Learning Library

← Back to Library

25 Years of IBM AI Evolution

Key Points

  • Rob reflects on joining IBM Consulting straight out of school, feeling unqualified to advise clients until he was thrust into a last‑minute Vio project, forcing him to self‑teach through intensive reading and rapid hands‑on learning.
  • He emphasizes that taking risks and learning faster than peers is essential in consulting, as much of the work involves figuring things out on the fly rather than following a predetermined plan.
  • The discussion shifts to AI, with Rob recalling the historical context of John McCarthy coining “artificial intelligence” in the 1950s and noting how IBM’s focus on AI has grown from a peripheral curiosity to a core strategic priority over the past 25 years.
  • Rob’s early experience illustrates the broader evolution at IBM, where consultants must continuously adapt to emerging technologies—now exemplified by AI—by rapidly acquiring new skills and applying them to client problems.

Full Transcript

# 25 Years of IBM AI Evolution **Source:** [https://www.youtube.com/watch?v=iaisOHeJLC4](https://www.youtube.com/watch?v=iaisOHeJLC4) **Duration:** 00:39:08 ## Summary - Rob reflects on joining IBM Consulting straight out of school, feeling unqualified to advise clients until he was thrust into a last‑minute Vio project, forcing him to self‑teach through intensive reading and rapid hands‑on learning. - He emphasizes that taking risks and learning faster than peers is essential in consulting, as much of the work involves figuring things out on the fly rather than following a predetermined plan. - The discussion shifts to AI, with Rob recalling the historical context of John McCarthy coining “artificial intelligence” in the 1950s and noting how IBM’s focus on AI has grown from a peripheral curiosity to a core strategic priority over the past 25 years. - Rob’s early experience illustrates the broader evolution at IBM, where consultants must continuously adapt to emerging technologies—now exemplified by AI—by rapidly acquiring new skills and applying them to client problems. ## Sections - [00:00:00](https://www.youtube.com/watch?v=iaisOHeJLC4&t=0s) **First-Day Consultant Reflections at IBM** - Rob recounts his early experience joining IBM Consulting, feeling unqualified and navigating unknown technologies while trying to prove his relevance as a new consultant. ## Full Transcript
0:00[Applause] 0:03[Music] 0:06how we doing good Rob this is our our 0:09second time we we did one of these in 0:12the middle of the P pandemic but not 0:14it's all such a blur neither of us can 0:16figure out when it was I know it's hard 0:18those are like a blurry years you don't 0:19know what happened right but um um well 0:22it's good to to see you to meet you 0:23again um I want I wanted to start by 0:26going back you've been in IBM 20 years 0:29is that right 25 in July believe it or 0:31not so you were you were a kid when you 0:33joined I was 0:35four so I I want to contrast present day 0:40Rob and 25 years ago Rob when you arrive 0:45at IBM what do you think your job is 0:48going to be at your career is what what 0:49do you think the kind of problems you're 0:51going to be addressing 0:53are well it's kind of surreal because I 0:56I joined IBM Consulting and I'm coming 0:58out of school 1:00and you quickly realize wait the job of 1:03a consultant is to tell other companies 1:05what to do and I was like I literally 1:07know 1:08nothing and so you're immediately trying 1:10to figure out so how am I going to be 1:11relevant given that I know absolutely 1:13nothing to advise other companies on 1:15what they should be doing and I remember 1:17it well like we were sitting in a room 1:21when you're a consultant you're waiting 1:22for somebody else to find work for you 1:24bunch of us sitting in a room and 1:26somebody walks in and says we need 1:29somebody that knows Vio does anybody 1:31know Vio I'd never heard of Vio I don't 1:33know if anybody in the room has so 1:36everybody's like sitting around looking 1:38at their shoes so finally I was like I 1:40know it so I raised my hand they're like 1:43great we got a project for you next week 1:45so I was like all right I have like 1:47three days to figure out what Vio 1:50is and I hope I can actually figure out 1:53how to use it now luckily it wasn't like 1:55a programming language I mean it's 1:57pretty much a drag and drop capability 2:00and so I literally left the office went 2:03to a bookstore bought the first three 2:05books on Vio I could find spent the 2:07whole week in reading the books and 2:09showed up and got to work on the project 2:12and so it was a bit of a risky move 2:15but I think that's kind of you would 2:18caution others against doing this well 2:20but if you don't take risk you'll never 2:22you'll never achieve and so to some 2:24extent everybody's making everything up 2:26all the time it's like can you learn 2:28faster than somebody else MH is what the 2:31difference is in almost every part of 2:33life and so it was not planned but it 2:36was an accident but it kind of forced me 2:38to figure out that you're going to have 2:40to figure things out you know we we're 2:41here to talk about a Ai and I'm curious 2:44about the evolution of of your 2:48understanding or IBM's understanding of 2:50AI at what point in the last 25 years do 2:53you begin to think oh this is really 2:55going to be at the core of what we think 2:57about and work on at this company 3:01the um computer scientist John McCarthy 3:04he was the he's the person that's 3:05credited with coining the phrase 3:07artificial intelligence it was like in 3:09the 3:1050s and he made an interesting comment 3:13he said he said once it works it's no 3:16longer called AI 3:18MH and that then became it's called like 3:21the AI effect which is it seems very 3:24difficult very mysterious but once it 3:26becomes commonplace it's just no longer 3:29what what it is and so if you put that 3:31frame on it I think we've always been 3:33doing AI at some level I I even think 3:36back to when I joined IBM in 99 at that 3:39point there was work on rules-based 3:43engines analytics all of this was 3:46happening so it all depends on how you 3:50really Define that term You could argue 3:52that you know elements of 3:54Statistics probability it's not exactly 3:57AI but it certainly feeds into it and so 4:00I feel like we've been working on this 4:03topic of how do we deliver better 4:05insights better automation since IBM was 4:09formed if you read about what Thomas 4:11Watson Jr did that was all about 4:13automating tasks yeah is that AI well 4:16probably certainly not by today's 4:18definition but it's in the same zip code 4:21so from your perspective it feels a lot 4:23more like an evolution Than A revolution 4:25is that a fair statement yes yeah which 4:27I think most great things in techn ology 4:30tend to happen that way yeah many of the 4:33the revolutions if you will tend to 4:35fizzle out but even given that is there 4:38I guess what I'm asking is I'm curious 4:40about whether there was a a moment in 4:42that Evolution when you had to readjust 4:45your expectations about 4:47what AI was going to be capable of I 4:49mean was there you know was there a 4:52particular Innovation or a particular 4:55problem that was solved that made you 4:57think oh this is different than what I 5:02thought I would say the moments that 5:05caught our attention 5:07certainly casprov winning the chess 5:09tournament nobody or deep blue beating 5:12CASRO I should say nobody really thought 5:15that was possible before that and then 5:18it was Watson winning Jeopardy these 5:21were moments that said maybe there's 5:23more here than we even thought was 5:26possible and so I I do think there's 5:28there's points in time where we realized 5:32maybe way more could be done than we had 5:35even imagined but I do think it's 5:39consistent progress every month and 5:41every year versus some simal moment now 5:45certainly large language models as of 5:48recent have caught everybody's attention 5:49because it has a direct consumer 5:52application but I would almost think of 5:55that as what Netscape was for the for 5:58the web browser yeah it brought the 6:00internet to everybody but that didn't 6:03become the internet per se yeah I have a 6:07cousin who worked for IBM for 41 years I 6:10saw him this weekend he's in Toronto by 6:13the way I said you work for Rob Thomas 6:16he went he went like this he 6:19goes he said I'm five layers 6:22down but so I always whenever I see my 6:25cousin I ask him can you tell me again 6:27what you do because it's always changing 6:29right I this is a function of working at 6:31IBM So eventually he just gives up and 6:33says you know we're just solving 6:35problems that's all we're doing which I 6:36sort of loved as a kind of frame and I 6:39was curious what's the coolest problem 6:42you ever worked on not biggest not most 6:45important but the coolest the one that's 6:47just like that sort of makes you smile 6:49when you think back on it probably when 6:51I was in 6:52microelectronics because it was a world 6:55I had no exposure to I hadn't studied 6:57computer science 6:59and we were building a lot of high 7:03performance semiconductor technology so 7:06just chips that do a really great job of 7:09processing something or other and we 7:12figured out that there was a market in 7:15consumer gaming that was starting to 7:17happen and we got to the point where we 7:20became the chip inside the Nintendo Wii 7:24the Microsoft Xbox Sony PlayStation so 7:28we basically had the the entire gaming 7:30Market running on IBM chips and so 7:33you're the every parent basically is 7:37pointing at you and saying you're the 7:39culprit probably yeah well they would 7:42have found it from anybody but was it 7:44was the first time I could explain my 7:48job to my kids who were quite young at 7:49that time like what I did yeah like it 7:51was more tangible for them than saying 7:53we solve problems or do you know build 7:55Solutions like it it would became very 7:58tangible for them and I think that's you 8:01know a rewarding part of the job is when 8:03you can help your family actually 8:05understand what you do most people can't 8:06do that it's probably easier for you 8:07they can they can see the books yeah um 8:10but for for some of us in the the 8:12businesses to business world it's not 8:14always as obvious so that was like one 8:15example where the dots really connected 8:18yeah the um there were a couple there's 8:21a couple let's talk about a little bit 8:22of this in the context of of AI I 8:25because I love the frame of problem 8:27solving as a way of understanding what 8:29the function of the technology is so I 8:32know that you guys did something did did 8:34some work with um I never know how to 8:37pronounce it is it Seeva Seeva Seeva 8:40with the football club Seeva in Spain 8:42tell me about tell me a little bit about 8:45that what problem were they trying to 8:47solve and why did they call you 8:50in every sports franchise is trying to 8:55get an advantage right let's just be 8:57that clear everybody's how can I use 9:00data analytics insights anything that 9:03will make us 1% better on the 9:06field uh at some point in the 9:09future and Sevilla reached out to us 9:13because they had seen some of the we've 9:15done some work with the Toronto Raptors 9:16in the past and others 9:19and their thought was maybe there's 9:21something we could do they'd heard all 9:23about generative AI that heard about 9:25large language models and the problem 9:28back to your point on solving problems 9:31was we want to do a way better job of 9:34assessing talent because really the the 9:37lifeblood of a sports franchise is can 9:40you continue to cultivate Talent can you 9:43find talent that others don't find can 9:45you see something in somebody that they 9:47don't see in themselves or maybe no 9:49other team sees in them and we ended up 9:52building something with them called 9:53Scout advisor which is built on Watson X 9:58which basically just 10:00ingest tons and tons of data and we like 10:04to think of it as finding you know the 10:06the needle and the hay stack of you know 10:09here's three players that aren't being 10:11considered they're not on the top teams 10:14today and I think working with them 10:16together we found some pretty good 10:18insights that's helped them out and how 10:19what was intriguing to me was we're not 10:21just talking about uh quantitative data 10:24we're also talking about qualitative 10:26data but that's the puzzle part of the 10:29thing thing that fascinates me how does 10:30one incorporate qualitative analysis 10:32into that sort of so you're feeding in 10:35Scout scouting reports and things like 10:38that I gotta real think about how much I 10:40can actually 10:44disclose but if you think about so 10:46quantitative is relatively easy yeah 10:49every team collects that you know what's 10:53the 40 yard dash I don't think they use 10:55that term certainly not in Spain uh 10:58that's all quantitative qualitative is 11:00what's happening off the field mhm it 11:03could be diet it could be habits it 11:05could be Behavior you can imagine a 11:08range of things that would all feed into 11:11an athlete's performance yeah and so 11:15relationships there's many different 11:17aspects and so it's trying to figure out 11:20the right blend of quantitative and 11:22qualitative that gives you a unique 11:24Insight how transparent is that kind of 11:27system I mean is it telling you it it's 11:29saying pick this guy not this guy but is 11:32it telling you why it prefers this guy 11:34to this guy is that I think for anything 11:37in the realm of AI you have to answer 11:38the why question yeah otherwise you fall 11:41into the Trap of the you know the 11:44proverbial black box and then wait I 11:47made this decision I never understood 11:49why it didn't work out so you always 11:51have to answer why without a doubt M and 11:54how is why 11:56answered sources of data the reasoning 12:00that went into it and so it's basically 12:02just tracing back the chain of how you 12:05got to the 12:06answer and in the case of what we do in 12:08Watson X is we have IBM models we also 12:11use some other open source models so it 12:13would be which model was used what was 12:16the data set that was fed into that 12:17model how is it making decisions how is 12:20it 12:20performing is it robust meaning is it 12:23reliable in terms of if you feed it to 12:26with the same data set do you get the 12:27same answer yeah these are all the you 12:30know the technical aspects of 12:31understanding the why how quickly do you 12:34expect um all professional sports 12:36franchises to adopt some kind of are 12:38they already there if I went out and 12:40pulled the general managers of the 100 12:43most valuable sports franchises in the 12:45world how many of them would be using 12:47some kind of AI system to assist in 12:49their 12:51efforts uh 120% would meaning that 12:54everybody's doing it and some think 12:56they're doing way more than they 12:57probably actually are so everybody's 12:59doing it I think what's weird about 13:01sports 13:02is everybody's so convinced that what 13:06they're doing is unique that they 13:09generally speaking don't want to work 13:11with the third party to do it because 13:12they're afraid that that would expose 13:14them but in reality I think most are 13:16doing 80 to 90% of the same 13:19things uh so but without a doubt 13:22everybody's doing it yeah yeah the um 13:26the other C that I love was there was 13:28one about a a shipping line Trion on the 13:31Mississippi River um tell me a little 13:33bit about that project what problem were 13:35they trying to 13:37solve think about the the problem that I 13:40would say everybody noticed if you go 13:42back to 2020 was things are getting H up 13:46held up in ports there was actually an 13:48article in the paper this morning kind 13:49of tracing the history of what happened 13:51in 2020 2021 and why ships were 13:54basically sitting at Seas for months at 13:56a time and at that stage we just we had 14:00a massive throughput issue 14:03but moving even beyond the pandemic you 14:06can see it now with ships getting 14:09through like Panama Canal there's like 14:11there's like a narrow window where you 14:13can get through and if you don't have 14:15your paperwork done you don't have the 14:18right approvals you're not going through 14:19and it may cost you a day or two and 14:21that's a lot of money in the shipping 14:23industry and the tricon example it's 14:26really just about when you're pulling 14:28into a port if you have the right 14:31paperwork done you can get Goods off the 14:35ship very quickly they ship um a lot of 14:38food which by definition since it's not 14:42packaged food it's fresh food there is 14:44an expiration period and so if it takes 14:47them an extra two hours certainly 14:51multiple hours or a day they have a 14:53massive problem because then you're 14:54going to deal with spoilage and so it's 14:56going to set you back and what we worked 14:59with them on is using assistant that 15:02we've built in Watson X called 15:04orchestrate which basically is just AI 15:08doing digital labor so we can replicate 15:12nearly any repetitive task and do that 15:15with software instead of humans so as 15:19you may imagine shipping industry still 15:21has a lot of paperwork that goes on and 15:24so being able to take forms that 15:26normally would be multiple hours of 15:28filling it out this isn't right send it 15:30back we've basically built that as a 15:32digital skill inside of Watson X 15:35orchestrate and so now it's done in 15:38minutes now did they real did they 15:41realize that they could have that kind 15:43of efficiency by teaming up with you or 15:45is that something you came to them and 15:47said guys you could we can do this way 15:49better than you think what's 15:52the I'd say it's always it's always both 15:55sides coming together at a moment that 15:57for some reason makes sense mhm cuz it 16:00you could say why didn't this happen 16:02like 5 years ago like seems so obvious 16:04well technology wasn't quite ready then 16:07I would say but they knew they had a 16:09need because I forget what the precise 16:12number is but you know reduction of 16:14spoilage has massive impact on their 16:17bottom line 16:18M and so they knew they had a need we 16:22thought we could solve it and the two 16:25together um who did you guys go to them 16:28though my point did they come to you I 16:30recall that this one was an inbound 16:32meaning they had reached out to 16:34IBM say we'd like to solve this problem 16:37I think it went into one of our digital 16:38centers if I if I recall it's a 16:40literally phone call yeah but so the the 16:43other the reverse is more interesting to 16:46me because there seems to be a very very 16:48large Universe of people who have 16:50problems that could be solved this way 16:52and they don't realize it what's 16:55your is there a shining example of this 16:58of some 16:59you just can't you just think could 17:00benefit so much and isn't benefiting 17:03right 17:05now maybe I'll answer it slightly 17:07differently I'm I'm surprised by how 17:11many people can benefit that you 17:12wouldn't even logically think of first 17:15let me give you an example there's 17:19a franchiser of hair salons M Sport 17:23Clips is the name my My Sons used to go 17:26there for haircuts because they have 17:27like TVs and you can watch so so they 17:29loved that they got entertained while 17:31they would get their haircut I think the 17:33last place that you would think is using 17:35AI today would be a franchiser of hair 17:38salons 17:40yeah but just follow it through the 17:44biggest part of how they run their 17:46business is can I get people to cut 17:48hair and this is a high turnover 17:51industry because there's a lot of 17:52different places you can work if you 17:53want to cut hair people actually get 17:55injured cutting hair because you're on 17:56your feet all day that type of thing and 17:59they're using same technology 18:01orchestrate as part of their recruiting 18:04process how can they automate a lot of 18:06people submitting resumés who they speak 18:10to how they qualify them for the 18:12position and so reason I give that 18:15example is the the opportunity for AI 18:18which is unlike other Technologies is 18:21truly 18:22unlimited it will touch every single 18:25business it's not the realm of the 18:27Fortune 500 or or the Fortune 1,000 this 18:31is the fortune any size and I think that 18:34may be one thing that people 18:35underestimate about AI yeah what about I 18:39mean I was thinking about education as 18:41as a kind of I mean education is the 18:44perennial uh Whipping Boy for you guys 18:48are living the 19th century right I'm 18:51just curious about if a if a 18:54superintendent of a public school system 18:56or the president of a university sat 18:59down and had lunch with you and 19:01said do the University first my costs 19:03are out of control my my uh enrollment 19:07is down my students hate me and my board 19:11is revolting 19:13help how would you s how would you think 19:16about helping someone in that 19:19situation I spend some time with 19:21universities I like to go back and 19:23visit Alma moders where I went to school 19:27and so I do that every year MH the the 19:30challenge I've Hol of universities is 19:32there has to be a will yeah and I'm not 19:35sure the incentives are quite right 19:37today because bringing in new technology 19:41let's say we want to go after we can 19:42help you figure out student 19:45recruiting or how you automate more of 19:47your 19:49education everybody suddenly feels 19:51threatened at a university hold on 19:53that's my job I'm the one that decides 19:55that or I'm the one that wants to 19:57dictate the course so there has to be a 20:00will so I think it's very possible and I 20:04do think over the next decade you will 20:06see some universities that jump all over 20:08this and they will move ahead and you 20:11see others that do not because it's very 20:15possible where how does when you say 20:18there has to be a will um is that the 20:21kind is that a kind of thing that you 20:23that people at IBM think about like when 20:26in this conversation you hypothetical 20:28conversation you have with the 20:29University president would you give 20:31advice on on where the will comes 20:35from I don't do that as much in a 20:37university context I do that every day 20:39in a business 20:41context because if you can find the 20:44right person in a business that wants to 20:46focus on growth or the bottom line or 20:50how do you create more productivity yes 20:52it's going to create a lot of 20:54organizational resistance potentially 20:56but you can find somebody that will 20:57figure out how to push that through I 21:00think for 21:01universities I think that's also 21:03possible I'm not sure there's there's 21:05there's a return on investment for us to 21:07do that yeah yeah yeah let's let's 21:11define some terms um uh AI years a term 21:16I told you like to use what does that 21:19mean we just started using this term 21:22literally in the last three months 21:25and it was a it was what we observed in 21:28ter 21:29which is most technology you build you 21:32say all right what's going to happen in 21:33year one year 2 year three and it's you 21:37know largely by by a calendar AI years 21:40are the idea that what used to be a year 21:43is now like a 21:45week and that is how fast the technology 21:47is moving and to give you an example we 21:50had one client we're working with 21:52they're using one of our Granite models 21:55and the results there we getting were 21:56not very good accuracy was not there 21:59their performance was not there so I was 22:01like scratching my head I was like what 22:02is going on what business were they in 22:04they were Financial Services the bank so 22:07I'm scratching my head like what is 22:08going on everybody else is getting this 22:10and like these results are horrible and 22:13I said to the team which version of the 22:16model are you using this was in February 22:20like we're using the one from 22:21October I was like all right now we know 22:24precisely the problem because the model 22:26from October is effectively useless now 22:28since we're here in February are you 22:30serious you you actually useless ABS 22:33completely useless yeah that is how fast 22:36this is changing and so the minute same 22:39use case same data you give them the 22:42model from late January instead of 22:45October the results are off the charts 22:48yeah wait so what exactly happened 22:50between October and January the model 22:52got way better but dig into that like 22:54what do you mean by the we are con we 22:57have built large compute infrastructure 23:00where we're doing model training and to 23:02be clear model training is the realm of 23:05probably in the world my guess is five 23:08to 10 companies MH and 23:10so you build a model you're constantly 23:13training it you're doing fine-tuning 23:16you're doing more training you're adding 23:17data every day every hour it gets 23:20better and so how does it do that you're 23:23feeding it more data you're feeding it 23:25more live examples 23:28using things like synthetic data at this 23:30point which is we're basically creating 23:32data to do the training as well all of 23:34this feeds into how useful the model is 23:37and so using the October model those 23:40were the results in October just a fact 23:43that's how good it was then but back to 23:46the concept of AI years two weeks is a 23:49long time and is that are we had a in a 23:53steep part of the model learning curve 23:55or do you expect this to continue along 23:57this at this 23:59Pace I think that is the big 24:02question and don't have an answer yet by 24:05definition at some point you would think 24:07it would have to slow down a bit but 24:09it's not obvious that that is on the 24:11horizon still speeding up yes how fast 24:15can it 24:17get we've debated can you actually have 24:20better results in the afternoon than you 24:22did in the 24:23morning really it's nuts yeah I know but 24:27that's that's why we came up with this 24:28term because I think you also have to 24:30think of like Concepts 24:33that gets people's attention so you 24:36you're basically turning into a bakery 24:38you're like the bread from yesterday you 24:40know you can have it for 25 cents but I 24:43mean you you do preferential pricing you 24:45could say We'll charge you X for 24:49yesterday's model 2 x for today's 24:52model I think that's dangerous um as a 24:55merchandising strategy but I get your 24:57point yeah but that's crazy and this by 25:01the way so this model is the same true 25:02for all models you're talking 25:04specifically about a model that was 25:05created to help some aspect of a 25:08financial 25:09services so is that kind of model 25:12accelerating faster and learning faster 25:13than other models for other kinds of 25:15problems so this domain was code yeah 25:20and so by definition if you're feeling 25:22feeding in more data so so more code you 25:25get those kind of results um it does 25:27depend on the model type yeah there's a 25:30lot of code in the world and so we can 25:33find that we can create it like I said 25:35um there's other aspects where there's 25:38probably less inputs available which 25:40means you probably won't get the same 25:42level of iteration yeah but for code 25:44that's certainly the cycle times that 25:45we're seeing yeah and how do you know 25:47that let's stick with this one example 25:49of this model you have how do you know 25:51that year model is better 25:54than big Company B down the street a 25:57client asked you why would I go with IBM 25:59as opposed to some there's some firm in 26:02the valley that says they have a model 26:04on this what's your how do you how do 26:05you frame your 26:08advantage well we Benchmark all of this 26:12and I think the most important is metric 26:14is price 26:15performance not price not performance 26:18but the combination of the two and we're 26:21super competitive there well for what we 26:23just released with what we've done in 26:25open source we know that nobody's close 26:27to us right now on code now to be clear 26:29that will probably change yeah because 26:31it's like Lea frog people will jump 26:33ahead then we jump back ahead but we're 26:37very confident that with everything 26:39we've done in the last few months we've 26:41taken a huge leap forward here yeah this 26:44it's I mean this goes back to the point 26:46I was making in the beginning so about 26:48the difference between your 20-some self 26:51and 99 and yourself today but this time 26:55compression is has to be a crazy 26:57adjustment so your the concept of what 27:00you're working on and how you make 27:02decisions internally and things has to 27:04undergo this kind of Revolution if 27:06you're if you're switching from I mean 27:08back in the day a model might be useful 27:11for how long years years I think about 27:15you know statistical models that sit 27:17inside things like SPSS which is a 27:21product that a lot of students use 27:22around the world I mean those have been 27:23the same models for 20 years yeah and 27:25they're still very good at what they do 27:27and so yes it's a completely it's a 27:30completely different moment for how fast 27:33this is moving and I think it just 27:36raises the bar for everybody whether 27:38you're a technology provider like us or 27:41you're a bank or an insurance company or 27:44or a shipping company to say how do you 27:47really change your culture to be way 27:50more 27:51aggressive than you normally would 27:54be does this mean this is a weird 27:57question but does this mean a different 27:59set of kind of personality or character 28:02traits are necessary for a decision 28:04maker in Tech now than 25 years 28:09ago there's a there's a book I saw 28:12recently uh was called The Geek way 28:14which talked about how technology 28:17companies have started to operate in 28:19different ways maybe than many you know 28:22traditional 28:24companies and more about being data 28:27driven 28:29more about delegation are you willing to 28:32have the smartest person in the room 28:35make decision as opposed to the highest 28:37paid person in the room I think these 28:39are all different aspects that every 28:41company is going to face yeah yeah next 28:45term talk about open when you use that 28:47word open what do you 28:50mean I think there's really only one 28:52definition of open which is for 28:55technology is open source MH and open 28:58source means the code is freely 29:02available anybody can see it access it 29:06contribute to it and what is tell me 29:08about why that's an important 29:12principle when you take a topic like AI 29:16I think it would be really bad for the 29:19world if this was in the hands of one or 29:22two 29:24companies or three or four doesn't 29:26matter the number some small number 29:28think about like in history sometime 29:30early 29:311900s the Interstate Commerce Commission 29:34was created and the whole idea was to 29:37protect Farmers from railroads meaning 29:41they wanted to allow free trade but they 29:43knew that well there's only so many 29:44railroad tracks so we need to protect 29:46Farmers from the shipping cost that 29:49railroads could impose so good idea but 29:52over time that got completely overtaken 29:54by the railroad 29:55Lobby and then they used that to 29:58basically just increase prices yeah and 30:00it made the lives of farmers way more 30:03difficult I think you could play the 30:05same analogy through with AI if you 30:08allow a handful of companies to have the 30:11technology you regulate around the 30:14principles of those one or two companies 30:15then you've trapped the entire world I 30:17think that would be very 30:19bad so is there a danger of that happen 30:23for sure I mean there's companies in 30:25Watson in Washington every week trying 30:27to 30:29achieve that outcome MH and so the 30:31opposite of that is to say it's going to 30:33be an open 30:34source because nobody can dispute open 30:37source yeah because it's right there 30:38everybody can see it yeah and so I'm a 30:42strong believer that open source will 30:43win for AI it has to win it's not just 30:46important for business but it's 30:47important for 30:50humans on the I'm curious about on the 30:54sort of list of things you worry about 30:56actually let me before 30:58let me ask this question very generally 31:00what is the list of things you worry 31:01about what's your top five business 31:03related worries right 31:05now top those are your first question we 31:08could be here for hours for me to 31:10answer I didn't say business related we 31:13can leave you know your kids haircuts 31:16got it out of 31:18the number one is always it's the thing 31:21that's probably always been true which 31:23is just people 31:25MH do we have the right skills are we 31:28doing a good job of training our people 31:31are our people doing a good job of 31:33working with clients like that's number 31:35one number two is 31:39innovation are we pushing the envelope 31:41enough or are we staying 31:44ahead number three is which kind of 31:47feeds into the Innovation one is 31:49risk-taking are we taking enough risk we 31:52without risk there is no growth and I 31:55think the Trap that every larger company 31:58inevitably falls into is is conservatism 32:02yeah things are good enough and so it's 32:05are we pushing the envelope are we 32:07taking enough risk to really have an 32:09impact I'd say those are probably the 32:11top three that I spend most of my time 32:13talk about the last term to Define 32:15productivity Paradox something I know 32:17you've thought a lot about what does 32:18that 32:19mean so I started thinking hard about 32:21this because all I saw and read every 32:24day was Was Fear about AI 32:29and I studied economics and so I kind of 32:33went back to like basic economics and 32:36there's been like a macro 32:38investing formula I guess I would say 32:41it's been around forever that says 32:43growth comes 32:46from productivity growth plus population 32:50growth plus debt 32:53growth so if those three things are 32:55working you'll get GDP growth 32:58and so then you think about that and you 33:00say well debt growth we're probably not 33:03going back to 0% interest rates so to 33:06some extent there's going to be a 33:07ceiling on 33:08that and then you look at population 33:11growth there are shockingly few 33:14countries or places in the world that 33:16will see population growth over the next 33:1830 to 50 years in fact most places are 33:21not even at replacement rates yeah and 33:24so I'm like all right so population 33:25growth is not going to be there 33:28so that that would mean if you just take 33:29it to EXT to the extreme the only chance 33:33of continued GDP growth is 33:38productivity and the best 33:41way to solve productivity is AI so 33:45that's why I say it's a paradox on one 33:46hand Everybody's scared half to death 33:49it's going to take over the world take 33:52all of our jobs ruin 33:54us but in reality maybe it's the other 33:57way which is it's the only thing that 33:58can save us yeah and if you believe that 34:02economic equation which I think has 34:03proven quite true over hundreds of years 34:06I do think it's probably the only thing 34:07that can save 34:08us actually looked at the numbers 34:11yesterday for a totally random reason on 34:13population growth in Europe I me see 34:16this a special bonus question we'll see 34:17how smart you are which country in 34:19Europe Continental Europe has the 34:22highest population 34:24growth it's small Continental Europe um 34:29probably one of the nordics I would 34:30guess 34:32close Luxembourg okay something is going 34:35on in 34:37Luxembourg I feel like we all of us need 34:40to investigate they're at 1.49 which in 34:42the day by the way would be a relatively 34:45that's the best performing country is 34:47one I mean in the day you'd be countries 34:49had routinely had 2 point something you 34:51know percent growth in a in a given year 34:54um last question you're writing a book 34:56now we were talk chat about it backstage 34:59um and now I appreciate the Paradox of 35:01this book which is in a universe where 35:04the model is better in the afternoon 35:06than it is in the morning how do you 35:08write a book that's like printed on 35:09paper and expect it to be 35:14useful this is the challenge and I uh 35:18I'm an incredible author of useless 35:20books meaning most of what I've spent 35:22time on in the last decade is stuff 35:24that's completely useless like a year 35:26after it's written and so when um we 35:30were talking about I was like i' would 35:31like to do something around AI That's 35:33Timeless yeah that would be useful 10 or 35:3720 years from 35:39now but then to your point so how do you 35:42how is that even remotely 35:44possible if the model's better in the 35:47afternoon than in the morning so that's 35:49the challenge in front of us but the 35:50book is around AI value creation so kind 35:53of links to this productivity Paradox 35:56and how do you actually get sustained 35:59value out of AI out of automation out of 36:05data science and so the biggest 36:07challenge in front of us is can we make 36:09this 36:10relevant past the day that it's 36:12published how are you setting out to do 36:15that I think you have to to some extent 36:18level it up to bigger Concepts which is 36:21kind of why I go to things like 36:23macroeconomics population geography as 36:27opposed to going into the the weeds of 36:29the technology itself if you write about 36:32this is how you get better performance 36:33out of a 36:34model we can agree that will be 36:37completely useless two years from now 36:39maybe even two months from now yeah and 36:41so it will be less in the technical 36:45detail and more of what is sustained 36:48value creation for AI which if you think 36:51on what is hopefully a 10 or a 20 year 36:53period it's probably we're kind of 36:56substituting AI for Tech technology now 36:58I've realized because I think this has 36:59always been true for technology it's 37:01just now ai is the thing that everybody 37:04wants to talk about um but let's see if 37:07we can do it time will tell did you have 37:10any inkling that the pace that this AI 37:12years phenomenon was going to that 37:15things the pace of change was going to 37:16accelerate so much because you had Moors 37:19law right you had a model yeah in the 37:21technology world for this kind of 37:24exponential increase in so were you 37:28were you thinking about that kind of 37:30acceler similar kind of acceleration in 37:35the I think anybody that said they 37:37expected what we're seeing today is 37:40probably exaggerating I think it's way 37:44faster than anybody expected yeah but 37:48technology back to your point of Mo's 37:50law has always accelerated through the 37:53years so I wouldn't say it's a shock but 37:56it is surprising 37:58yeah you've had a kind of 38:00extraordinary uh privileged position to 38:03watch and participate in this revolution 38:05right I mean how many other people have 38:07been in that have 38:10ridden this wave like you 38:12have I I do wonder is is this really 38:15that much different or does it feel 38:17different just because we're here 38:19meaning I do think on one level yes so 38:22in the time I've been in IBM mobile 38:26happened social Network happened 38:29blockchain happened AI so a lot has 38:32happened but then you go back and say 38:33well but if i' had been here 38:35between 1970 and 38:38'95 there were a lot of things that are 38:40pretty fundamental then to so I wonder 38:43almost do we do we always exaggerate the 38:45time frame that we're 38:48inh I don't know yeah but it's a good 38:52idea 38:54though I think the ending with the 38:56phrase I don't know it's a good idea 38:59though it's probably a great way to wrap 39:02this up thank you so much thank you Mal 39:05[Applause]