. And

they never gave us. There's some

littering from electricity. Obviously,

you have laws that prevent this company these companies to pick up their lithium and so on and so forth. Their second

claim is that it supplants critical thinking. One, we have responses. For

thinking. One, we have responses. For

example, lots of universities go back to closedbook exams. Number two, obviously use it in an additive way to learn more facts as you write your essay. Third,

there are increased development of watermarking technology that uses lots of data to show where and when AI text has been generated. The point was lots of CG was bluster. It said AI is bad.

Destroys your critical thinking. They

wave their arms around. They did not prove it structurally. On [snorts] the rest of the debate, why are we going to win? The biggest reason why we win is

win? The biggest reason why we win is look at the motion mimicking imitating human behavior. To think like a human,

human behavior. To think like a human, to understand a human and persuade a human, you need to learn on real data, not biased data. What the AI researchers doing AI alignment like David Africa do?

They train AI on a cross-section of data across all of society. If you're a government using AI to affect policy, if you're a company using AI supply chain, if you're a scientist using AI to affect proteins, or a social scientist looking

at AI to measure economic outcomes, you are incentivized economically, socially, politically to make sure that your data is accurate. The key way that that AI

is accurate. The key way that that AI learns is through feedback. You tell it, here are my values. Here are my responses. The response from government

responses. The response from government is no, but it tricks you. It doesn't get real impact or doesn't actually tell you, you know, real things. So, you're

you're tricked in the short term. Two

responses. One, if an AI just makes two completely different arguments to you as closing government, as opening government says within hours, you stop trusting it. So the incentive is to be

trusting it. So the incentive is to be consistent. Second, AI learns over time

consistent. Second, AI learns over time that people have values like integrity, like like individualism, like liberalism. These values need to be

liberalism. These values need to be interpreted into binary code. AI has to understand what our values are, what our fears are, what our motivations are. And

AI given all the economic imperatives that we described is going to play a major role in society either way because it already is. Now I think the mistake that top half does is that they think of AI as chat GPT. Chat GPT is the tip of

the iceberg on AI. The vast majority of machine learning is being used in supply chains by researchers by scientists in vehicles. There are so many other uses

vehicles. There are so many other uses of AI that interface with people that research people with understand people which are not chat bots and which are used additively alongside people rather than directly interfacing. Closing

government tries to refute this empirical fact with very strange structural analysis. They say the mar

structural analysis. They say the mar market for AI chatbots is very large.

You do not become a dominant company like chatbt by charging people $20 a month. This is a very simple market.

month. This is a very simple market.

What you do is you license the software, not the chatbot, but the deep learning, the reinforcement learning, the machine learning out to companies, out to researchers, out to all of society and then they use that technology. Much of

AI will not be consumerf facing. It will

be employment and it will be agentic.

Right? Think about AI agents, things that shop for you, things that do tasks for you, things that actually replace human labor. Those things all need to

human labor. Those things all need to understand us because they interact with us every day because they affect governments and policy makers because they're obviously going to be used because of how powerful they are. Before

I go on opening, motivated by the incentives you described, companies will use AI to understand people on our side too, but because it won't be aimed at manipulating us. It won't focus on our

manipulating us. It won't focus on our weaknesses and our failures and our insecurities. So AI on your side

insecurities. So AI on your side understands the most base version of us.

Why would it care about us? Because we

should satisfy the preferences of the most base versions of us. Most of

advertising is an attempt to get at these things that's flawed and imperfect. It sells us products which

imperfect. It sells us products which are marginally good for us but not perfect for us. There are many attempts to advertise to people to persuade people that are imperfect and flawed and give us things which respond less to our

values and preferences. Last of all, the AI has to be so godlike in your imagination to believe opening government. To think the US government

government. To think the US government will just leave the president's phone lying around for some malicious foreign actor to put a big red button on it that's going to cause them to to drop a nuclear weapon instead of, you know, having anti virus software or so on. To

think that it's possible to manipulate people into obvious lies without the AI on the other side because of how cheap and democratized AI is pushing back on those assertions. It's not clear why it

those assertions. It's not clear why it empowers governments. Governments are

empowers governments. Governments are always empowered. AI is democratic.

always empowered. AI is democratic.

People have access to it. They use it to push back. So to the extent that there

push back. So to the extent that there is propaganda, it's propaganda from both sides and has a minimal impact. How do

you weigh us into the round? The key

thing that opening opposition struggle with is that they focused on some marginal impacts of better outputs. So,

for example, you could have AI friends, etc. To which opening government successfully says, well, these are a somalum. They're not always going to be

somalum. They're not always going to be good. And if OO agrees with their

good. And if OO agrees with their mitigation, as they do in a POI, then you're often not going to value this AI friend very much because, you know, it's AI. Does it actually resolve loneliness?

AI. Does it actually resolve loneliness?

Maybe it gives you a little bit of happiness in the short term. Probably

not in the long run. The second is this unspecified claim about economic benefits because of some supply chain optimization which they never cared to impact or significantly weigh in this debate. What's far more certain is that

debate. What's far more certain is that to teach AI to understand us, AI has to know us like the back of its hand. It

has to know our values, our preferences and our fears. Now the two important pieces of weighing on this number one even if you initially collect this data for chat bots to use on a consumer level that data is then resold and

cross-pollinated across the entire AI industry. So when you're training to

industry. So when you're training to build controls on AI, so your AI alignment research is trying to help companies build models that have stop buttons that say, "No, this is unethical. No, don't let this person do

unethical. No, don't let this person do this. Don't let them put that output

this. Don't let them put that output in." They're better at mimicking and

in." They're better at mimicking and understanding people because AI ultimately thinks through repetition.

What's it what it's maximizing for on our side of the house is understanding people and their values. And therefore,

it iterates and iterates and iterates closer to what we actually believe. Yes,

this may be used by some bad actors. But

the point is that the general average view of people is about respecting dignity, about respecting rights, about respecting honesty. These are things

respecting honesty. These are things which are socially glorified and exist in society which AI learn from with the unique access to data because you're economically motivated to train on this data. CG says you can still give it this

data. CG says you can still give it this data. There's no economic motivation to

data. There's no economic motivation to do so because there's no reward. B, it's

obvious that mimicry is a level way above the ability to write your essay.

At the end of the day, consider this. No

teams in this debate are going to be able to prove massive pie in the sky claims like the massive use of water or way more use of AI. AI is here to stay.

It's in our societies. It's in our governments. It's in our corporations

governments. It's in our corporations and it's in our phones. Do you want it to relate to you to understand it? Do

you want to understand that output and have it explain what it's doing to you?

Do you want to be able to do research to stop it from doing evil things? Only on

our side of the house are you able to achieve this. Very proud to oppose for

achieve this. Very proud to oppose for the last time. [applause]

I'd like to thank that speaker for that speech. Uh, did you enjoy the finals?

speech. Uh, did you enjoy the finals?

[cheering] Massive thank you to all the finalists and let's give them a one big round of applause at the end.

[applause] Yeah.

Thank you. I I'll give it to somebody from to explain what's going on now, I guess.

Everyone, moment of your focus, please.

What we will do now is take your things and leave the venue. You'll be back here at 7:30 sharp and we'll start the closing ceremony there. Salt.