How JPMorgan's embrace of AI could change banking for us all

JPMorgan aims to become the first major bank fully powered by AI. What does that mean for the future of banking?

Hugh Son, banking reporter for CNBC. He broke the news of JPMorgan’s adoption of agentic AI in his piece “Here’s JPMorgan Chase’s blueprint to become the world’s first fully AI-powered megabank.”

Karim Lakhani, professor of Business Administration at the Harvard Business School. He specializes in technology management, innovation, and AI. He is the co-author of the case study “JPMorgan Chase: Leadership in the Age of GenAI.”

The version of our broadcast available at the top of this page and via podcast apps is a condensed version of the full show. You can listen to the full, unedited broadcast here:

Part I

MEGHNA CHAKRABARTI: New AI products and strategies are being released at lightning speed. It was only three years ago that ChatGPT was first released to the public. Then quickly came AI tools that the public doesn't see but have an impact on everyone.

Things like AI assisted data processing or even decision making in the health insurance industry say. And in fact, I should correct myself, those tools were already at use before the public got access to things like ChatGPT, and many of those technologies were in the predictive AI category, which analyzed data to forecast outcomes.

Then the game changed again with the introduction of generative AI tools that synthesize brand new content. Everything from writing to music to code to all the AI slop that is sloshing around our digital lives. Now there's a third wave of AI tools coming. And these are called agentic AI. And to be honest, most of us here had never even heard of it until just a couple of months ago, and for a very specific reason. Now, first.

What is agentic AI? IBM defines it as "an artificial intelligence system that can accomplish a specific goal with limited supervision. It consists of AI agents, machine learning models that mimic human decision-making to solve problems in real time." Okay. So what is the thing that raised our awareness about agentic AI?

Just a couple of months ago, one of the largest banks in the world announced a massive plan to quote "fundamentally rewire" itself for the AI era. And how? By rolling out agentic AI across its $850 billion business. And that company is JPMorgan Chase.

JAMIE DIMON: It affects everything, risk, fraud, marketing, idea generation, customer service, and this is the tip of the iceberg.

CHAKRABARTI: That's JPMorgan CEO, Jamie Dimon. And although the company initially backed away from AI several years ago, it has since very much changed its tune. Dimon spoke at the America Business Forum in Miami in early November.

DIMON: It's going to cure a lot of diseases that we've all been afflicted with, okay. My guess is the developed world will be working three and a half days a week in 20, 30, 40 years and have wonderful lives.

CHAKRABARTI: There is just as much evidence, both current and historical, that AI disruption may fundamentally damage some people's lives. Those who end up out of work, out of money, and without purpose.

Nevertheless, Dimon sees JPMorgan's embrace of AI as a force that can shape the future.

DIMON: The way I look at is  every application, every job, every customer interface, everything we all do, and you're all gonna have your own personal assistants. You're gonna have agents that do research for you every time you wake up in the morning.

CHAKRABARTI: JPMorgan Chase plans on using agentic AI to automate every behind-the-scenes process at the bank. Every one of the company's 300,000 plus employees will be expected to use it. Dimon says every week 150,000 employees are using AI systems the company already has in place. And last month he spoke with Bloomberg's Tom Mackenzie and acknowledged that agentic AI will change how jobs are done at JPMorgan and in the broader finance industry.

DIMON: I think people shouldn't put their head in the sand. It is going to affect jobs. So think of every application, every service you do, you'll be using AI. Some to enhance it. Some of it'll be you doing the same job, you're doing a better job at it. There will be jobs that it eliminates, but you're better off being way ahead of the curve and retraining people.

So we train and redeploy a lot of people. So for JPMorgan, we're successful. We'll have more jobs, but there'll probably be less jobs in certain functions.

CHAKRABARTI: And eventually JPMorgan's agentic AI will come to you. The bank wants every client experience to be quote, "curated" with AI concierges, but perhaps it's the behind-the-scenes analysis and decision making that could utterly transform finance.

JPMorgan isn't just a Wall Street heavyweight; it's a market mover. It's a too big to fail bank. It's a corporation whose decisions have the power to strengthen or topple the U.S. economy, which means you. So Hugh Son joins us now. He's a banking reporter for CNBC, and he was given exclusive access to JPMorgan's next steps in embracing AI.

He even got a chance to see the new agentic AI at work, and he joins us from New York. Hugh, welcome to On Point.

HUGH SON: Thanks for having me, Meghna.

CHAKRABARTI: Okay, so tell me more, first of all, in more detail the definition of agentic AI as you understand it, within the ecosystem of JPMorgan.

SON: So everybody here is familiar with generative AI and ChatGPT, chat bots essentially.

Now, agentic AI is a step further. It is, however, built on gen AI, it is built on this LLM technology. The difference is, I liken it to the difference of having an assistant versus having a colleague or a coworker. And by that, I mean you can certainly ask ChatGPT to summarize long documents, write your emails.

Even create images on some of these other platforms. However, it's an ask and deliver. Now agentic AI, the difference there is they can take multi-step processes, deliver on a goal. And autonomously perform those, the tasks required to come up with something. And an example of that would be, come up with a website consolidating all of your media appearances, Meghna. And perform that, debug it, create the steps for it, and present it back to me.

There are already platforms that can do something like that. And that is, we are at the dawn of agentic AI.

CHAKRABARTI: Okay. I can tell you that the slide deck of my media appearances would be one thing. On Point, but I'm very captivated by your analogy of it's more like having a colleague than an assistant that delivers parts of something that you're looking for. So colleagues make decisions in any business process or any team. And the thing that captures my attention the most is the decision-making aspect of generative AI. Can you tell me more about that?

SON: In any complicated job we can talk about websites, we could talk about writing a story, which is what I do, and then occasionally talking about it.

There are dozens of decisions to make. And agentic AI is in its infancy of being able to make those decisions on its own. And with something called reinforcement learning, which is penalties and rewards for a job well done. It's almost like humans, a human toddler would learn, which is fascinating on its own.

And so the emergence of agentic AI, of being able to make decisions, what's going to please Hugh or Meghna in this case of making a website at the end of the day. And if at the end of the process, whether it takes minutes or hours to come up with a website of all your media appearances of On Point. And you say, this is good. But I'd like this and that. It will learn from that. And so it's really, the fascinating thing is the parallel with the way people learn, the way we learn.

CHAKRABARTI: Yeah. It occurs to me that the public's understanding or even familiarity with agentic AI is so new that I accidentally called it generative AI a minute ago.

So this is just like a new phrase, even in the popular discourse about AI technologies. Hugh, you wrote, you essentially broke the news about AI and JPMorgan in an article called “Here’s JPMorgan Chase’s blueprint to become the world’s first fully AI-powered megabank.” Now you got to see some of this agentic AI at work at JPMorgan.

What did they actually show you?

SON: Yeah, so I was the first outsider to see LLM Suite, which is their sort of unimaginatively named portal. Now the cool thing about LLM Suite is that it taps into, and they want to be model agnostic. So OpenAI's model, obviously that's on there. That was base case for both JPMorgan, Morgan Stanley, Goldman Sachs, the big three.

They added an Anthropic, which is turning out to be really good at lots of different things in the corporate sense. ChatGPT, more consumer. Anthropic, more corporate.

And the first use case they gave me was a fascinating one, which is for those listeners who know more about investment banking, and I'm assuming most do not, there are the farm system of investment banking. Our young, bright college grads. Usually from top universities, and they are called analysts. So they have these two-year analyst programs they start at summer internships when they're still between, I think junior and senior year or even sophomore and junior year.

I think junior and senior year. They ultimately feed into the, and I'd liken them to the foot soldiers of Wall Street because there are many of them. Most of them don't make it. High mortality rate, and these are the kids, these are the 22-year-olds who are up all night. And so the classic example is you have your managing director who's pretty high up, pretty senior. Say, yeah, I'm meeting X, Y, Z company tomorrow morning. I'm going to get on a plane at 6 a.m.

You, my team. And I guess he's talking to his VP, who then talks to the team of analysts, ultimately. You come up with a pitch deck. Okay, you give me a 30-page pitch deck about X, Y, Z company. Their opportunities, their risks, maybe their need for capital, maybe their need for M&A, who they could acquire. Okay. And so that's the classic model. And then, and you've heard many complaints about this over the years, especially recently, that you have five or six of these junior bankers working all night.

Really until 2, 3 a.m. Killing themselves, essentially delivering this. And if there's a little margin off, or God forbid a factual error in this pitch deck ... people get screamed at.

Part II

CHAKRABARTI: Let's get back to your example. I can see every movie ever made about Wall Street in my head and those giant rooms full of like a lot of computers and young, fresh out of college faces.

Bleary-eyed after crunching the numbers for a slide deck that the exec expects first thing in the morning. Instead, what would the agentic AI do?

SON: The demonstration for me is that it did it in about 30 seconds.

CHAKRABARTI: Okay. My immediate next question is that sounds like a job wipe out for entry level analysts at JPMorgan.

SON: So to be clear, it is 80% of it. And it's not perfect. And the prompt was, you're a technology banker at JPMorgan Chase. Tomorrow you're meeting with a CEO and CFO of Nvidia, come up with a five-page pitch deck, including all the recent news. And so it's pulling in, we talked about agentic taking multi-step tasks autonomously.

It's pulling in data sources. And putting it into what is probably a template, right? These aren't, you're not recreating the wheel with each pitch deck. And so what will happen then is I think you'd have a senior banker sprinkle some of their own magic and their own humanity onto it.

And but still radically collapsing the time to your point, Meghna. And so what I think that means is one of the ideas being tossed around at several Wall Street firms. They are reticent to say this on the record. They are not behind closed doors and not for attribution, which is, there's just going to be less of a need for these people.

And it's not, in that Marvel movie, I think the last Avenger, there's been so many, the Avengers, where they snap a finger and half of humanity goes away. That's not really the case in this. It is you still need people. It's just that you won't need as many. Because the ones that you have will be using AI and some of the scut work, and it is scut work, will go away.

You still need people. It's just that you won't need as many.

Hugh Son

CHAKRABARTI: So we're going to come back to the jobs issue in a second. Because through your reporting, I also learned that really the people who probably are less at risk are those who are client facing. But even there, agentic AI is going to be sort of part of the client experience. So we'll come back to that in a second.

But how much decision-making power does JPMorgan plan on giving its agentic AI? Because the example that you gave us makes perfect sense. Okay. To be honest. It's actually a fairly low-level job. But could we see a future in which this agentic AI is doing the analysis and then make either making a decision or providing advice to a senior exec on, say, should JPMorgan provide the leverage to the Saudi Sovereign Wealth Fund to purchase Electronic Arts? That would, that is a huge deal, right? That's a multi-billion-dollar deal. Could we see a future in which the AI is just, yeah, let's do it.

Or even just approves of it.

SON: You've probably heard of this concept of human in the loop, and what I think is possible for an end state, which could be years out, could be single digit years, it could be decades, maybe never, is that you can have a situation in which the vision for this is a human at a control panel wakes up in the morning.

This is the news you need to know, Hugh. These are the actions you need to take. In the case of this proposed M&A deal, this is the bull case. According to our data flows, we sense that this entity, their revenue's grown really fast. However, their capital lines are depleted, perfect scenario for raising capital.

Perfect time to give them a call. Here's their number. And the human in the loop for relationship-based businesses is never, is not going away any time soon. And so I would think of it as commodified things. Hugh Son applies for a credit card. Yes? No? Hugh Son has already been approved for this client relationship. Now we need to do a bunch of KYC. Stuff that is rote will go that route to full automation. It's rules based. Stuff that is relationship based, won't go away anytime soon. And so that's the way I think of it.

Rules-based stuff is very prone to AI, relationship-based things, that's much harder to displace.

CHAKRABARTI: Yeah. And this is interesting because the different kinds of businesses that JPMorgan and Chase is in, the amount of people that this could have an impact on really changes, like customer credit card stuff.

Millions and millions of people are making those applications to JPMorgan Chase all the time. And so this is one of the reasons why, you know, seeing the various levels of decision making is very interesting to me. Because for the average American, there could be instantaneous decisions made about their financial lives or their access to credit, which could have an impact.

So Hugh, hang on here for just a second. Because I do want to continue to hear about your experiences with this agentic AI and also the inside story of JPMorgan Chase and its adoption.

Actually, let's just jump to that right now. I had the understanding that Jamie Dimon and the bank as a whole, not that long ago, were not the biggest fans of incorporating a lot of AI into JPMorgan. And what changed?

SON: So my knowledge of that storyline isn't the strongest. I will say it does rhyme with a lot of the most recent, much less spectacular technology, which is Cloud. And so they, and a lot of their peers in the big banks were very hesitant to get into Cloud because of security reasons.

And look, these are old school people in their 60s. And they've done, and banking has been one of those kind of industries that hasn't changed for two or three decades in some disciplines. And so I wouldn't be shocked if they, 2023, they're looking at it. Late 2022, ChatGPT comes on the scene.

They're just looking at it and they think it's a toy. And what happens over the time is, if I were to speculate, they've obviously seen the traction, quite often it's people's younger colleagues or their children.

CHAKRABARTI: I was just going to say.

SON: Have you seen this? And what else happens is their internal bankers, their TMT bankers, their tech media and telecom bankers say, you've got to get your eyes on this. And they go to the West Coast, and they make a tour of OpenAI and Microsoft Alphabet, et cetera, and they talk to the evangelists. And then the next thing you know is Jamie Dimon in his 2024 annual investor letter says, this could be as big as electricity.

Or the printing press.

CHAKRABARTI: Can I just make a note about that? Because the cut that we played at the top of the show, of Jamie Dimon saying, we are, we could be entering a world where developed nations, people, will only have to work for three and a half days a week.

That is, I am not a Luddite, but that is the kind of smoke blowing that you hear coming out of Silicon Valley.

SON: Yeah, that's a head fake. (LAUGHS) So when we all went home during COVID, there was for many, for several years and including up to the number two at JPMorgan Chase, this idea of hybrid being the go forward, because it's resilient. When we're in Hurricane Sandy and you lost some offices and some capabilities you had to revert to, back offices across the river.

There is something inherently resilient about hybrid. And what does Jamie want you to do? He wants you in the office five days a week. Now a scenario in which Jamie Dimon or somebody like him is running JPMorgan Chase. You're never going to work three days a week. I would put money on that, Meghna.

CHAKRABARTI: I guess what I was referring to is something that you told our producer, that you told him that the bankers meet with AI clients in California or AI makers and smoke what's passed to them? I would think that they would be far more critical thinkers.

Am I just being too cruel to very wealthy?

SON: Are you talking about the bull case on AI and how we could change society or something else?

CHAKRABARTI: Both. The bull case on how we can change AI. And why so quickly people like Jamie Dimon went from being skeptical about AI to announcing that the bank would basically become, as the headline of your story said, an AI powered mega bank.

SON: Yeah. All the incentives are in the bull case. And there's this tantalizing future for JPMorgan. And it was in the top of your segment, 300,000 plus employees. Many of them, many of them are doing processes and behind the scenes processes in which a client is taken care of, but they don't really, quite frankly, care how they're taking care of, as long as it's not, there's no mistakes.

And so you have lots of people there who are employed, the lubrication of the pipes of finance. If those people can, if you can get away with doing 25% more business with 10% for fewer people, which is by the way, the articulation of one specific business line, the consumer bank and specifically fraud and that type of thing, operation staff, they call it. And that's three to five years. And that's conservative by the way. What does that do to your stock and ultimately your own net worth? Up into the right.

CHAKRABARTI: Yeah. Okay. Up into the right. Hugh Son, hang on here for just a second. Because let's hear for a moment from a very high-ranking officer at JPMorgan itself, this is Derek Waldron.

JPMorgan's head of Chief Analytics, where he is tasked with, quote, accelerating the firm's AI and machine learning agenda. And this is a clip from when he spoke with a representative from McKinsey and Company earlier this year, and he talked about the ways in which JPMorgan is already using AI.

DEREK WALDRON: It's used in tens of thousands of different ways. Lawyers use it to scan contracts, read contracts, compare contracts, generate contracts, credit professionals, use it to read terms, compare covenants, extract information, sales professionals and frontline bankers use it to help distill information and prepare for the meetings.

And I could go on and on.

Let's bring Karim Lakhani into the conversation now. Karim is a professor of business administration at the Harvard Business School, specializes in technology innovation and AI, and is the co-author of a Harvard Business School case study called “JPMorgan Chase: Leadership in the Age of GenAI.”

Professor Lakhani, welcome to On Point.

KARIM LAKHANI: Hi, Meghna. Good to be with you.

CHAKRABARTI: It's great to have you. Let me first ask you, actually, again, just I'd like to step back and understand the internal transformation in JPMorgan that led to this. Do you know more or do you have a bigger sense as to how Dimon and his other C-Suite colleagues became such fans of agentic AI?

LAKHANI: Yeah, absolutely. Our case study came out in April. We used it for our MBA classes at HBS. And I had a chance to get a view about this since almost about a year ago, we've been tracking this transformation taking place at JPMorgan. And look, there are two key events. As Hugh mentioned, there's definitely like the ChatGPT moment that they all encountered and immediately their reaction was, let's shut this thing down. And which is a natural reaction in a bank, an institution like JPMorgan, because oh my god, AI run wild.

All of our people are going to be abusing AI. It's gonna be crazy. But at the same time, there are basically like three heroes in this story about JPMorgan. Lori Beer, who's the chief information officer. She has a $18 billion technology budget.

CHAKRABARTI: (LAUGHS)

LAKHANI: Which is unbelievable. Teresa Heitsenrether, who is the chief data analytics officer. And Mary Erdoes, who is the CEO of Asset and Wealth Management.

[Lori Beer, Teresa Heitsenrether and Mary Erdoes] all saw this as a massive opportunity to transform how work gets done at JPMorgan, but also how they serve their customers.

Karim Lakhani

And so they were saying, we have to find a way to use these tools internally and to actually figure this out. And so they were on a fairly fast journey to figure out how they could actually adapt these large language models internal to JPMorgan in a secure, safe environment. And then to ultimately drive the democratization of it. And so it was, I got a front row seat of the conversations as they were emerging.

And that's what allows us to write these kinds of case studies. And it was really remarkable to see an organization of that size pivot. And I think the reason was that this was, like, oftentimes when I see technology leaders, sorry, business leaders, CEOs and C-Suites, the technology is done for them, right?

They will say, 'Hey, IT flunky, you go do my tech for me.' And then they'll get some dashboard or some report. The difference with generative AI has been that you actually have to do the AI yourself and all of these leaders, Lori, Mary, Teresa, did the AI themselves and could see how it was transformative for them.

So it was no longer like getting a report about it or getting a 50-page eye watering deck about it. It was like, oh, I'm using this to solve my own problems. And if it helps me in my tasks as a leader of the bank, what else could we be doing about it? And so that shift of leaders using the tools to solve their own problems in their own work was the big moment when they said, 'We gotta figure this out.'

That shift of leaders using the tools to solve their own problems in their own work was the big moment when they said, 'We gotta figure this out.'

Karim Lakhani

CHAKRABARTI: Oh, interesting. Obviously, I want to bring a healthy skepticism about the human impact of any massive AI transformation that we have going undergoing now. But to be entirely fair, it also seems to me that finance is the perfect industry to introduce more and more sophisticated AI tools. Because so much of it is analysis. And is agentic AI simply just that plus a little more, or maybe a lot more decision-making post the analysis?

LAKHANI: I think the way I would think about this is that at the moment, most of this is going to be augmentation. So that it helps you as a banker, as analyst to make decisions. The systems will recommend and then you'll make decisions.

Now there are some settings like today even when you use your credit card and there's a fraud notification and your card is blocked automatically, that's like a machine learning system basically deciding, I don't need to go to a human to make a decision. The patterns tell me I should just shut this down.

And the AI tools are getting this way now. The agentic tools are getting this way as well. But as you were discussing, making a recommendation to make a billion-dollar investment and a big fund, they can make the recommendations, but the decision will still be up to humans.

Part III

CHAKRABARTI: Even though I said that finance is like the perfect industry for AI to do a ton of work, it's also an industry that has to be extraordinarily careful with its data, right? Just for the safety of the economy, for the bank itself and its security, and plus just all the money that we're giving to the banks. And in order to also satisfy federal regulation.

AI, by definition though, requires like the constant eating and analyzing of data. I presume that JPMorgan isn't sending their data out for AI to analyze and just bringing it back in. What are the security concerns here?

LAKHANI: Yeah, 100%. And look, every industry cares about data, right?

So health care has the same concerns, manufacturing has the same concerns and so forth. And every company that I talked to, the first worry is, what happens to our data? How is it secured and so forth? And if you look at a company like JPMorgan, this part of their $18 billion tab is cybersecurity, right?

They're a massive threat vector for people to come in and disrupt them. And so they have, and along with working with the hyperscalers, they have worked very hard on ensuring that the data stays within their premises and is controlled. And they can use the model's superpowers, but the data stays with them.

And I think that's part of the magic or the trick that companies need to do to understand how to utilize these superpowers from these companies, yet preserve their data, preserve the security and the sanctity of the data. So that's part of the vast shift happening in the broader tech sector to ensure that technology becomes available in this secure way.

CHAKRABARTI: In fact, I think you've said that it was JPMorgan that really pushed, say, OpenAI and the other AI companies to create enterprise models, right? Where it would stay within the four walls of the bank?

LAKHANI: Absolutely. Both JPMorgan and then just up the road from you at Moderna as well.

So Moderna has also been a big the pharmaceutical company, the biotech company has also been a big proponent of using generative AI for the internal usage. And both of them simultaneously, in two very different sectors, said in order for us to use these tools beyond sort of toy applications you have to work through on the security concerns.

And Microsoft and Google and OpenAI and all these companies and Amazon, they've all worked pretty hard to make sure that they can give you those guarantees for data security.

CHAKRABARTI: Okay, Hugh, I wanna go back to the issue of jobs. Because this is something that with any technological revolution, this is the first and one of the sharpest pain points for not just a company, but a country as a whole.

So this is JPMorgan Chase CEO Jamie Dimon, and he specifically addressed recently the concerns about job loss when discussing agentic AI. And he says the way to address these concerns is through considered decision making.

DIMON: And I will give you a thought exercise. I think there are 2 million commercial truckers in the United States, and if somehow you can push a button, convert it to AI, it's cheaper, faster, safer, less CO2. You want to do it, but you shouldn't do it if you're going to put 2 million people out of jobs. And the way to handle that is retraining, income assistance, slow it down, early retirement, relocation programs, and it's all doable. And that would have to take both government and business doing that together.

CHAKRABARTI: Hugh, is JPMorgan even launching any such kinds of assistance programs like Dimon just talked about?

SON: I think it goes down to the business lines. They've got, as we talked about, 300,000 plus employees. I do think that there is at least the possibility that if you've been, I hate to say, replaced by a AI robot, or more likely, replaced by processes that are empowered by AI and by people who are wielding AI, and you fall on the wrong side of that trade, unfortunately, in the coming years, you will have, at the very least, the ability to apply for other jobs internally.

There are always job openings at the same time. There are job closings. There could be training. I haven't heard about that yet. Be a nice thing and we will hold them to account on that.

CHAKRABARTI: Yeah, because as you wrote in your story, these are questions that still, the answers are slowly unfolding.

Will these companies actually retrain workers displaced by AI? But also, I'm looking at your story. You reported that in May, the company's consumer banking chief actually told investors they have a prediction about jobs within the company that may change due to AI. Do you remember what that is?

SON: And this is specifically the operations use case?

CHAKRABARTI: Yeah.

SON: I mentioned earlier, which is unsexy work. It is onboarding accounts. KYC. It is making sure that the flows happen, and they happen predictably when you're talking about trillions and billions of dollars in payments.

And that's something, you can imagine a scenario in which people who are wielding AI replace processes that were not completely manual either. It's not like each account and each transaction was being checked by hand. Obviously, as Karim mentioned, machine learning algos are doing much of it, but there is more scale at play.

And so if you're trusting AI agents in AI processes to be more of what humans used to do, which is that mid-level manager, you can rise and if you succeed in that, you'll be a higher up in the food chain, managing greater amounts of flow and that's certainly what they see.

And I think the meta story here, Meghna, is that it's not just that they're telling their, it's not an internal message, not only an internal message. It is obviously and so people cannot be surprised that, hey, this is coming. But it's also an external message. The messaging is, we are going to be one of the winners of this trade, therefore we merit a rising valuation.

The messaging is, we are going to be one of the winners of this trade, therefore we merit a rising valuation.

Hugh Son

And that's part of the story here.

LAKHANI: If I may add something here.

CHAKRABARTI: Please go ahead.

LAKHANI: I think the jobs thing is a very critical part of this story and story that's going to be with us for the next few decades, actually, in the banking industry. We've seen this movie before, right?

As you recall, ATMs were introduced in the '70s and '80s, and there was this big worry that all these bank teller jobs would disappear. And in fact, the opposite happened. The number of teller jobs increased because what ATMs did overall was that they reduced the cost per branch, right?

You could have fewer tellers per branch, and you could open up more branches. And so from 1980 to about 2010, the number of bank tellers actually increased. And the job changed from just handing your money to being a relationship advisor, being an advisor on selling you products from the bank and so on and so forth.

And so I think the part of what we need to keep reminding ourselves is that as technology starts to work along the functions that other people, that humans did, that there's also a transformation that takes place. And if you go back to the history of our country, the agriculture sector, in the 1800s, used to employ close to 50% to 80% of American population.

And then today only 1.2% of the U.S. workforce is engaged directly in agriculture and farming. The rest has been automated, and we have more food than ever, and we are still at historically lowish unemployment rates. And so I think we have to keep thinking about these types of technologies in the end as transforming the nature of work that humans do.

What's happening is it's now happening across all sectors. Not just in banking, but across the board. And we have to go through this transition period that Hugh is so eloquently talking about. That we have to actually start caring, right? And the caring responsibility is with the leaders, right? The leaders have to care about what this means for their companies and for their workers.

The leaders have to care about what this means for their companies and for their workers.

Karim Lakhani

And of course, our dysfunctional governments, to actually invest in the caring part as well. But I think that responsibility lies squarely with the Jamie Dimon's of the world, the CEOs of the worlds, but also our political leadership. To make sure that these transitions work in our favor, not against us.

CHAKRABARTI: Agreed. Because the agriculture example is an interesting one. Because you're absolutely right about the change over the past century in terms of now we are at a place where American farms are super productive, but with far fewer people. You were making the point I was actually going to ask about, which is frame of reference, right?

In terms of time. And I think that the immediate frame of reference in terms of the change in disruption that AI can have on the lives of millions of people is the time reference is now. And the impact is large. And we won't get into this now. We've done lots of shows about this and we'll do more, but like the potential for social disruption is also there.

LAKHANI: 100%. Yeah. And I think the thing to note, and I think this is again part of why leaders need to pay attention. All types of leaders need to pay attention, is that the technology is getting better exponentially every six months, nine months, 18 months, and we're not used to seeing exponential technology increase in these shorts amounts of times. In terms of performance. And the performance, really what I say is that this is about underlying expertise.

What we're seeing is that the cost of expertise is declining. Because these models are basically inherently encapsulating human expertise when they're deployed. And so the cost of expertise is declining, and if companies and organizations are nothing but bundles of expertise, finance, marketing, sales, production, interviewing, for example, and that cost is declining, right?

Then how we think about jobs, how we think about roles and functions and processes becomes a really urgent issue for all of us.

CHAKRABARTI: I have to say to be perfectly honest, I think about this a lot. That, I don't know, agentic AI or even generative AI right now could be, I guess not sitting in my place, but in the computer that I'm looking at and doing this show without me.

LAKHANI: I did use the agentic AI,  you know what Hugh was saying, I created a website for you. I think your producer has it, while you and Hugh were talking.

CHAKRABARTI: Stop. (LAUGHS)

LAKHANI: So I went ahead and created a website for you. It's done. Yeah. You can share it later.

CHAKRABARTI: Okay. (LAUGHS)

LAKHANI: But also, what I did was, in preparation for this interview, I put in my case study and my teaching notes, and then I said, I'm being interviewed by Meghna. How should I prepare for it? And so then it worked for about 10 minutes and then gave me a bunch of questions you might be asking me and how I should show up in front of you. So even for my prep now, as somebody who does media interviews and so forth, right?

It's changed dramatically.

CHAKRABARTI: Okay. I have to ask one question, then there's something else very important I want to go to. Actually, my next question is going to be about the risk of too big to fail in AI. Did the AI tell you that I might ask about that?

LAKHANI: It asked me about the job site for sure, and it asked me about the social concerns for sure.

Let me see if it asked about, it wrote a long report, so I didn't even get a chance to read all of it.

CHAKRABARTI: (LAUGHS) Oh! You need an AI to summarize it.

LAKHANI: Exactly. Exactly.

CHAKRABARTI: Okay so let me just ask this question then. Because every time we talk about the big banks and JPMorgan. In particular, my mind goes back to 2008, 2009 and 2010.

SON: Of course.

CHAKRABARTI: And the CEOs of the world's largest bank or America's largest bank sitting there, raising their hands under oath saying, yeah, we made mistakes, but we were so glad that the American taxpayer bailed us out.

So we that in mind. I just want to play one more clip here. This is from IBM and an informational video they have publicly about agentic AI. And in the video, they highlight some concerns about the autonomy part of AG agentic AI.

As autonomy increases, so do risks like misinformation, decision making errors and security vulnerabilities, many organizations are still catching up with the generative AI risks and agentic AI just amplifies them.

Note with outcomes like these, there are even fewer humans in the loop.

CHAKRABARTI: So we just have a couple of minutes left and I'd love to hear from both of you. This question, which is, we already know that the banks have proven to the government or the government has responded that these banks are too big to fail.

How is accountability extracted when we're marching into a future where the AI is actually potentially making a lot of these decisions that could put the economy at risk? Hugh, your thoughts?

SON: It's gotta be explainable. And so if the OCC or the Federal regulators say, how did you, what happened here?

Can you explain this? You've gotta be able to open up the kimono, show them the steps that led to an underwriting mistake. You've gotta be able to show your work. And I think that's always going to be the case. It's always going to be the case that ultimately a human being will be responsible for high level decisions.

And an executive will be there. There's never going to be an AI CEO. And if there is, then I think we're all in trouble.

There's never going to be an AI CEO.

Hugh Son

CHAKRABARTI: Professor Lakhani, go ahead. I'll give you the last minute or so.

LAKHANI: Yeah, so look, just as ... we can scale the benefits of AI exponentially, we can also scale the harms of this exponentially.

And this requires, again, a new set of skills for our leaders and for our regulators. Our leaders no longer can outsource the technology decisions to the basements of the organizations, but it has to be at the C-suite. And that's why JPMorgan is interesting because the technology leaders report directly to JPMorgan. That's the first thing. The second thing is from a regulatory perspective, like our regulators need to get as smart about AI and technology and become users of the technology themselves as they are with everything else. So you can't say, we're gonna do an audit for you and then we're gonna do a paper-based audit.

You have to do algorithmic audits. You have to understand their systems and the capabilities. And so we are almost in this arms race where our regulators need to catch up with technology, not with more paper or more meetings. And I think that's the shift that I think we all need to be thinking about in our governments, in our regulatory apparatus, in health care, in finance, in construction, you name it, where all these technologies are being deployed that we need to all up our game.

We often say machines won't replace humans, but humans with machines will replace humans without machines. And this imperative is not just for leaders, but for all of us, as we face what these technologies can do for us.

The first draft of this transcript was created by Descript, an AI transcription tool. An On Point producer then thoroughly reviewed, corrected, and reformatted the transcript before publication. The use of this AI tool creates the capacity to provide these transcripts.