Revisionist History - IBM CEO Arvind Krishna: Creating Smarter Business with AI and Quantum

Episode Date: November 27, 2025

Today we're sharing a recent episode from another show Malcolm hosts, Smart Talks with IBM. Malcolm Gladwell sits down with IBM Chairman and CEO Arvind Krishna in a special live episode of Smart Talks with IBM. They discuss the groundbreaking potential of quantum computing, the transformative impact of AI on business, and how Krishna’s visionary predictions from the 90s continue to guide IBM’s innovations. This is a paid advertisement from IBM. The conversations on this podcast don't necessarily represent IBM's positions, strategies or opinions. Visit us at https://www.ibm.com/think/podcasts/smart-talksSee omnystudio.com/listener for privacy information.

This is an I-Heart podcast. Guaranteed Human. I'm Jonathan Goldstein, and on the new season of heavyweight... And so I pointed the gun at him and said this isn't a joke. A man who robbed a bank when he was 14 years old. And a centenarian rediscovers a love lost 80 years ago. How can a 101-year-old woman fall in love again? Listen to heavyweight wherever you get your.

your podcasts. Hello, hello, I'm Malcolm Gladwell, and you're listening to Smart Talks with IBM. This season, we've been bringing you stories of how IBM works with its clients to solve complex problems, like helping L'Oreal re-imagined how scientists approach cosmetic formulation or enabling Skuderia-Ferrari-HP to connect with fans in new ways. But in this episode, we're going to zoom out and look at the bigger picture. Earlier this month, I had the chance to meet the person who's shaping IBM's future. It's CEO and chairman, Arvind Krishna.

We sat down in front of an intimate live audience at IBM's New York City office and talked about his uncanny ability to anticipate where technology is heading, the future of AI, and his passion for quantum computing, which he says is as revolutionary as a semiconductor. Thank you, everyone. Thank you to Arvind. you're a difficult man to schedule for one of these things. So we're enormously pleased that you could join us.

Let's start with a, I have all these cousins, I have two cousins who work for IBM their entire career. I would ask them, what does IBM do? And they would always give me different, confusing, complicated answers. What's your answer? What's your simple answer to that question? IBM's role is to help our clients improve their business by deploying technology. that means you're not ever gated to one product.

It is what makes sense at that time. But it is about improving their business, not just giving them a commodity. And to go to the next layer, I would say we help them through a mixture of hybrid cloud and artificial intelligence, and a taste of quantum coming down the road

is kind of where I would take it. That's what IBM is. So you are technology agnostic in some sense? I'm product agnostic. Product agnostic. I'm not technology agnostic. Yes. But if I, 25 years from now, IBM could be doing things that would be unrecognizable to contemporary IBM?

It is completely possible. Yeah. It could be that in 25 years from now, the only software IBM does is open source. It could be that the only computing you do is quantum computers. And if I add those two, people today would say that's not the IBM of today. Is it even simpler to say you just, IBM solves problems at the, At the highest technical level? If you say highest technical level, yes.

Like the guy who invented the barcode. He was solving a problem. Retailers wanted to scale. Many of you may not know. It was an IBMer who invented the barcode. By the way, not somebody who was a PhD, not somebody who was a deep researcher. I think it was actually a field engineer.

Oh, really? Yeah. And lasers were out, and you could use lasers to scan things, but they could be upside down, they could be muddy, they could be partly scraped off, and he came up with the idea of the barcode. Yeah.

And that changed inventory management forever. The world needs to know that IBM invented the barcode. You guys should do a better job publicizing this act. I am sure our CMO will listen to this podcast, and we'll get that idea. Tell me, you started at the Thomas Watson Research Center. What were you doing when you first started IBM?

I started in 1990, and that was the era in which computers and networking were beginning to converge. And for the first five years, I was actually building networks. So let's remember this was pre-laptops. Laptops came in 92 or 93, but it was clear to us that they were going to come, so portable computing. And I spent my first five years building what today you would call Wi-Fi. We used to have these debates, can we build it, it's got to be small enough.

I mean, like, it can't be more than 100 grams was kind of our thought, because if it's more than that, you're on a 3,000 gram laptop, why would anybody put this on? And the debate used to be, why would anybody want to walk around untethered? Won't you want to attach a big, thick cable into it and sit down because that was the thought. That's how terminals worked. And I spent five years having a lot of fun, building many iterations of those, and making progress on that. If I had a conversation with your 1990 self about what the next 30 years are going to look like, is it possible to reconstruct what were your predictions at that age

about where the company and where the industry was going? It was more about where technology was going to go, I would say, than where industry would go. I would have told you that networking and computers would fuse. In 1990, that was a weird thought that some researchers held by the late 90s that was obvious, that it became the internet, I would have told you that I believe that video streaming will be the primary way people will consume video. You would have said that in 1990. Absolutely.

Now that didn't take five years, that took 20, but it happened because you could do it technically except it was just too expensive and too cumbersome. And if you've been in technology, like in 1985, I would have told you the end of the Internet is old. Because when I went to grad school, every one of us had those days an Apple, Mac or Lisa on our desks. They were all connected by a network. We were happily sending email to people all around the country.

We were doing file transfers. So, okay, you had to be a little bit more aware of the technology and didn't have a browser. That took 10 years to get the browser. That took five years to be a business. But when you see the speed and the pace of technology. in usually 10 or 15 years, the cost point and the consumerization is at a scale that you couldn't imagine 10 years ago

until you've seen a few of those cycles. Wait, did you make the leap to... Sorry, this is fascinating. I'm curious, but how far did you take? That's a really fundamental thing to have gotten right in 1990. How far did you run with that idea? We were pretty convinced that what we used to think of as linear television or broadcast

would become digitized. That was a given. Two, with cable already the preponderance of how people got it, that if you put packet television over cable, then that becomes the way it will go. I fundamentally believed, actually way back 87, that on-demand movies would become the way people would consume movies. So those were all things that I could have predicted.

Now, I didn't personally work on all those. I mean, after networking, I moved on to doing other things. But those were easy to predict. If you had a conversation in those years with someone in the television industry and you gave them those predictions, did they see it? Were they convinced of this? I'm actually going to take it back to wireless networking. I think one of the reasons I do what I do today,

which is at the intersection of business and technology, is because of what I saw happened with, Wi-Fi. So you build these wireless networks. And then you say, hey, the market is going to be millions, tens of millions, billions of users. And the business looks at it and says, we think the market is confined to warehouse workers doing inventory. You can't look at them and say, why not people in their homes? Because they couldn't imagine outside how people bought things at that time. And so I became convinced that I can't just help invent it.

I got to think about now how do you market it, to whom do you market it, what are the routes, how do you make it easy enough? And that was probably, I mean, I'm making it simple now. That was probably a five to ten year evolution of myself in those days. You know what this reminds me of when the telephone is invented in the 1870s, it doesn't doesn't take off for 40 years because the people running a telephone business, they didn't want women using it because they were worried that women would gossip with their friends. They didn't understand that that's actually what telephone was, right?

But it's an exact parallel. Yes, it is. You see it again and again. What is the source of that blindness? So there's a gap, in other words, between the invention, the technological achievement, and the social understanding of the technology. Why is there such a gap? I think that the gap is fundamental and rule.

rooted in a lot of academic disciplines. So even channeling some of your work, though you don't intend it to be used that way, you can say a lot of things are data-driven. If it is data-driven, then by definition you're looking at history. If you're looking at history, that means you're looking at existing buying patterns. If you look at existing buying patterns, you forget all of those who have created massive value in time. all created markets, meaning they've all created new markets.

And I think that is why the world is fascinated with people like Steve Jobs, for example. He imagined a market that didn't exist. So I think that is the gap. And then if you can get the technology, the business acumen to scale a company, and that imagination of making a market is how you create, I think, massive value. You've got to get all three pieces going. It's not enough. In other words, you were thinking it's not enough to invent something new.

I need to make a business case for it simultaneously. And that's what gets you thinking along the path that leads you to this job? Oh, yeah, I would tell you. If you had met urban in 1994 and you had talked about the stock market or about a balance sheet, I would have looked at you like, okay, I got what those words are, I can parse them, I have no idea what they are, I have no intuition on what they are. I couldn't tell you why it's relevant or why it's not. But then you began to think, okay, why do companies get higher values?

Okay, that's the stock. What does that capture? If I have to spend working capital, then that's from the balance sheet. So you learn. I mean, I figure I'm willing to learn. I'm willing to read. However, the best way I read is, like, to go to your balance sheets, here you can read the book.

It's pretty damn dry. Much easier to go talk to a financial expert who's around the corner. And people are, if you're curious about what they do, they're really happy to share their expertise. And over time, you learn more and more, and they actually become part of your network within the company. And that's how you can both learn and evolve yourself and actually gain the extra skills.

You know, to be a successful business leader, do you have to unlearn or deviate from some of the things that made you a successful scientist? I actually believe the exact opposite. Yeah. But use what you're really, got at as a foundation, but don't make it the only thing you use. So then how do you add the other skills? And there's many ways. You can have people that you trust who help you add those

skills. You can gain some intuition, maybe not the depth of expertise. I want to be deeper on certain areas of electrical engineering than I'm ever going to be, let's say, in finance or marketing. But I want to be curious about those. I don't want to dismiss them. So you build on skills, and then you have to say, but I need a complete and holistic view, so I'm going to be a little deep, not very deep on all of those. And you've also got to learn to trust your intuition a little bit. Yeah. I forgot a question that I wanted to ask about the predictions of 1990 Arbent.

What did you get wrong? Oh, lots of things. I think that people were thinking that in those days, and it's not my phrase, but I'll come back to it. I think most people thought that the communication companies would turn out to be the winners of how networking got carried.

If you all think through the 90s of the investments that were being done, by let's not take the names of all of the telecom carriers, didn't turn out to be the case. Actually, I think that's a business model case. The reason is they all had in their heads that you can charge people by the minute.

Because they had been doing that already. Because they've been doing that for a hundred years. And in the end, the winners are networking were those who set flat price, 30 bucks a month or 50 bucks a month or whatever. And that was just too much of a leap for that. You think it's as simple? That is the most parsimonious explanation for why you think they fail? No, there were a couple of other more technical things.

One was written by somebody who was inside one of these telecom companies, and he labeled his article, The Rise of the Stupid Network. So telephone people believe that the network should be really smart. The end device is dumb. If you think about the telephone, telephone is dumb. It doesn't actually do anything. It doesn't a bunch of relays. And the network is smart.

It routes you. It figures out where to send it. It does echo cancellation. And the current internet is completely dumb of the inside. It just takes the bits and shoves them out the other end. All the intelligence is the computer at the end. That's probably a bit more of a fond explanation. But business model didn't help them either.

Yeah. Wait, did 1990 Arvind think that the network should be dumb or smart? I'm not sure I thought about it deeply, but everything I worked on, the network was dumb. The network moved bits. That's all it did. Yeah. Because even I, in those days, understood, I couldn't imagine all the applications. So if all you're doing this voice, maybe the network can be smart.

But if you're doing all those other things, how could the network possibly know all those things and be smart for it. If you, so you've been CEO for, what, five years? Five years. Wait, so in your five-year increment, what was your most misunderstood decision? Where you ended up being right, but everyone thought you were crazy? 2018, I proposed to our boat that we should buy a company called Red Hat. IBM does proprietary, but that was open source.

The stock fell 15% on the day we announced it, and today most people, people who turn around and say, this is the most successful acquisition that IBM has done in all time, and probably the most successful software acquisition in history. So it was completely misunderstood because people didn't see that you actually did need a platform that could make it agnostic across multiple cloud platforms, across on-premise environments, so you've got to have a view of what it could be, and we drove it to a place where I think today it stands as the leader in its space. So how did you come to believe this heretical notion?

So cloud was happening. You could ask yourself the question, should we spend a lot of capital and chase cloud? Okay? You're five years to be generous, maybe longer behind at that point the two leaders. So you could spend maybe $10 billion a year and a lot of businesses tend to do that.

Okay, it's so important, it's going to be half the market, I can't not. My view was, we'll always be five years behind. They're not dumb and they're not slow. So if you're going to be there, you're going to be best case, a distant third, worst case, maybe a fourth or a fifth because there's Chinese also in the mix. Why would you do that? Instead, is there a different space you can occupy? Instead of competing with them, can you become their best partner?

in which case you write their success. If I want to be their best partner, then what are the set of technologies that would be useful? So you kind of flip the problem is how I thought about it. How hard was it to convince people needed convincing before that acquisition? Probably six to nine months of breaking my head with no success. And then six months of building. the momentum once a couple of people began to see it.

Yeah. You're very persistent. Oh, yes, very. Would you describe that as your defining trait? I am very persistent and I'm very patient. I'm also probably very impatient, but I'm not a yeller and screamer. I don't rant and rave. But as I say, if I think we're going to do something,

I can be remarkably stubborn about it. We will do it. If I got your family, put them up on stage and ask them this exact question. Is this how they would answer as well? They would tell you I'm very stubborn. They might not agree that I don't rant and rave. Well, you know, one of the principal observations of psychology

is that our home self and our work self are uncorrelated. Once you know that, you know everything. Wait, I'm curious. One last question about that. How long does it take for you to be vindicated with Red Hat? Probably it took five, maybe four years. I think by 2023, so 2018, we announced it. We took the big stock draft. It took a year to close, 2019. So if I count, not that I'm counting that much, but July 9th, 2019. as the day that we got all the approvals, it took another few weeks to actually transfer the money. But from there, probably 2023, the world woke up and said,

hey, you guys deserve credit for this. This was actually a great move, not a bad move. Yeah. But this is interesting because this is a real gamble. If it doesn't work, you're not sitting in this chair right now, right? Oh, for sure. There were two steps. One, if it was obviously not going to work,

I wouldn't have been selected. And two, if it hadn't worked after that, that's why CEOs can be short-lived. Can I ask you a sort of a personal question? How much sleep did you lose over this? Once we had made the decision, none. Can you give me pointers on how you do this? Because I wake up at 2 a.m. every morning, and over much more trivial things than this. Once a week, I'll probably wake up at 2 or 3 in the morning.

I acknowledge it because I wake up and my brain is running. And once it's running, I don't even try to go back to sleep. I mean, I pay it's there. Go get up and do work and make yourself productive. You're going to be tired by 4 in the afternoon. That's fine. You'll sleep well that night. I have actually learned a long time back.

You can't do it across. You can't do it early morning, through the day and late at night. So an hour before I think I want to go to bed, I will actually change what I'm doing. Meaning, I will start reading. something interesting to me but completely outside the scope of work. I may read a biography.

I might read somebody who's quantificating on demographics and population but I won't read it on leadership because that's too close now. 20 years ago, I might have it. That would have been different. I won't read it on deep science

because that's too close to what we do for a living. So it's got to be outside the things that will make my brain churn about work. But it's got to be something that is dense enough to occupy your brain, so it shift gears. Sorry, I want to dwell on this just for a moment, the red hat thing. Was there someone or is there someone who you went to and explained the logic of this?

And they saw the logic of this, and that meant a big difference to you? Getting their support made a big difference. You'd be surprised. I'm remarkably open inside. I mean, when I have, are there probably a half-dozen to a dozen people inside that I'll talk to and I'll be completely open about, hey, this is what I'm thinking, I don't know, here are the risks. I'm open about those also. It's not just the benefits.

I think these are the risks, but I think the benefits outweigh the risks. I talk about that to people all the time. So whether, for example, I mean, I'll take names. I think our current CHRO, Nicol who introduced us, she has been in that loop since at least 2015 for me.

If I look at our CFO, Jim Kavanaugh, he's been in that loop probably since 2013 and the IBMers will probably wonder what the hell intersection did you guys have?

It didn't. When I talked about learning finance, I would go to him and say, hey, explain this to me. I don't understand why it's like this. And to me, it's okay.

You're patient, you go learn. If I think about many of the people in the software business, they've been having these discussions with me for always. I mean, now I'll acknowledge I can get probably impatient and acerbic, but it's meant to be a discussion. I mean, like, let's have the discussion. If you have a strong point of view, I got it. Nobody has a, going to be perfectly correct. But I always look for, if you have a strong point of view, that means it's from a different perspective than mine.

So what do I learn from that is the question, which has. helps to improve my point of view. That makes sense. I actually think that each person should try to build a community of 100 people inside your enterprise and 100 outside that you can call up. I have no hesitation. Somebody introduced me to a long time back to a CEO on the outside. I call them up all the time and say, hey, do you have five minutes?

I'm just thinking about something this way. The CEO of Red Hat who left IBM in 2021. We probably talk every two or three months. on a random topic. By the way, it becomes mutual. He'll ask me my opinion on some things. Now, by the way, three or four times, he might do something different,

but he wants my opinion. The other way around. If I gave you my phone number, can I be on that list? I would just be fascinating. I don't know if I can help you, but it would be really fun to get the call. Sure, you can.

Do you think that we can ever succeed unless people who influence opinions say things about us? So you may not think deeply about maybe the physical, of quantum computing, but would you think deeply about why and what moment may make it much more attractive to a large audience? Sure you would.

You'd be far better as a thinker on that topic and probably most of the people. I was thinking, you know, when you were making your comments about your 1990 self and streaming, that the rational thing would have been for there had been a reserved board seat for every, for every television, television network from someone from the world of technology, which I'm 100% sure they did not have that in 1990. But they, their board was probably composed of people like them. Let's talk a little bit about technology now. There's so much, so much of the changes going on right now are accompanied by a great deal

of hype. What are we overestimating? What are we underestimating? Look, let's go back to 1995, the internet, because I think that the current moment is very much like the internet moment. Actually, all the moments in the middle were much smaller. I think mobile streaming were much smaller. Internet was the major moment.

If you remember back to 99 and 2000, people claimed there was a lot of hype. Would we say that the Internet of today has more than fulfilled all those expectations and more? Yes. Along the way, did 8 out of 10 of the companies that were invested in heavily go bankrupt? Yes. I actually think of that as being the huge positive. of the United States capital system. That investment happened, eight out of ten went broke.

By the way, those assets didn't go away. They got consumed at 10 cents to the dollar by somebody else. Who could then make a lot of money? But the two out of ten, just take two. It probably has paid for all the capital. If you just take Amazon and Alphabet, aka Google, just those two have probably paid for all the capital of that time. So that's what's going to happen this time.

There will be a lot of tears, but in aggregate, there will be a lot of success. And I think that's a fundamental difference between the U.S. model and almost all other countries. On all other countries, they're desperate to keep all the companies alive. So that means you're diluting. That's a horrible thing. So to me, let the system work. It's worked really effectively. By the way, not just now.

I mean, all the way back to railways and electrification and you mentioned telephone system. you can keep going on, oil, I mean, consumer goods, it goes on and on. I think this system is very effective. It deploys capital. It senses this is a big market. It's completely willing to over-deploy capital in the short term, not the long term. That results in more competition. So it actually improves the rate of innovation.

That means what might have taken 20 years takes five, and the winners emerge. Exactly the same is going to happen this time. Yeah. I saw that. I grew up in Waterloo, and BlackBerry, of course, was from Waterloo. Yep. Everyone used to work for BlackBerry. Yep. Blackberry goes into its dive, and that's the best thing that happened to Waterloo. Because it was not just capital, but talent. Yep. Talent just went to... So many other companies.

So all these smart people went on, did the next really more interesting thing. Yeah. But wait, you haven't answered... So what is you an idea that we are underestimating at the moment that's in the current kind of suite of... innovations? So I don't think AI is being underestimated because when you look at the amount of capital and the amount of things chasing it, I think it's incredible. I do think that a lot of enterprises are deploying it in the wrong place. They're running after shiny experiments.

There's a lot of basic things you can do to use AI to improve the business today. So that's what would be just my one advice to them. Pick areas you can scale. Don't pick the shiny little toys on the side. Then I think... For example, then, if anybody has more than 10% of what they had for customer service 10 years ago, they're already five years behind. If anybody is not using AI to make their developers who write software 30% more productive today, with the goal of being 70% more productive, that's not to say you'll lead less. you'll just get more software done, then they're not.

And I would turn around and tell you, I think only maybe 5% of the enterprises are on both those metrics today. Yeah, wow, yeah. And the one that is completely underestimated, I kind of put it like this. Quantum today is where GPUs and AI were in 2015. And I bet you every AI person is thinking and hoping,

I wish I had started doing more in 2015, as opposed to wait it until 2022. Quantum today is there. So it's not good enough that you can get a big advantage. But if you learn how to use it, then in five years, you'll be ready to exploit what comes. Yeah, we're going to get to quantum in a moment,

but I have a couple other AI questions. You know, I, as you know, where this conversation is part of this thing that we do with IBM Smart Talks, and the last episode I did was on Kenya, which has a massive default, forestation problem. And they got together, IBM took all the NASA satellite data, ran it through an LLM, and gave them this incredibly precise 10 meter by 10 meter analysis of what trees to plant, where to plant them exactly where the, you know, an astonishing kind of blueprint about how to fix their country ecologically. And it made me think, when we analyzed the potential of AI, are we making a mistake by spending too much thinking about the developed world? when it's actually the developing world

where the greatest ROI for this is? To me, look, software technologies are wonderful in the sense they can scale and they can be an and. So you don't have to do one or the other. You use deforestation. How about the use of pesticides and fertilizers? We overuse it. We tend to, for irrigation,

we tend to just flood everything as opposed to say, okay, only that one needs it. You could do all those things together ten times effectiveness and that all would apply to the development. the developing world. How about remote health care, or telehealth using an AI agent? So the examples are numerous.

In the developed world, I believe we are running out of people. I know that nobody likes to hear it. Most of the Far East is going to have half the number of people by 2070 compared to today. That's not that far away. If I look at Europe,

both rates are far under sustaining or keeping population flat. I mean, the U.S., also, depending on which number you want to look at, is either 1.6 births per women or 2 or 2.1, why are the three numbers? 1.6 is to women who are born in the United States. It becomes 2.0 if you include immigrant women. It becomes 2.1 if you include children who are immigrating in. So you've got to decide where the trend is obvious.

This is going down. So AI in the developed world is going to be essential because to keep our current. quality of life. You need more work done. Well, what's going to do the work if there aren't people to do the work? So the problems are different in the two places. Yeah. It gives you, in the developing world, you get access to a suite of technologies and things at a price you could never have been able to afford. Correct. That was my, in talking to the Kenyan thing, it was like the whole, it's this, it's maybe one of the largest ecological projects of its kind of 15 billion trees they want a plan. And that is one country that might get it done because they do take a lot of pride in

their ecology and in sort of returning to the land and giving back. Yeah. Yeah. What's different about IBM's version of AI versus some of your... So we are not a consumer company. So we have no focus on a B to C chatbot. And the reason I say that is, if you're making a B2C chatbot, Does it help you to make it even bigger and more computationally inefficient? And the short answer is yes, because you have a certain number of users, and you kind of say, I kind of say this jokingly, if I add finish to French capabilities, I can probably add 5 million users. If I add writing a haiku, I might be able to add another 5 million users.

If I add writing an email in the voice of Steinbeck, I can probably add another 5 million users. Do all those things. If my goal is to help a company summarize the legal documents in English, that can be a model that's 100th the size, as effective, probably higher quality, but I don't need to go wide. So if you're focusing on the enterprise, that actually takes away the focus of having to go to extremely large models, which by definition are going to be computationally expensive, power hungry, and demand lots and lots of data. So I can turn and tell the enterprise, you don't need to worry about copyright issues, about all those because you can create on a much smaller amount of data.

And now, by the way, tuning it for you yourself is a weekend exercise. It's not a six-month on a big supercomputer cluster somewhere out there. That's one big difference of what we do. Second, we are very focused on helping those problems that can give people immediate benefit, where we have domain knowledge.

So our domain knowledge is, around operations, is around programming and coding, is around customer service, is around customer experience, logistics, procurement. Let's stay in the areas where we have a lot of expertise. And then three, we kind of apply it to ourselves. And so we are not asking our clients to be the first experiment on it. We say, you can leverage what we did. We are happy to bring out all our learnings, including what needs to change in

the process, because the biggest change is not technology, is getting people to accept that there's a different way to do things. Are there challenges to explaining what makes you different to potential customers? For sure. The shiny object is always attractive. Well, I can go and try chat GPT. Why don't you have your GPT version? Do you use chat GPT?

I have used it. I asked you a question recently, which I thought was really simple, and it made up about 10 people. Anyway, I had a bad experience. I actually think that that's a fundamental issue with all LLMs as they get larger. Yeah. Because you had to ask, what was the original insight that led to these? It was a reward function with intent.

So if it has learned by using a reward function, its reward function comes from giving an answer that satisfies you. So if it thinks that if it makes up an answer that will satisfy you, how will you stop it? Why do we think this is different than the clever college kid who doesn't know an answer but a bullshit's the way to an answer? It's exactly the same.

It's like the example of clever hands. Do you know that story? The horse that they thought could speak and all it was doing was pleasing its master? Yes, it is a little bit of clever hands. Yeah, it's like dogs kind of imitating and looking. Yeah. What would you identify as the most significant bottleneck

in the development? of AI? What's slowing us down right now? I am not convinced that LLMs is the way to get much beyond where we'll get incremental improvements, but I, for one, don't believe that LLMs are going to get us to super intelligence or AGI. So I'll park that on the side and simply say, we have to find a way to fuse knowledge And how do you represent knowledge as opposed to have to statistically rediscover it each time I ask a question? And how do we fuse knowledge with LLM?

Maybe then we'll get to leaps and beyond today. On LLMs alone, my view is I think we can get a thousand X efficiency in power and cost and compute from today. So if you make something a thousand times cheaper, would people use a lot more? of it. Yes. And I think those answers lie as is usually in compute through advances in semiconductors, advances in software and advances in algorithmic techniques, all three. But how come we're not working in any of those three?

We're just taking the current semiconductor and going more. We're taking the current algorithmic techniques and not really trying to invent new ones. So I think those are all happened, less than five years. But why, you see there's a, there's a, you see, there's a, the, you know, there's a, you know, We're in a moment where people are not pursuing the optimal strategy for exploiting this technology. Why? Because when you see a few people running really hard and they're willing to invest any amount of money, so efficiency is not the focus, people feel if we don't do the same, we'll get left behind.

So... Is this a case where there's too much money? Humans have never had four more, right? Ever. But is this a consequence of overinvestment in the field? Going back to my internet analogy, if two out of ten are going to succeed, how do you guarantee or how do you improve the odds that you are one of those two? So if you pause to say, I want to become more efficient, that's not the way to win.

So first you win, then you become efficient. Yeah, yeah. Let's talk about what is... I was told your favorite topic was quantum. It is. What? Before you even go any further, why is quantum your favorite topic? We only had two kinds of compute in the history.

So, 1945, I was to use that year for ENIAC, all the way until 2020, we had one kind of compute, classical, what today you would call a classical computer. Then GPUs and AI came around. So you would say the intuition there is you went from sort of bits, which is algebra or high school algebra to including neurons which is captured in linear algebra, but that give you a different kind. But it can do problems that are really hard to do. I don't say impossible, just hard to do on normal computers.

Quantum adds a third kind of math. Yes, the physics properties which really get people energized and their imagination going and we use all these words about entanglement and superposition. But maybe because I'm a bit of a math guy, the real thing is it does a third kind of math to make it really simple. A third kind of math that comes from the field of abstract algebra. It does the math. You can use Hamiltonians for those who like physics, or you can use the word Lee algebra

for those who like abstract mathematics. If you can do a third kind of math, which algorithms are suited to that third kind of math? So, it excites me because we can now approach algorithms that you just could never do on the other two. It's impossible. Now, it's different than AI. It's not data-intensive, it's compute-intensive. So we kind of had compute and supercomputers, then we went to data, which is AI.

And now if you say there's another class of problems that require lots of compute, that's quantum. A couple months ago was at the Watson Research Center and they have, you know, on the ground floor, they have those behind the glass, those incredibly exciting-looking machines. But where are we in the timeline of this? Three to five years away from shocking people. What does shocking people mean? Do something that nobody thought was possible in that timeline.

Does an example come to mind? I was actually pleasantly surprised. So one of our clients, HSBC, last week published a result that using a quantum computer, bond trading was 34% more accurate. than their prior technique. 34%. This is an industry that's used to 1%. Correct.

Or 0.5%. Yes. That's astonishing. Now, that was not at a scale where they could turn it into production today. But that was sort of their original thought experiment and that's what they did. Now, can you imagine when will somebody...

So you were correct. You talk about an industry where one basis If I remember, I may be wrong, like $13 trillion of money kind of moves around in the financial industry each day, right? So a basis point would be $13 billion, something like that, right? One over $10,000. So when you think about the kind of profit that people can make, if you can tell somebody that you can come up with a better price than your competition, by just one basis point.

they would actually gain the entire market share. Yeah. So I think something around there, or something in the world of materials, can we make a better battery? Could we make a solid-state battery? Mm-hmm. Which means your risk of fires, heating decrease dramatically.

And the reason, sorry, to ask a really naive question, why is it that a quantum computer would be better at solving a battery problem than our existing methods of computing? of computing. So the equations of quantum mechanics and chemistry and how things interact are well known. To solve them, there are no known techniques. These are not like closed form, you know, it's not like the square root of a quadratic equation. So the only way to solve them is to explode the state space.

So if you have a few hundred electrons, you need two to the hundred states, I'm sorry, you don't have that much memory. It's impossible. it takes a really, really long time on a normal computer to solve those problems. Right? But that's a simpler problem. If a quantum computer operates in the equation domain, it doesn't need to explore the state space. It can actually solve it.

That's why I called it a different kind of math. That's the kind of math it does. So in a couple of seconds, it can tell you this is how that material will behave. Oh, I see. So you've taken what could take years to a few seconds. Yeah. That's a pretty big change. Yeah.

Yeah. It's speaking a different language. Speaking a different language. So any kind of problem that comes along that's specific to that language. Correct. Which is not all problems. Yeah. That's what I call it.

It's one more kind of math. Yeah. What's an example? So, so many questions. Give me another example of a kind of problem that a quantum computer would love. This one is a bit more speculative, but, and I'm going to use a little bit more. to use a little bit of poetic license. So let's take a post office in a mid-sized country.

They probably burn a billion gallons of fuel per year delivering packages and letters because most postal in advanced country says every house, every address each day. The way to optimize this is we can formulate the problem. It's called the traveling sales problem. Solving it is really hard. So people have heuristics. Let's suppose today our heuristics get us to witness. 20% of the optimal answer.

Let's suppose a quantum computer can get you the next 10%. Well, if I can get 10% of a billion gallons, that I think is 100 million gallons of a math is right. And in the country, I'm thinking about that could be 800 million pounds of saving to one entity in one year. And the associated carbon footprint, climate change, wear and tear, less mileage on vehicle, I'm not even counting all that. These are pretty attractive problems to go after.

So if I look at the interest recently, New York has started a whole program in some places. Illinois stood up a quantum algorithm center, which being a number of the universities, the governor there was heavily behind it, et cetera. So I wouldn't say that this is widespread. This is why I'm saying three to four years for that moment. But there's enough people who are deeply cognizant who are saying, wait a moment. We kind of get it.

This is a new kind of bad. What are the new problems we can solve? And the fact that we have about roughly 200 clients who work with us, very early stage, small experiments, is because their intuition is I can do something here that I couldn't do in other places. Three to four years is not a long time. No. But if I'm in a battery business and I don't have a line out to a quantum computing experiment, I have a problem.

Don't I have a problem? Yeah, you'll probably be out of business in 10 years. Or maybe you could write a big check and buy the technology from somebody else. Where does quantum rank in the kind of great inventions of the last 150 years? Equal to semiconductor. And I think that if semiconductors vanished, modern life would stop. Like just stop. Yeah.

no electricity, no automobile, no streaming. You can imagine the yells from all the kids who ever hear that, no streaming. And is that, it's funny because don't, as someone who's outside this world, I feel like quantum is under-discussed relative to its potential for transforming society. Because I use my internet example.

95 was the moment with Netscape, that internet came on people's consciousness. I said, in 85, I considered it to be this is a solved problem because it needs something that makes it accessible, easy. That was the browser. The Netscape browser is what brought it easy to understand. We have probably, as I said, about four to five years from that moment. That's why it's under-discussed because the moment I say,

in you've got a math, I probably lost 99% of the audience. If I go to quantum mechanics, I've probably lost 99% of the audience. So you, as CEO for the last five years, have been really the birth mother for a lot of the quantum computing work. I'm curious. So you come in, when you started as CEO, was this your first priority? I had already started investing in it back in 2015 when I was leading IBM research. So let me acknowledge And like nobody should try to copy it

I've had a I'll call it a weird career I was a researcher At some point if you had asked me I said I'm one of those people You know throw pizza under a door And like leave me alone I don't want to talk to people

Then I decided I was interested in the business Then I went and started acquiring companies And doing that Then somebody told me Hey why did you start doing some business strategy Then I went back to research And led our research division For a couple of years

And when the people described it to me, I asked some questions. So it wasn't a big investment at that time. It was, hey, can we make a computer not just a science experiment? Can it run by itself all night? Can you think about software so that even people are not deep in quantum mechanics can begin to use it? And they began to do those things. So over three, four years, did I get enough confidence? Yeah, okay, this is something that can really work.

And then you've got to nurture it to where it gets bigger and bigger until you get the confidence that, okay, now it's a big bet. And what was the moment when you realize now it's a big bet? Probably two or three years ago. And how do you decide, as the head of a company like this, how much money, how many resources and how many people and what kind of prominence to give to an idea like that? So three layers. the set of people who actually have the knowledge

and the intensity to fundamentally advance the technology if I could find more, I would hire them. So I'm constrained on people on that one. Because normally there's only so many people who can do these things. Two, you've got to be careful. If you push too hard on timing, you'll get people to take so much risk

that actually the thing will fail. So that's the art of between the leadership on the project and me to say, okay, how hard can you push, but not so hard that you cause it to fail because then they can compel to commit timelines that are just impossible. Yeah. How do you, this is fascinating, though.

So it's ultimately a question of judgment of trying to figure out what's the sweet spot between enough pressure to keep them ahead of the pack, but not too much pressure so that they start taking risks. How do you calibrate whether you're hitting that sweet spot? I mean, do you reassess every few months and say, I think I'm over-correcting or under-correcting at this moment? So, one, you got to have what I call,

and this is channeling a word from one of my favorite books, The Geek Way. How open can you be? So I want to press hard, but the team knows that they're allowed to push back and really argue back hard. That means you'll get to probably that correct Goldilocks pressure. Two, the people themselves should want to go as hard as possible, but not harder than possible. So that is then personality of leadership. Yeah.

Does that make sense? But you have to be someone who people feel comfortable being honest with. Yes. Absolutely. And people feel comfortable being honest with you. I believe so. Yeah. Has there been a moment in this path with quantum where you did think you were pushing too hard?

No. because I think that the leadership there will argue back with me any day of the week. I don't think that they feel that they have to ford. Do you drop by Saturday night at 10 p.m. to see if people are working? I tend to text people and ask questions and like I'll read something and say, hey, are these people doing this? And if they can answer me in reasonable terms, I have. actually then say, great, they're already watching the competition, they're watching the literature, they're watching the science. I don't need to push hard. If they're already ahead of it,

then me, I can answer my question. I'll say, thoughtfully, not always completely accurately. If you're thinking about it on the road, I don't need to push. Yeah. One last question I wanted to ask you. Do you have the most interesting job in America? I believe that it's the most interesting job, which I wouldn't give up for anything. Yeah. It does. It also sounds like you're enjoying yourself. I enjoy it as long as, look, my role and goal should be to make the enterprise thrive. As long as I'm making the enterprise thrive and our clients delighted, I love it.

If I don't, somebody else should do it. Arvin, this has been so much fun. Thank you so much for taking a time and a fascinating, completely fascinating conversation. I wish I was one of those people who could help you out with quantum, but I'm afraid I'm not. In a few years. Good. Thank you so much. Thank you so much.

is a production of Pushkin Industries and Ruby Studio at IHeartMedia. To find more Pushkin podcasts, listen on the IHeart Radio app, Apple Podcasts, or wherever you listen to podcasts. I'm Malcolm Gladwell. This is a paid advertisement from IBM. The conversations on this podcast don't necessarily represent IBM's positions, strategies, or opinions. Thank you.