Betting on responsible AI

Today on banking on AI, we chat with Chad Koziel Associate Vice President, layer six Gen AI delivery.

He walks us through his journey through AI, explaining how TD is approaching it, and gave some really calming advice for those who may be a little bit anxious about AI.

Chad, thanks so much for being here.

We're really, really excited to have you on the podcast.

You know, the first question a lot of people have is they say banking, AI, they they don't really think it goes together that often or it doesn't come to mind right off the top.

So when you hear that, I'd love just to get a sense of, you know, how you react to that and describe sort of what you're doing at TD Bank when it comes to AI.

So the first reaction is defensive.

And it's just it's defensive because, banks have been at the forefront of AI for decades now.

Right.

So beginning with the earliest models around and how we adjudicate risk, how we how we approach our customers and bring them in, how we support them that all of the what you're seeing in AI today is founded on those basic models.

So banks in some ways have gotten incredibly good at embedding AI across all of all of our operations for all of our customers.

The today like what, what people think of today in generative AI is empowered by scale.

And so if we look at where scale exists, it is in banking.

It is where we have millions of people who depend on products, produced by a company like T.D.

every day.

And it's that scale that has both enable genii, but then also pushes T.D.

and other banks rapidly to adopt it and makes it viable to do so quickly and well.

And so to that end, like TD is pushing hard and and fast to ensure that those two terms are synonymous.

They're not they're not antagonists to each other.

And that our colleagues and customers see the benefit of that.

So when you talk about scale, that's interesting.

So I say, you know, what are you doing in your day to day?

What does that look like?

How do you describe it to your friends who don't work in a necessarily?

And what's going through your mind in terms of what I'm doing today?

How are we going to scale this?

When somebody asked me, the first thing I'll say is I work in data and and I, I say that just as a bit of a play because of the attention that AI is getting.

Just quite frankly, I would rather fly under the radar a little bit.

That's that that's just sort of me personally. Right?

But engaging.

Engaging with the question. Right.

What I would tell people is, I bring AI to market.

I put AI in the hands of colleagues and customers.

That's purely my job, to make sure that when TDA is investing the enormous sums that it is to be at the head of AI, that we are delivering them, realizing the benefit that people are using it, loving it.

And if those things aren't happening, then I'm not doing my job.

So when you look back at the history of how you ended up here, you don't need to go through your resume.

But in terms of your journey to get to TD.

Walk me through that a little bit.

So without going into to all the details of professional poker and metallurgical engineering and deal making, we have.

To talk about that at some point. But maybe not today. But maybe not today. Right?

The what brought me to did was complexity and scale.

So I'd made I'd made some early steps into at the time, data science and data analytics in 2014.

But it was forgive the term, but it was small potatoes in terms of the the size and scale complexity that I got to work with.

And I crave that.

And frankly, there are only a small number of companies that offer that type of challenge, a very small number globally.

And so that that's what incited me to join today.

And then from there, it's been a journey from a digital analytics, capability into digital customer experience, and then finally into running an AI shop.

And you would have seen through your experience at TD in terms of the evolution of it, like I from my perspective, is moving faster than it ever has, and every day it seems to get faster.

I'd love to get your perspective on from when you arrived here to where we are today, and then we'll get into what AI is solving for.

But I'd love just to get a better sense of that journey being on here at TD.

I mean, your perception isn't wrong.

So there was we can divide AI a TD into let's call it like three three sort of life stages.

So the first is pre 2015 2016.

And this is where we're using traditional statistical models for garden variety sort of problems that a bank has.

Around 2014, 2015, 2016 we have the rise of an algorithm called XGBoost, which just enabled us to do those things so much better, so much better, so much more robustly.

And so this leads to TD acquiring layer six.

This was the fundamental capability.

And there was a period of dominance from 2015 to about 2023, 2022 where that modeling paradigm, that AI paradigm reigned supreme.

November 2022.

Very specific, very specific.

This is the launch of GPT three, GPT three.

And this is the seismic shift, right?

So we can pinpoint to a few research papers, critical research papers, other factors that led up to that moment.

But November 2022 was when someone, anyone could open up a web browser and observe magic in the form of a AI that would legitimately pass the Turing test and would legitimately interact with them as a human would.

And not just that, that conversational capability, but also the precision with which it could complete tasks that were here to for completed by humans.

That, that just that was the gasoline and the fire all in one.

And so from that period onwards, it has been a hard sprint in AI.

It's been a question of how well.

So in Genii, we can see that it's pretty easy to see what an AI can solve and look at it and say, hey, it looks like it can do that.

It's a lot harder to prove we can do it well and then to deploy it.

And so this three year almost sprint now is around ensuring that TDI can prove and can do all of these tasks extremely well.

And then to scale, to deploy, to get to market in a responsible way.

And so your perception that there was a seismic shift absolutely on point.

And the real question I think is where on the we're on that curve are we.

And I don't know the answer to that, but the shift was there.

And we will see the after effects for years to come.

As from a product and engineering standpoint, from a science standpoint, we get better and better at mobilizing that.

So we're seeing the issues customers are dealing with and clients that they're dealing with.

What's the team and the teams around the bank looking to solve for, and how are they making AI a central piece of that solution?

So maybe there's there's two primary fronts to that.

So the first is that the obvious first uses of ChatGPT are around.

Help me work through a problem right here.

Something that I can probably do it myself.

Could probably do it a little quicker if I had somebody looking over my shoulder, somebody who is an expert in it.

And so the first push is going to be everybody should have that, everybody should be accelerated by that.

And that will not just be our colleagues, right.

Who experience it through Microsoft Copilot capabilities.

It'll be our customers as well.

We're building those that they have that person looking over their shoulder saying, hey, here's here's how you can think about, for example, like your spend history.

Here's how you can think about these products and services, what they could offer you.

The other piece is in terms of actually having a model, a gen AI solution, and I, an AI product, perform some of those tasks, some of those tasks that are heinously complex but are also sufficiently repetitive

that they're really no fun to do, right, no fun to do.

And they require different experts to intersect, to interact, to get something done. Right.

And so we can we can think of all these sort of long paper based processes that occur at a bank, and you might be shocked to discover we have a few of those.

Right.

And a little surprised, actually.

Well, we all are right.

But, the fact is that that those are the types of things that you imagine, all these, all these, these paper processes of applications of forms and such, these have enough variation that through any sort of traditional rules based approach, any traditional, really basic model based approach, they were impossible.

You needed a human.

You needed a human to look through.

You needed a human to identify whether there was something wonky.

You needed a human to ask. Wait a second.

This looks more complex.

We need to we need to look at this other thing as well.

And for the very first time now we have AI's that can do that.

And so that's that's probably the transformational aspect.

And you can just imagine for someone like a customer who previously may have had to wait days, weeks for something to wind its way through, the opportunity to complete that in seconds, minutes is now there.

And that's what we're hoping to share with our customers pretty soon.

Changes things drastically, and so, and that brings me to my next sort of inquiry in terms of, you know, AI exists everywhere.

If you're not doing it now, you're behind the ball in smaller places, like you said, you know, smaller potatoes, you can move and develop and design and at different speeds because the bank at the end of the day is a large, heavily regulated organization.

There's no way around that.

So I'd love to just hear from your perspective the uniqueness of working on something so cutting edge in an environment that we know has lots of checks and balances.

I mean, the uniqueness is, is, is this, do you like a challenge?

Like it's, we can go faster.

We can deploy easier at a small company.

Maybe it's easy.

We can, we have better infrastructure at Big Ten, big tech companies to have built their organizations from the ground up.

In the digital age.

At G.D., you get the intersection of all of that complexity, the regulatory oversight, the, the challenge of proving and, frankly, like, that's fun.

It's it's it's awful on some days, but it's it's generally fun to work through that type of scale problem.

And so what it also, you know, sort of paradoxically, enables TDI to do things that those smaller companies can't.

The, the scale enables us to look at concepts like, I presume, and novel AI capabilities that those smaller companies can also, paradoxically, our RTD is built on trust our customers trust us where we are regularly regulated because of the types of data,

the types of interactions that we have with our customers.

And so that puts us in, in my opinion, in, in both a, a position where we must be thoughtful, it must be careful, it must be responsible, but also an exciting position where if we maintain that trust as the regulatory bodies are a part of that, we can accomplish things with the size and types of data that it has

that those big tech companies with all their infrastructure cannot.

And I think that's exciting in a way.

We'll be able to do things for our customers that nobody else in the world can.

For all of the innovations you're seeing at meta, at Google, at Microsoft, at OpenAI, they won't be able to do these things for our customers, that we will.

And I think that's pretty cool.

I think it's great.

I think it's some good taglines in there.

Actually, for our AI future, I would like to dive in just a little bit in terms of, you know, we talk about AI's coming and the humans involved, humans not involved.

What this looks like.

This is also probably another party question that you get quite often.

People to some degree are very excited, also a little nervous.

So when you're chatting through that nervous individual in terms of what this looks like, what do you what are you going to tell them?

What do you let them know in terms of where this is going?

I start by asking like what?

What have you experienced this been?

What makes you nervous but make you scared?

There's I mean, so we like we talked about this November 2022 moment right where magic appeared.

Right.

A sufficiently advanced technology indistinguishable from magic effectively.

And that necessarily raises some questions.

It raises some anxiety in terms of the capability that has.

So I but I would start with asking you know what.

So I mean maybe like let me toss it back.

Like you know what mystifies you.

What are you anxious about?

I think the reality most people look at is, yeah, I think they're in one of two camps.

They're either they actually don't know that much about AI, and it just sort of is a fear because it's the unknown, or they've dug it a.

Little bit by May, just most people don't.

Most experts in generative AI and in AI, including myself, don't.

That's an interesting perspective.

But sorry, like, you know.

In terms of I think a lot of people go to what does this look like for the future when it comes to employment, when it comes to how we live, how we operate?

I think that's the biggest fears that come in.

And I know there's no singular answer.

There's no crystal ball that we can look into.

But when you hear things like that, what goes through your mind?

I mean, my first answer is always, I don't know, I, I, I will gladly, shy away from that question and I'll just, just share that if we look at the last let's go back to 2017.

So attention is all you need is the seminal paper that kick this off and brought Transformers into the world after a fashion that was Google.

The next big milestone was OpenAI and GPT three and 2022.

Before that, OpenAI was seen as a second tier player in this space.

This year, the major market movement that occurred in January was with the launch of Deep Seed.

And what we could quibble as to whether Deep Seeded is truly the earthquake that the prior two events were.

What we can't argue about is that they are all coming from different places.

And so what you're seeing is, with the pace of innovation, with the spurts of, revolutionary growth that are occurring, they're not coming from predictable points.

They're not.

Again, you have Google, OpenAI deep seeded in this case, but these are not the same players.

And so if we look to what that future is, I would be extremely hesitant about trying to predict it, even from my position as a as a semi.

If I had to make a guess, turn me down.

I have to do on this podcast.

Which is fair.

Which is fair, right?

So if I had to make a guess, my guess would be that we're going to see some tremendous advances in science and engineering that overcome a lot of the major gaps that you see.

So when you use these tools, if you use them enough, you play with them, you'll break them.

You'll break them because they lose track of things.

They they forget, they don't they can't remember for that long.

They don't have access to real time emerging, emerging information natively.

We're going to see a lot of those challenges start to go away.

So we're going to see a lot of these become, much more integrated into our daily lives in much the way that we see, potentially a human, a support.

Now, in terms of what that means for jobs, Microsoft Copilot is probably the start.

We can imagine everyone working everywhere is going to have something supporting them every day.

We're going to see that probably people who are better at using knowledge and are more likely, I wouldn't say better.

I'll pick that back.

More willing, more more assertive and challenging.

The tools and making use of them are going to get much better at their day to day lives in all sorts of different ways.

You're going to be able to have your own personal assistant.

It's going to scale that as a capability.

Heretofore, that was the realm of the extremely wealthy.

You're going to have something that can make appointments for you, that can that can get you and get you a reservation that can cancel those things, that can, help you solve the problem of running late in the morning.

And how do I react to that with my kids, with, with my work, etc.?

It's going to be embedded, and people who are willing to make that leap, much as we saw in the digital age, which is probably our first push into that scale, are going to be able to potentially live, you know, much more productive existences at work, going to have access to these luxuries that were previously inconceivable.

That's my hope of the next sort of three years, five years, the big question, if I may, is whether we see another revolutionary change.

So if we look from this, GPT 345 was released, I think that was last week.

We're seeing an evolution constrained.

We're not seeing Quantum Leap from version to version over and over again.

And so the real question to me is, does all of my thinking on this get invalidated by a revolution that occurs, and the amount of external investment pouring in suggests that that might happen?

It increases the odds of it happening significantly compared with, you know, just a few years here years ago.

And we will see of how earth shattering that future is.

We'll see whether we get that revolution.

Even that last question.

So wonderfully, you know, all kind of came and tied together in terms of where we may be going.

I want to go a little bit back again, if we can, just in terms of your history in this interesting world of AI, how you ended up here, you mentioned something about poker, and I'd love just to hear a little bit more of that history and how you ended up in that world and what got you into it.

I mean, so I my, my professional history starts with, failing out of metallurgical engineering.

It wasn't for me.

I won't get into the details, but it was messy.

And, you know, left to my own devices as a really early, really early 20s kid, I had some skills.

I knew how to play poker.

I did it, not the, I think that's.

Like the setting, like, of above, a film that, you know.

Well, there's not a cast in, you.

Know, probably probably a grittier film if I'm going to be on it.

Right.

So, I mean, you imagine professional poker, you can see it on TV, right?

The vast majority of players of professional poker players are either sitting behind a screen 16 hours a day or sitting at a lonely casino table 16 hours a day, waiting for the right person to walk in that they feel, you know, in this turn, a fish, okay, so they can feel they can take money from I think.

I'd be a fish in this situation and.

And I mean, and most, most people who go into casinos are.

Right.

And so I was that guy sitting at the table waiting right.

Most days.

Friday nights and Saturdays are exciting because you get a, you get a crowd that, that you can, you can play with there.

But the rest of the time it's this grind.

It is a grind every day to make sure that you are sharp, you're disciplined, you're making the right decisions over and over again.

And the the life lesson from it, if you will, is that I was making money.

I was profitable, but the swings are wild, huge, and so the mentality isn't around.

Did I make money on that and did I?

Did I win that hand?

It is.

Did I make the right decisions in that hand?

Because if you make those decisions, you achieve what's called an expected profitability and expected value, right.

And so, ironically, that is probably the fundamental of what brought me into I this concept of not do I have the right outcome in every case, but are we consistently making the right decisions that lead us to make the right investments on behalf of colleagues and customers?

Are we placing good bets over and over again, and are we okay with failing when we place that good bet using good information and still got it wrong?

That's that's that's sort of like what poker is to me now.

Ironically, today I cannot play poker.

I don't enjoy it anymore. Okay.

So after about a year and a half or so of doing that I I, I was done.

You're done.

I was done for 16 hours a day of grind, that discipline.

That was tough, right?

And so at that point, I entered the corporate world, went through a series of degrees, you know, eventually landing here.

Right.

But that expected value mindset is probably the central, aspect of how I try to make decisions.

That's such a strong connection.

What an interesting story.

And you're using that obviously in your day to day when I, you know, talk to people in your field.

I'm always curious as to, you know, that one story that you have that you tell people and you know, they're going to get excited, probably because it excites you as well.

And if you have an example of that or something like that, or just let me know, like what excites you about this world?

I mean, there's there's a few moments to me.

So as an example, I think my big awakening moment was when I watched GPT 3.5 destroy the interview process that I was using for candidates, and it was at that moment that I realized we needed to undertake a fundamental shift.

Destroyed. In what way?

It's a it in seconds and not just a state in terms of answering the questions, but in terms of understanding and intuiting context in terms of, avoiding some of the traps that we would set within us in terms of identifying what questions it should ask

and when we when I saw that, I realized that how we are going to work, how we are going to, communicate the things we're going to look for in people are going to very fundamentally change.

And for me, that was that was the exciting part.

I would I would challenge anybody to use an image generation tool or to use ChatGPT.

If you've never used it to help you solve a problem.

And just an example, recently, you know, gone camping with my family a few times more thinking, okay, we'd like to maybe take this.

It's just something a little bit more intense.

Lighten our gear.

Get ready to go out into the wilds, if you will, and using ChatGPT or a GPT in this case to help me think through the process at a really detailed level of what gear I should change out.

How would I go about that?

How could I do that affordably?

And other people might, in that case, give up and say, you know what?

Like, I'll just I'll stay with this life experience of camping that I've got now.

I'm not going to pursue it that next degree because I don't know how to approach it now. I can now other people can.

now. I can now other people can.

And that's just one great.

Example of, the reality of it.

And to your point, you. Can.

That is scalable.

Everyone can ask that question to a certain degree if they have the technology in front of them. That's it.

It that that is it.

It is the, the scale, human level, reasoning that has never been possible before. They

amazing that all the conversation that we had has been just really insightful.

Is there anything that you you like to make sure is put out there to the general public in terms of AI, and particularly when it comes to AI in the financial world.

I would offer an assurance.

So when if I'm on, if I'm on the street, I'm hearing all sorts of stories and a lot of stories about an AI gone wrong, a company using AI in ways they probably shouldn't.

What I would say is that G-d is going to use generative AI.

We are going to use AI.

We are going to push it right to the limit, but we're going to be responsible about it.

We are going to do it properly in a way that benefits our colleagues and customers.

And that's the message I would send that.

Thank you so much for sitting down with us today. Thanks for having me.