WEBVTT - IBM Vice Chairman Gary Cohn Talks AI Finance 0:00:02.520 --> 0:00:07.880 Bloomberg Audio Studios, podcasts, radio news. Please to say that 0:00:07.920 --> 0:00:09.719 our next guest, he sits right at the intersection of 0:00:09.760 --> 0:00:13.680 macro policy, AI enterprise tech. Gary Cohen is the former 0:00:13.720 --> 0:00:16.400 president of Goldman Sachs, former director of the US National 0:00:16.440 --> 0:00:19.160 Economic Council, and he's been vice chair of computing giant 0:00:19.200 --> 0:00:21.040 IBM since twenty twenty one. 0:00:21.040 --> 0:00:22.920 He joins US right now. Very great to have you here, 0:00:23.120 --> 0:00:23.960 Thanks for having me. 0:00:25.320 --> 0:00:27.160 I don't even know where to start with it because 0:00:27.360 --> 0:00:29.640 you cover everything, but I actually do want to start 0:00:29.680 --> 0:00:31.720 first off, because we haven't talked a lot about technology. 0:00:31.720 --> 0:00:33.680 There are a lot of disruptive forces going on from 0:00:33.760 --> 0:00:35.879 what's going on in the Middle East, and obviously we 0:00:35.920 --> 0:00:38.320 talk a lot about private capital, but AI, I think, 0:00:38.320 --> 0:00:41.040 by anyone's measure, is going to be transformative, whether that's 0:00:41.080 --> 0:00:42.360 going to be good or bat. You're at IBM, one 0:00:42.400 --> 0:00:44.040 of the biggest computing companies in the world, and I 0:00:44.040 --> 0:00:46.000 do want to point out, aren't you having a conference 0:00:46.360 --> 0:00:47.320 right now in Boston? 0:00:47.360 --> 0:00:49.920 When you're here in La your IBM think conference, this 0:00:49.960 --> 0:00:50.760 is your biggest conference. 0:00:50.760 --> 0:00:55.040 Why are you here because you're here, No, it's like 0:00:55.600 --> 0:00:58.440 we do have our largest conference in the year going 0:00:58.440 --> 0:01:01.400 on right now. I think conference in Boston. As you said, 0:01:01.880 --> 0:01:05.120 thousands and thousands of our clients. They're rolling at a 0:01:05.160 --> 0:01:08.600 lot of our new technology showcasing our new technology. But look, 0:01:08.600 --> 0:01:10.319 we're a global company and a lot of our most 0:01:10.360 --> 0:01:13.600 important clients are here as well. So I made the 0:01:13.680 --> 0:01:15.840 decision to be here. The rest of the IBM senior 0:01:15.920 --> 0:01:18.319 leadership team is in Boston right now, and some of 0:01:18.360 --> 0:01:20.160 my team is actually going out tonight to get to 0:01:20.200 --> 0:01:20.959 Boston for tomorrow. 0:01:21.040 --> 0:01:22.319 Well, that's what I'm curious about too. 0:01:22.400 --> 0:01:25.760 I mean, obviously there was a very market decision to 0:01:25.800 --> 0:01:28.240 be here, and I do wonder about when we talk 0:01:28.240 --> 0:01:30.360 about this AI transformation. We talk a lot about it 0:01:30.400 --> 0:01:33.559 from technology, but this is becoming a finance story as well. 0:01:33.720 --> 0:01:36.280 I mean, you need capital to make good on the 0:01:36.319 --> 0:01:39.240 promise of what this is. Where does IBM actually fit 0:01:39.400 --> 0:01:40.520 right now into all of that. 0:01:41.040 --> 0:01:43.560 Well, it's an interesting question. I'll leave the capital piece 0:01:43.600 --> 0:01:47.000 of the side. But when you talk about the AI 0:01:47.160 --> 0:01:50.240 architecture of the world today, because we are really redefining 0:01:50.440 --> 0:01:54.960 what AII architecture is, it's hybrid cloud, multi cloud all 0:01:55.000 --> 0:01:58.440 over the world. It still has on premise computing, and 0:01:58.600 --> 0:02:01.400 it's data and data manner. You know, we talk about 0:02:01.440 --> 0:02:04.960 AI all day long, but really what is AI. It's 0:02:05.000 --> 0:02:08.080 a system that goes out and finds data and does 0:02:08.120 --> 0:02:10.200 the research and does the computation. 0:02:09.760 --> 0:02:10.520 On the data. 0:02:10.600 --> 0:02:12.440 So at the end of the day, being able to 0:02:12.480 --> 0:02:14.600 manage the data wherever your data is. 0:02:14.639 --> 0:02:15.280 And that's hard. 0:02:15.320 --> 0:02:18.120 Sometimes you have very organized data, sometimes you have very 0:02:18.120 --> 0:02:21.000 disorganized data, but being able to pull from that data 0:02:21.080 --> 0:02:22.919 real time to make sure you get the right answer. 0:02:23.240 --> 0:02:25.800 We're sitting in the middle of the data management business 0:02:25.800 --> 0:02:28.200 as well as the on premise computing as well as 0:02:28.200 --> 0:02:30.480 the hybrid cloud structure, and those are we think those 0:02:30.480 --> 0:02:32.919 are three very important components of where the world's going. 0:02:33.160 --> 0:02:35.800 Well, I think about your time at IBM, as Romain mentioned, 0:02:35.840 --> 0:02:38.280 I mean you joined in twenty twenty one, and you 0:02:38.320 --> 0:02:41.959 think back on the conversation around AI in twenty twenty one. 0:02:42.320 --> 0:02:44.959 I mean, chat, GPT wasn't really on the scene yet. 0:02:45.000 --> 0:02:48.320 It wasn't, you know, so ubiquitous in modern life the 0:02:48.360 --> 0:02:50.280 way it is right now. And I just wonder, you know, 0:02:50.320 --> 0:02:53.360 how the shape of those conversations have changed. When you 0:02:53.360 --> 0:02:54.600 think about the last five. 0:02:54.440 --> 0:02:56.640 Years, it's changed dramatically. 0:02:56.680 --> 0:02:59.040 So when I got the IBM, as you said, five 0:02:59.080 --> 0:03:01.760 and a half years ago, everyone in the world was 0:03:01.840 --> 0:03:04.120 convinced that the world was going to cloud. Everything was 0:03:04.160 --> 0:03:06.720 going to be in a cloud, and every model was 0:03:06.760 --> 0:03:08.040 going to be a SaaS model. You were going to 0:03:08.080 --> 0:03:10.919 basically pay for consumption. Well, guess what, it didn't really 0:03:10.919 --> 0:03:12.760 work out that way. A lot of things went to 0:03:12.800 --> 0:03:14.799 the cloud and people found out the cloud it has 0:03:15.080 --> 0:03:18.919 enormose amount advantages, but on premise computing, and the mainframe 0:03:19.360 --> 0:03:21.400 still is a vital, important. 0:03:20.960 --> 0:03:21.680 Piece of quittent. 0:03:21.680 --> 0:03:24.800 In the United States, over ninety percent of financial transactions 0:03:24.800 --> 0:03:27.840 in the United States go through an IBM mainframe every day, 0:03:27.880 --> 0:03:29.200 every minute of the day. 0:03:29.440 --> 0:03:31.200 So this whole move that. 0:03:31.120 --> 0:03:32.440 We're going to put everything in the cloud and the 0:03:32.440 --> 0:03:34.440 cloud is safe and the cloud is perfect. 0:03:34.680 --> 0:03:36.000 It's not worked out that way. 0:03:36.120 --> 0:03:38.760 Now we've got this hybrid architecture, so being in the 0:03:38.760 --> 0:03:40.640 cloud as part of it, being a hybrid cloud as 0:03:40.640 --> 0:03:42.520 part of it. Being in the mainframe is part of it, 0:03:42.640 --> 0:03:45.280 And that's the way the world's evolving here. Then you 0:03:45.320 --> 0:03:48.400 bring in these large language models, which are a whole 0:03:48.480 --> 0:03:51.200 new set of technology, and you're right. When I got 0:03:51.240 --> 0:03:54.000 the IBM, you know, the world of AI was it's 0:03:54.120 --> 0:03:57.160 kind of interesting, maybe we should look at it. You know, 0:03:57.280 --> 0:04:00.000 sometimes it hallucinates, sometimes it gives you a great answer. 0:04:00.280 --> 0:04:01.680 And I think we all sat up there and said, 0:04:02.400 --> 0:04:05.120 you'll probably get this worked out. Not not crazy, but 0:04:05.240 --> 0:04:07.400 I'll tell you I'll tell you something. Now we're at 0:04:07.440 --> 0:04:10.520 the exact same place with quantum computing today. We're in 0:04:10.520 --> 0:04:14.240 a world where quantum computing kind of works. Sometimes might 0:04:14.280 --> 0:04:16.120 not give you the right answer, but it's going to 0:04:16.160 --> 0:04:18.559 get better and better over time. So in my short 0:04:18.600 --> 0:04:21.200 period of IBM, I've seen this sort of ebb and 0:04:21.279 --> 0:04:24.040 flow of new technology come in and even go back 0:04:24.160 --> 0:04:26.839 earlier in my career. You know, when we brought in 0:04:26.880 --> 0:04:29.279 new technology. You tend to see these ebbs and flows 0:04:29.640 --> 0:04:31.600 and what you think at the beginning of a cycle, 0:04:31.920 --> 0:04:33.440 it's not what the end of the cycle is going. 0:04:33.400 --> 0:04:33.800 To look like. 0:04:34.000 --> 0:04:36.039 Well, so your point on quantum computing, I mean, we 0:04:36.120 --> 0:04:38.919 typically talk about it in the context of, you know, 0:04:39.000 --> 0:04:41.520 this is being presented as the next AI, that's how 0:04:41.520 --> 0:04:44.320 disruptive it's going to be. But do you think that 0:04:44.600 --> 0:04:47.840 you know corporate America, that you know, America at large 0:04:47.880 --> 0:04:51.000 is prepared for that Because you think about how disruptive 0:04:51.080 --> 0:04:54.359 AI has been, you could make the case that a 0:04:54.360 --> 0:04:56.279 lot of folks were not positioned for that. 0:04:57.120 --> 0:05:00.080 So the world is not prepared for quantum. But the 0:05:00.120 --> 0:05:02.120 world was not prepared for AI three years ago. That's 0:05:02.120 --> 0:05:04.320 why I think the analogy is interesting. Three years ago 0:05:04.360 --> 0:05:06.160 people said, like AI, I don't know if we'll ever 0:05:06.200 --> 0:05:06.640 affect me. 0:05:06.680 --> 0:05:07.840 I don't know if I'll ever use it. 0:05:08.320 --> 0:05:11.160 No one would have ever talked about how many llms 0:05:11.440 --> 0:05:13.599 are downloaded per day, per hour, per second. 0:05:13.960 --> 0:05:16.120 So quantum is in a very similar place. 0:05:16.160 --> 0:05:19.200 I think today people are saying, okay, interesting, not sure 0:05:19.200 --> 0:05:21.760 what it is, not sure what it does, not sure 0:05:21.760 --> 0:05:24.600 it will affect my life. But I said, we've seen 0:05:24.640 --> 0:05:28.279 this movie. Quantum is going to be real. It is 0:05:28.360 --> 0:05:31.200 going to affect people's lives. It may not affect people's 0:05:31.200 --> 0:05:34.880 lives in a real time consumptive pattern the way people are. 0:05:34.839 --> 0:05:36.760 Using large language models today, but. 0:05:36.760 --> 0:05:39.280 It will have dramatic effect in the healthcare system, will 0:05:39.279 --> 0:05:41.720 have dramatic effect in the risk management system and pricing 0:05:41.760 --> 0:05:45.280 for banks. It'll have portfolio optimization in some of the 0:05:45.640 --> 0:05:48.880 chemicals and science and physical science world. What can go 0:05:48.960 --> 0:05:52.600 on in the quantum model, It's just nothing short of extraordinary. 0:05:52.720 --> 0:05:54.840 But we're still getting there. I don't want anyone to 0:05:54.920 --> 0:05:57.960 think we're there today. We're evolving, we're going through error 0:05:58.040 --> 0:06:00.760 correction rate, and the machines get better better every day. 0:06:00.880 --> 0:06:03.080 We've got about eighty five systems out in the world 0:06:03.080 --> 0:06:06.360 today with over three hundred clients using our quantum system. 0:06:06.400 --> 0:06:08.120 And it's you know, it's it's people. You would think, 0:06:08.160 --> 0:06:10.880 it's the Boeing Airlines, it's the JP Mortganes, it's the 0:06:10.880 --> 0:06:14.719 Cleveland Clinic looking at human genomes. So all different parts 0:06:14.720 --> 0:06:18.200 of the ecosystem are in the quantum world, playing around 0:06:18.200 --> 0:06:21.720 with it, experimenting with it, getting really interesting data out 0:06:21.720 --> 0:06:21.920 of it. 0:06:22.040 --> 0:06:23.279 There's a lot of upside to this. 0:06:23.400 --> 0:06:26.080 But as you know, as somebody obviously steeped in economics, 0:06:26.120 --> 0:06:27.880 there's still a lot of concern about the potential for 0:06:27.960 --> 0:06:32.279 job losses, certainly with AI and the potential productivity improvements 0:06:32.279 --> 0:06:34.280 that come with that. How do you balance that out? 0:06:34.279 --> 0:06:36.240 I mean, if you put your nec hat back on here, 0:06:36.400 --> 0:06:38.119 I mean, how do you think about that and making 0:06:38.160 --> 0:06:41.920 sure that whatever benefits a crue to the society overall, 0:06:42.360 --> 0:06:43.760 that it is going to be a rising tide. 0:06:43.839 --> 0:06:46.320 Let's everyone, this is not a unique discussion. 0:06:46.760 --> 0:06:50.120 Every time we've gone through a seismic technological invention in 0:06:50.240 --> 0:06:53.720 the world or in the country. We have obsessed about 0:06:54.120 --> 0:06:57.640 job loss, and by the way, we will see job 0:06:57.720 --> 0:06:59.200 loss in jobs. 0:06:58.880 --> 0:07:01.040 That are employing people today. 0:07:01.400 --> 0:07:03.960 But what we have seen historically, and I don't think 0:07:04.000 --> 0:07:06.400 this time will be different, what we see is massive 0:07:06.440 --> 0:07:10.880 productivity gains. When you see massive productivity gains, the economy grows. 0:07:11.480 --> 0:07:14.240 Is the economy grows, we still need to put people 0:07:14.280 --> 0:07:16.080 to work in different types of jobs. 0:07:16.280 --> 0:07:16.960 So you can go. 0:07:17.000 --> 0:07:19.680 Back in history and look at all the technological advancements 0:07:19.840 --> 0:07:23.200 and you'll read stories that are horrifying, saying all these 0:07:23.200 --> 0:07:24.960 people are going to lose their jobs, Oh my god, 0:07:25.000 --> 0:07:27.920 what's going to happen, And miraculously the economy grows. 0:07:27.920 --> 0:07:28.720 People get jobs. 0:07:28.840 --> 0:07:32.160 Like today, we have a massive shorter of shortage of 0:07:32.200 --> 0:07:35.360 people in the trades. You know, if you're a person 0:07:35.400 --> 0:07:37.560 that can put down a cement floor and get a 0:07:37.640 --> 0:07:39.800 perfectly level for a data center, you're. 0:07:39.640 --> 0:07:41.560 Earning hundreds of thousands of dollars. 0:07:41.640 --> 0:07:43.560 If you're a person that can wire that data center, 0:07:43.640 --> 0:07:46.120 you're a person that can bring in the heating, ventilation, 0:07:46.160 --> 0:07:49.480 air conditioning, all air conditioning system. Those jobs are massive 0:07:49.560 --> 0:07:52.720 payers today and there's a huge shortage of people. So 0:07:52.760 --> 0:07:55.000 some of the people that are doing tasks that they 0:07:55.040 --> 0:07:58.200 may not love to do. Because the quantum I'm sorry, 0:07:58.240 --> 0:08:01.320 the AI machine is better at media tasks. They can 0:08:01.360 --> 0:08:04.080 go from potentially doing an unsatisfactory job to a very 0:08:04.120 --> 0:08:05.520 satisfactory job in the trade. 0:08:05.760 --> 0:08:07.640 What do you think we start to see those productivity 0:08:07.640 --> 0:08:10.600 improvements actually show up in the actual economic data. I mean, 0:08:10.640 --> 0:08:13.160 I think back to Kevin Warsha's a confirmation hearings where 0:08:13.200 --> 0:08:14.920 he kind of made the case how some of those 0:08:14.920 --> 0:08:17.840 productivity improvements, at least in his view, could actually set 0:08:17.880 --> 0:08:19.480 the stage for actually cutting rates. 0:08:19.840 --> 0:08:21.320 Do you think we're at that inflection point? 0:08:21.320 --> 0:08:23.480 Because as you know, I mean for decades, I mean, 0:08:23.480 --> 0:08:25.600 productivity effectively stalled in this country. 0:08:25.800 --> 0:08:28.080 Do you see this as being a true accelerator of that? 0:08:29.080 --> 0:08:30.800 AI is a true accelerator. 0:08:31.680 --> 0:08:34.880 You don't see the productivity gains day by day. You're 0:08:34.920 --> 0:08:36.560 going to have to measure them over a quarter or 0:08:36.640 --> 0:08:40.360 month or year by year because companies are adapting. You know, 0:08:40.600 --> 0:08:44.320 we're sort of going from the interesting science experiment of 0:08:44.400 --> 0:08:48.040 AI to the economic value. We're in that transition from 0:08:48.080 --> 0:08:51.760 science to economic value. We at IBM just to be honest, 0:08:51.960 --> 0:08:55.640 we have used ourselves as client zero, meaning all the 0:08:55.760 --> 0:08:58.360 AI tools that we're out selling to our clients, we 0:08:58.400 --> 0:08:59.800 have implemented AM on ourselves. 0:09:00.200 --> 0:09:02.960 We have saved over four and a half. 0:09:02.720 --> 0:09:05.720 Billion dollars of expenses and our head count has not 0:09:05.760 --> 0:09:08.079 gone down. So we have been able to take people 0:09:08.080 --> 0:09:10.079 out of jobs they didn't like doing move them into 0:09:10.160 --> 0:09:13.319 jobs with much higher value added to them. So take 0:09:13.360 --> 0:09:16.480 someone in an HR department. You know historically you're buying 0:09:16.520 --> 0:09:18.000 a house, so you're renting an A department. 0:09:18.160 --> 0:09:19.160 You need a reference letter. 0:09:19.240 --> 0:09:20.880 You need to know how long you've worked at, what's 0:09:20.880 --> 0:09:24.480 your jobs, how much you've heard before AI. You would 0:09:24.480 --> 0:09:26.480 call up somebody, you would get them on the phone. 0:09:26.480 --> 0:09:28.160 You'd say, hey, this is so and so I need 0:09:28.200 --> 0:09:29.840 you to send a letter to so and so. They 0:09:29.840 --> 0:09:31.360 would have to go pull your data out of your 0:09:31.360 --> 0:09:33.360 file and send out the letter, and you'd hope and 0:09:33.400 --> 0:09:35.960 pray they do that quickly. Now you go to what's 0:09:35.960 --> 0:09:38.880 an HR bought. You put in your information, you tell 0:09:38.920 --> 0:09:40.800 them the sent to as many places you want, and 0:09:40.840 --> 0:09:43.280 after you hit sent, it's going back out the people 0:09:43.360 --> 0:09:47.559 doing that work that's not highly satisfying work. We've taken 0:09:47.559 --> 0:09:49.640 people that were doing that work in HR and now 0:09:49.679 --> 0:09:52.520 we have them out recruiting new talent to the firm, 0:09:52.520 --> 0:09:54.960 we have out mentoring to people to the firm, So 0:09:55.000 --> 0:09:57.200 we're getting much higher productivity out of that. 0:09:57.760 --> 0:10:00.600 Right, I do want to bring this into the context 0:10:00.640 --> 0:10:02.720 of the economy of what we're talking about with AI, 0:10:02.920 --> 0:10:05.640 because certainly that has been a big benefit to the 0:10:05.760 --> 0:10:08.800 US stock market, And there is the idea out there 0:10:08.840 --> 0:10:10.839 that you think about what's going on when it comes 0:10:10.840 --> 0:10:15.680 to energy prices right now really impacting lower income consumers, 0:10:15.720 --> 0:10:18.360 and then you have the wealth effect of AI really 0:10:18.360 --> 0:10:22.720 translating into higher income consumers who are invested heavily in assets. 0:10:22.720 --> 0:10:24.600 And the general idea is that we're going to see 0:10:24.640 --> 0:10:28.280 this K shape economy basically be exacerbated. Gary, and I 0:10:28.320 --> 0:10:31.120 wonder you know where you fall on that sort of logic. 0:10:31.480 --> 0:10:34.679 Well, look, we have a fairly massive wealth effect going on. 0:10:35.200 --> 0:10:37.280 You know, I look at the economy today is asset 0:10:37.280 --> 0:10:39.760 owners and non asset owners. You know, if you own 0:10:39.840 --> 0:10:43.760 assets today, your assets probably appreciate. Whether it's whether it's 0:10:43.760 --> 0:10:47.240 a house, it's a car, it's an investable asset. Most 0:10:47.280 --> 0:10:50.800 of these underlying assets are going up in value versus 0:10:50.960 --> 0:10:53.240 the other side of the k, which is a which 0:10:53.280 --> 0:10:56.120 is a large predominance of our country who is really 0:10:56.200 --> 0:10:59.840 suffering and really having a difficult time as their input 0:10:59.840 --> 0:11:04.320 car meaning their food cost, their energy costs, their insurance costs, 0:11:04.360 --> 0:11:06.960 their price of cars, price of used cars are all 0:11:07.000 --> 0:11:09.839 going up every day, and their wages are not keeping 0:11:09.920 --> 0:11:12.160 up with that. This is this is a problem. I mean, 0:11:12.240 --> 0:11:14.800 we can't sweep this under the rug. We have to 0:11:14.880 --> 0:11:17.560 understand that we've got a fundamental problem going on in 0:11:17.600 --> 0:11:20.960 the country, and look it's starting to resonate with more 0:11:21.000 --> 0:11:23.560 and more people. I think that AI is part of 0:11:23.600 --> 0:11:26.480 the solution. AI can train people. AI can move people 0:11:26.480 --> 0:11:29.120 in the higher paying jobs. Like I said, the jobs 0:11:29.120 --> 0:11:30.600 are out there. We just have to forget how to 0:11:30.600 --> 0:11:31.120 train people. 0:11:31.559 --> 0:11:34.000 All right, Gary, got to leave it there, so appreciate 0:11:34.000 --> 0:11:35.800 your time. I'm sure it's a busy conference for you, 0:11:35.880 --> 0:11:39.600 that is Gary Cohne. He is the vice chair of IBM.