How to Design a Data Strategy for AI, with IBM Chief Data Officer (CXOTalk #793)

How to Design a Data Strategy for AI, with IBM Chief Data Officer (CXOTalk #793)

Show Video

Today on Episode #793 of CXOTalk,  we're speaking about data and AI.   Our guests are Inderpal Bhandari, the global chief  data officer of IBM, and Anthony Scriffignano, the   former chief data scientist at Dun & Bradstreet. Inderpal, welcome to CXOTalk. It's great to see   you. Please tell us about your work at IBM. I'm actually a full-time chief technologist.   When I first became chief data officer in 2006,  there were just four of us globally. I was the  

first in healthcare. Then the profession and the  related professions like chief analytics officer,   transformation officer, that took off, and I  happened to be fortunate enough to ride with it,   and I've done this job full-time, so IBM being  the fourth and perhaps the most complicated.  At IBM, my strategy, data strategy, has been  to make IBM itself into an AI enterprise and   then use that as a showcase for our clients  and customers because our clients look very   much like us. That's what I've been doing  for the last seven and a half years or so.  Anthony Scriffignano, welcome back to CXOTalk.  You're a good friend. It's great to see you.   Tell us about your work these days. Thank you very much, Michael. It's   great to see both of you. As you mentioned, I was with  

Dun & Bradstreet for quite a long time (over 20  years). Right now, I'm doing a number of things.   Probably front foot is as a distinguished fellow  with the Stimson Center, which is a think tank.   "Think tank," I'll put in quotes because there  are a lot of what I would call applied research   or action research where they get involved  in doing things, not just writing about them. 

I've been involved with things that are called AI.  The term has been around probably since the '50s,   but I've been involved with it  as it's become computational from   its birth. I know Inderpal has as well. There are lots of things going on in the   world right now in terms of regulatory focus on AI  as well as new types of AI becoming the greatest   new shiny object and everyone pays attention  to them. I stand for the science behind it.  • What do you have to believe? • What has to be true in order for you to   do that thing that you think is so cool? • And why is it better than   what you're doing today? • And what is the cost of it?  I've tried to ask those emperor's  new clothes kind of questions,   and that's the role I'm playing right now. We're talking about data and AI. I think where  

we need to start is, when we talk about an AI data  strategy, what actually is that? Inderpal, do you   want to maybe take a crack at that to start? AI is only as good as the data that is used   to train that AI because AI has a training  sequence and then an inference sequence. The   training sequence has to do with seeing all kinds  of related data so that it can then train itself   to figure out what the right output is when it's  shown an input that it may not have seen before.  If the data (to begin with) is flawed or low  quality, the AI will not work effectively.   It's the garbage in, garbage out, that  kind of phenomenon that you would have.  They go hand-in-hand. And very often, you think  of people talking about AI. If they haven't  

really looked at the data but they embark on an  AI strategy, that is going to be very high risk.   It'll most likely fail because they'll have to go  back and straighten out the data strategy first   just so that it's fit for purpose. Now, when you say fit for purpose,   what that means is if you know what the business  objective is that you're trying to serve. So,   it could be something quite narrow like a  specific objective. It could be something like   I want to understand what segments of my business  should I try to expand to increase my top line.  In which case if it's segments of business –  data about your clients, about your products,   et cetera – those things become very  important. You'd want to make sure  

that that data is of very high quality. On the other hand, if it's something at   a strategy level, which is kind of what happened  when I joined IBM, I mean IBM wanted to be a cloud   and AI company. To be a cloud and AI company,  eventually, we landed at the point that, well,   let's transform ourselves internally before  we show this off to our clients and customers.  That became a strategy that was enterprise-wide.  And we realized that now, well, for instance,   not only do we have to make sure that our  structured data is in order but also our   unstructured data because we are going to go after  this and transform ourselves into an AI company.  There are two aspects there that are relevant.  One is at a strategy level when you're aligning  

to the business strategy, or it could be  more narrow to a specific business objective.  Anthony, the challenge of aligning the data  strategy to the business objectives is something   that many organizations struggle with. What  thoughts or advice do you have on making that   work? You've seen so many different scenarios. You would really have to unpack what Inderpal   just said quite a bit to really get at  the essence of it. And I did (when I   was listening to him), but he was using  some terminology very carefully there. 

A lot of times, organizations don't have one  strategy: Make more money. Grow... fill in the   blank. The things that we learn in business school  – you can serve your shareholders, you can serve   your customers, you can serve your employees –  it's kind of hard to do all of those things at   the same time because, very often, optimizing  for one is less optimal for one of the others. 

The strategy of which we speak when we start to  talk about AI has some very serious implications,   these methods that we're talking about. And  I should say that these days it's rare that   only one method gets applied. Very often, there  are many methods being applied simultaneously.  There are some commonalities.  One of the commonalities is that   the quality of the data has many dimensions. Truth: If your AI is going to ingest data, it's   going to probably presume it's all true. Well,  all data is not necessarily simultaneously true. 

It may have been true at the time that it  was created but maybe not so much anymore   at the time that it's curated. So, how old is  the data? Is it still true? How would you know   that it's still true before you consume it into  an algorithm or an approach that presumes that?  I love to say that, when we go to court, we swear  to tell the truth, the whole truth, and nothing   but the truth. That's because those are three  different things, and those are three different   ways to manipulate veracity or understanding. When you get back to this concept of strategy,   well, whose strategy? What part of the  organization specifically? What objective?   How would we know when we were successful? Asking those questions is very often a source   of contention because the people in the room that  all think they want the same thing realize they   don't. And when you unpack it a little further,  they realize that to get what the guy on the   left wants, you have to get less of what the  person on the right wants, and it's not pretty. 

It's not really a technical problem as much  as it is an alignment problem and getting   everybody to agree on what they want so that  we would know that this strategy of which we   speak is actually what this AI of which we  speak is delivering. It's really difficult.  These roles (like the chief data officer, the  chief transformation officer), the reason these   are CXO roles is because of what Anthony just  said. You have to be part of that discussion.  It's not so much like there is a concrete business  objective. Sometimes you get into those situations  

where it's very clear-cut. But more often than  not, it's a strategic discussion in terms of   understanding, clarifying, perhaps even adding to  the business strategy; and then relating it back   to what you are trying to do with data and AI. Unless you're in a position to have that kind   of conversation (and you also have the wherewithal  to pull that off), you won't really be successful.  

That's why these are CXO roles because it's really  part of the negotiation that goes on to align the   business strategy to the data or the AI strategy. Maybe I can just add a little bit to that. The   whole concept of being in  the room is so important.  Back in the day, the goals and the objectives  would come down from on high, and the folks   with the keyboards would just make it so. It  doesn't work that way anymore, and it can't   really work that way anymore. It's so critical  that the folks in the roles that Inderpal and I  

have had have a seat at the table; understand  what went into the ask and not just the ask.  Very often, what folks want and what they  need are two very different things. And so,   without being arrogant about it, you ask  a lot of questions and you get at what   they really needed in the first place, which is  probably not what they started out asking for.  You're talking about organizational alignment with  business strategy and, at a high enough level,   this is true for every business decision  that needs to be made. Yet, when you hear  

people talking about AI and data strategy, the  conversation turns very quickly to: What kind   of data do we need? Where do we get that data?  What's the technology that we're going to use to   aggregate and to manipulate that data? What kind  of models are we using? Now I'm confused because   you're talking about one thing and I hear the  entire world talking about something different.  The world tends to focus on  the hammers and the nails,   and it tends to focus on the tools that are going  to be used for the purpose. If I come to your   house and I say, "Do you want to put an addition  on your house?" and I've met with the architect   and I understand your objectives, let's talk. If someone else comes to your house and says,   "I'm going to build you a beautiful  addition, and I'm going to use the hammer,"   you don't really care about the hammer.  Of course, the hammer is important.  It's very important that we have the right  data, the right tools, the right technology,   the right people. It's people, process, tools,  and mindset. All of those have to be aligned  

in order to get this right. But it starts with  making sure you're focused on the right mission.  Yes, that never changes, that piece  of aligning back to the business.  The way I would put it is, yes, no matter  how promising the technology, no matter how   dramatic the advance, et cetera, it doesn't let  the organization off the hook for coming up with   a sound business strategy and then of aligning  these elements to that business strategy. That's   still going to be very much needed. In fact,  maybe even more so than before (as you try to   go after these new approaches and methods). Be sure to subscribe to our YouTube channel   and hit the subscribe button at the bottom of our  webpage. You can subscribe to our newsletter, and  

we'll tell you and notify you about our excellent  upcoming shows and guests. We have lots of them.  Would you say that the business side is more  difficult or harder to achieve than the data   and technology foundations (in your experience)? I would say that if you take something new,   you probably want to draw the distinction between  a mature technology and a technology that's more   recent, nascent, or emerging. If you take  something new like that, like the ladder,   then there is a tremendous amount of  complexity on the technology side as well.  Early on, when getting into this game, when  you were working on AI (for instance, at IBM),   it became very clear, as we went forward,  that there were four elements that had   to move in lockstep: data, technology,  workflow, and culture. Those four kind  

of had to move at the same time. Otherwise,  the adoption was not going to be effective.  The technology piece, emerging technology, at  that time cloud was emerging. There were a lot   of AI techniques that were emerging, the deep  learning stuff with GPUs and things like that.  You had to make all that stuff work together, so  there is significant complexity in the technology   piece. But there is also significant complexity  in the data piece and the workflow piece and then,   eventually, in the culture piece of the  organization. The stuff that we were  

talking about in terms of the negotiation,  working with the C-suite, there's a lot of   the cultural aspect that goes into it. There are many organizations one could   go into and, essentially, as Anthony said, they  would still want to give you a set of objectives   and say, "Here. Go off. Implement this. We really  don't want to hear from you about anything else.   These are your marching orders. Go off and  implement this." But that's the wrong approach   when you're trying to bring in an emerging  technology and use it to impact the business. 

Anthony, we have a question exactly on this topic  from Twitter, from Arsalan Khan. Maybe you can   share your thoughts on this. He says, "When we  talk about alignment, there's business strategy,   enterprise business architecture, change  management, culture, and now data strategy."  

All right, Anthony. What's your prescription  then to make all these layers work together   and align? It sounds almost impossible. Almost impossible is a synonym for possible. So,   if you said it was impossible,  then we have to talk. Right? 

First of all, thank you for the question (from  someone who knows how to ask a good question).  I would say it's really important that you  start with the question, with the objective.   Everybody wants to jump to the technology. They  want to jump to the data, the deal, the thing.  There are two factions in the room. The  one faction is focused on the revenue,   the growth, what's going to happen to the  organization. The other faction is focused on,  

"All right. Let's get going. Let's start doing  stuff. Let's start cooking in the kitchen."  I'm usually the one somewhere in the middle of  those two saying, "Let's make sure we're answering   the right question here." I'm not slowing you  down. I'm actually making sure we get done in   a way that we don't fall over the finish line. It is very difficult to get all those factions in   the same place. Probably the most important thing  you have to do is be able to listen to each other   and not start immediately talking about hammers  and nails or immediately start talking about what   color we're going to paint the finished product. Somewhere in the middle is: Why are we doing   this? What are we not doing while  we're doing this? Do we know?  There's a big difference between can we do it and  should we do it. What are we giving up while we do  

it? What about compliance? What about regulatory? How do we know that the data that we have is the   right data to make the decision you want?  Just because you believe it and you have   your confirmation bias and you found  one or two pieces of data that support   your hypothesis doesn't make you right. We have to ask these difficult questions,   and there's a very fine line between being right  and being dead. You have to be able to ask them   in a way that doesn't annoy. It can annoy them a  little bit, but you have to annoy them just to the  

point where they don't kick you out of the room. Keep asking those "help me understand" kind of   questions until we get to a shared  understanding of what it is we're   trying to achieve and the opportunity cost  of all the other things that we're not doing.  Inderpal, you're a technologist. So, if  this is strictly then a business issue of  

organizational alignment, why do technologists  play such an important role in this discussion,   such a foundational, fundamental role? I think the best way to think of my role,   of people in similar situations, is that  of a change agent. The catalyst for the   change is the technology, but the change has to be  affected in the organization and in the business.   You have to be able to bridge those  two to be able to do this successfully.  It's a transformation, and the transformation  typically has those elements that I talked   about for what we do, what I do: data,  technology, workflow, and the culture.  I'll give you one other thing. There is  a lot to be done in terms of changing  

the culture of an organization when  you try to bring about this change.  What we saw at IBM when we pushed forward with our  data and AI strategy was that the adoption of the   platform was triggered far more by the bottom-up  measures that we put in place. We had a team that   was empowered to engage with other teams that were  working in the business doing workflows, quote to   cash, procurement—things like that—supply chain. We have an empowered team on the technology side   which didn't really need to come back for  direction or instruction. But if they found  

a like-minded team, they could go ahead and move  forward with the transformation. We found that   85% of the adoption actually came from that path  as opposed to the top-down path, and so forth.  It really is all about how you  affect the change. But obviously,   if the catalyst is the technology, then you've  got to be able to walk that walk as well. 

But you can't discount the other side  of it. You have to really be the bridge.  Michael, I smiled when you called Inderpal a  technologist and he very diplomatically didn't   respond. I think you can tell by that answer that  you have to be much more than just an expert in   the technology to get what he just said right. In large organizations, what's happening right   now is a massive federation of data and  AI capability. It's not like you go to   the room where the people that know how to  do that live and ask them to do it for you.  Almost anybody can get these capabilities on their  desktop. It doesn't mean it's the right place to  

do it, but they can start doing it there. Everyone feels like they're an expert. Just   like when we all first got—I'm trying  not to name a product, but I think I   can say—Harvard Graphics. In the days even  before PowerPoint where, all of a sudden,   we could all lay things out on the screen, we all  thought we were experts in design, layout, font   selection, and all of that. Of course, we weren't. There's an old joke where the punchline is death  

by PowerPoint. We all know versions of that joke.  And I'm not picking on PowerPoint. [Laughter]  Federating a capability like that across an  organization or across the world comes with   some risk that those who really are practitioners,  who know what the difference is between what you   can do and what you should do, understand  the implications of going down a certain   path and the difficulty of changing course  once you get too far down that path. They   have to be able to hear what's going on. For what Inderpal is describing to work well   85% of the time, it does require an organization  that actually talks to each other or at least   talks up to people who talk to each other (up  and down). But that's not always the case.  You can't just throw everything out in the middle  of the floor and say, "Here you go. Everybody,  

play with this and do whatever you want."  That will not work. That will end in tears.  You have governance. You have focus on  these foundational pieces. What about the   interface between the technology and what you're  describing? The whole world (and organizations,   by and large) tend to focus on that technology  piece. Can you now maybe talk a little bit about   technology management as it relates to what you  were just describing, and also selecting the right   kinds of technologies and especially, selecting  the right kinds of data to match with the problems   that you're trying to ultimately address? I would add time. That is still relevant.  I think I have a good example for  you. When the pandemic broke out,  

I don't think anybody was really expecting  that. All of a sudden, organizations shifted   to almost exclusively working from home. There are laws about what data you can   access from home and what data you can  access at your desk. You have a different   firewall when you're working in the office  than you do when you're working at home.  You've got developers that used to be  co-located that are not co-located anymore. 

Organizations had to absorb all of that change  while still trying to serve their customers and,   in some cases, failure to do so  could have been like a death.  There is an urgency about this as  well. You can't take forever to do it.   And you have to have good discipline in place  so that when the unexpected happens (in the   middle of the other unexpected that was  already happening), you have the resiliency   to survive that and come out of that stronger. I'm not going to suggest (although I could) that   IBM is one of those organizations, but mature  organizations that get it can do that. We saw   a lot of organizations that weren't so mature not  getting it (in the middle of all that disruption).   So, it's a very big question you're asking. The example of the pandemic was particularly  

instructive, and I think it goes to your data  and AI questions earlier in the segment as well.  When the pandemic hit, in terms of being able to  run your business (for instance, make financial   forecasts, make forecasts about your supply  chain, about your procurement abilities,   et cetera), all the models that were  in play were essentially useless   because we had now embarked on a  situation that was completely new.   No matter what technology we had in there from  an AI standpoint or a model standpoint, it had   been trained in a completely different world. That was Anthony's point, right? It may not   be true now. [Laughter] In fact, what was true,   though, was if we were able to get the data,  accurate data, pristine data, into the hands   of the people who were running those different  departments, along with an overlay of what was   actually happening in the pandemic—where  Covid-19 was breaking, what were in the   incident reports in different areas—if you could  geographically then overlay that on what these   guys were working on (whether it be financial  forecasts or sales, which they expected to close,   or procurement sites that were in danger, and  things like that), they could make something out   of it and move forward with it. That's, I think,  also an instructive example of the relationship   between data and AI and how that plays out as  things really unfold that are truly unexpected. 

Let me draw first blood on saying something super  nerdy. There's a concept. I call it decision   elasticity. I kind of stole it from economics. How wrong can you be and still make the same   decision effectively? You don't have  to be perfect to make a decision.  Inderpal is talking about training. There's an  implication there that you have longitudinal data,   data from the past that you can project into a  near-term future that looks reasonably similar.   And you can measure the elasticity of your  decisions: How wrong are they? Then if they start   getting wronger and wronger (to coin a term),  then you can stop and re-examine those methods. 

The problem is, when you have something  completely disruptive, there is no data.   And the most dangerous situation you  can find yourself in is when the world   is changing faster than the data that describes  it. That's exactly where we were at that moment.  You can't just throw your hands up and say,  "Well, wait. When you have five years' worth   of data, come back and I'll retrain  everything and we'll be good to go."  You have to have methods in place  that are effective in a situation,   and that's what this environment taught  us. That you can't just rely on one type of   learning, one type of projection into the future. At that time, I was very involved with watching  

bad guys do bad things. Well, when  there's disruption, the best bad guys,   especially if they think they're being  watched, they change what they're doing.  If you model based on what they were doing,  you're modeling how the best ones are no   longer behaving. Kind of a dangerous  thing to do, right? But we know this. 

The flipside of that coin is if you know that  the environment changed such that the bad guys   are going to probably try to take advantage  of it, that many of them are probably going   to do that unartfully. And so, you may be  more easily able to see them as they run.  You turn on the light and the little creatures run  away. You can see that, and so there might be an   opportunity there along with that risk. Sometimes these situations,  

I would say, are almost never all bad or all good.  There's always something in it that can teach you.  There's always something in it that can make what  you're doing better if you have enough time to   breathe and observe what's going on and use the  energy in the best possible way. It doesn't mean   the bad thing will stop happening, but it may mean  that you emerge from it in a better way because   you took that time to be more thoughtful about it. We have a question from Twitter. Lisbeth Shaw   says, "The issues you're describing are true  of any business or technology transformation.   Are there particular points, issues, that are  more problematic for AI-enabled initiatives?   Can you kind of drill down into that?" If you look at the advent of AI,   the progression of AI, it's moved very,  very quickly in the consumer space   but not so fast in the business space.  That's because, in the business context,   people don't trust AI, and they  don't trust AI for multiple reasons. 

We talked about some of the issues about  the data, so the robustness of the data,   the quality of the data, the currency of the  data. Then we also get into issues that have to   do with the fairness of the algorithms, that  the results they produce are going to treat   people fairly (if the data pertains to people). You have the issue of privacy being invaded   in terms of the algorithms discovering something  new. There's the famous example, or infamous   example, of a lot of the retailers looking  at the shopping patterns and shopping data,   and then inferring that this person is pregnant  and mailing their home. It turns out to be a young   lady, and it was really a complete invasion  of her privacy. So, those aspects come in.  Then there are the issues around job displacement  and things of that nature. If you're applying AI  

in the enterprise, there are two flavors, often. There's the automation flavor, which has to do   with when things are kind of straightforward. You  go from one step to the other, you know what those   steps are, and you can automate all that, so  there's job displacement associated with that.  But even on the decision-making side where  the AI is helping make a decision, there's a   decisionmaker in play, and they have to trust it.  They have to say, "Well, this won't displace me." 

Extending that further, the executives, as you  put AI. We kind of know by now that the AI has   to be infused into the major workflows of the  business, things like procurement, supply chain,   et cetera. That's the kind of IP that doesn't  get published in papers or patented or anything.   Those are the trade secrets of a company. They have to be able to trust whoever the   vendor is of this software that this is not  something that's going to disintermediate.   Furthermore, the decisionmaker that's  working with the system has to understand it.  Years ago, I did this computer program called  Advanced Scout that ended up being used by every   coach in the NBA. I remember the first time it  had a counterintuitive finding; it basically asked  

the coach to play two backup players in a playoff  game that they were on the verge of elimination.  He was very concerned about that because  he felt, "If I do this and I lose,   I'm going to lose my job and reputation as  well, in addition to the series." We kind of   solved that problem by letting him see the video  clips of when those two players were on the court,   but that's the explanation piece. If you tell a doctor, "Amputate the   left leg," they're going to have all kinds  of questions. Okay, why amputate? What other  

options were considered? Why is amputation  the right one for this patient, et cetera?  Explanation is another big part of it, and  the AI systems today don't do a good job of   all that. Those are the special aspects  of AI and trust that come into play.  I think that was a fantastic list,  and I won't bane to add to it,   but I will suggest another dimension to it. Great question: How are the AI issues?   What's special about the AI issues? I would say another one is that you   have the opportunity to fail faster and at larger  scale. There's a tendency (once these sorts of   systems are implemented) someone says, "Well,  it's 99% accurate," "It's 92% accurate," "It's   87% accurate," and you assume that means that  87% of the time the prediction will be right. 

Well, no. That's based on the past, not the  future. Very rarely do we measure fast enough to   stop every conceivable bad thing from happening. Inderpal hinted at something, which is   an observer effect. People, when told  what to do by a "machine" will sometimes   think they know better or not want to be told what  to do by a machine and do something different just   because a machine told them to do it (to prove  that they can do something), and not necessarily   thinking it out loud like that. The question I get asked a lot is,  

"What about someday, will people be reporting  to robots or robotic bosses of some sort?"  You say, "Oh, of course not. I would  never do that." Then the GPS tells you   to turn left or right, and you do. Outlook  tells you to go to a meeting, and you go.  We're already taking a lot of  direction from automation. I  

won't call it AI necessarily but from automation. The human effect of what we do as human beings to   accept or reject that device is essential to  get at trustworthy AI, to get at making sure   that we don't marginalize others (that are already  marginalized) more because they don't have access   to these technologies. The concept of good and  not good kind of depends on where you're sitting   sometimes (whether it's good or not good). There are certainly lots of volumes, books,   committees focused on trustworthy AI  and explainability. There's legislation   as we speak being considered that  will hold the feet to the fire of   anyone who is implementing anything called AI. To say that it's not being adopted by business,  

the adoption is lower, I think, in some  ways because of some of these human   factors. It's not a lack of technology. It's a  reticence to just push that button so quickly.  Technology will always outpace regulation,  so you have to be careful, or you could   find yourself in a world of hurt where now  they're coming after you because you used   that technology that made a better decision.  Good luck trying to prove that sometimes.  This is a question from Hue Hoang, and  he says, "We can sometimes measure the   cost of implementing data solutions, but  how can we measure the operational costs   when a business decides not to implement certain  solutions such as governance or data quality?"  The opportunity cost, the  cost of not doing something.   Thank you, Hue, for that question. That's a  big one, and I think it's an important one.  If we're going to decide not to do something,  we should decide not to do it on purpose,   not just because we got tired of arguing  about it or because we didn't want to   take the effort to get all the data that  will be necessary to make that decision.  One of those annoying questions that I usually  bring into the conversation is if we're going to   decide not do this (because there's some other  thing that we want to do, and that other thing   has been deemed more important), great. Then  let's make that decision. But let's understand  

the opportunity cost, the cost of not doing it. In many cases, it does become clear-cut because   you might have regulations that then levy huge  fines. For instance, in the European Union. GDPR,   for instance, if you don't have the right  setup for governance and privacy and so forth,   you'll be hit by a major fine. In other cases, though,  

when they're making these decisions, they might  choose not to do the governance of the data,   but it'll end up reflecting in the actual  output that's being produced. Then somebody has   to go back and fix it, so keeping tabs on that. I'm assuming here that you've lost the argument   and they've gone ahead with it. Keeping tabs  on that and raising that every time it happens,   I think very quickly you'll be able to  make a difference in the way people are   viewing it because nobody wants a disaster. If they've skipped that step, which has major  

magnitude, sometimes that'll happen. Collaboration  is the name of the game, so you just want to keep   an eye on it, warn people that this is going  to happen, and every time it happens or even   before it happens, you raise your hand and say,  "Look. I told you about this. Now let's do it."  This is from Iavor Bojinov, who is a professor  at the Harvard Business School. He's also been   a guest on CXOTalk. He says this. Inderpal, maybe  I'll ask you first. "Is there anything different  

between generative AI and more traditional AI,  and how should organizations approach this?"  I think the best way to think about generative  AI, the promise is that you can do things   conversationally. Just as you and I can have a  conversation, and we can discuss something and   try to get to some resolution, that's the whole. Now, if you apply that in a large organization   and say, "I've got some intelligence that can  now conversationally help me do client support,   employee support, my IT operations, et  cetera," that's hugely, hugely promising.  On the other hand, the way these systems work  today, the best way to understand generative AI   that I've been able to get my mind around it is,  in a sense, each word is predicted. Then the word,   essentially, that word is fed back into  the input. Then the next word is predicted.  It's almost like when you and I are  talking, I'll sometimes do this. I'll  

go out on a limb. I'll start saying something,  and the thought hasn't fully formed. Usually,   I manage to come out of it. But many times,  I'll end up with my foot in my mouth.  The generative AI techniques are  essentially going out on a limb every time,   which is also why they're not always consistent  with the response. You know you might have the   same prompt give you a different response because  it's working off a probability distribution. 

I think there's a tremendous amount of  promise but also a tremendous amount of   work that needs to be done to address some  of the issues that we've raised earlier.  Anthony, differences between generative  AI and traditional AI and implications   for the enterprise – pretty quickly, please. Generative AI is making stuff that didn't exist   before based on stuff that it observed, and that  stuff can be hexed, it can be images, it can be   anything that we as humans consume. The challenge  to it is that you look at all this stuff in the  

past, and you kind of compute on it and do a lot  of math. Then you generate something that looks   like a human said it, and a human didn't say it. When the world changes and the corpus of data   that it's looking at didn't change fast  enough, that nuance gets lost, and we lose   the ability to understand something nuanced. So,  if the purpose is to provide customer support   based on frequently asked questions, or if the  purpose is to summarize a whole bunch of things   that you should have read and didn't have time,  it's a fantastic idea. If the purpose is to write   some new thought leadership on something, maybe  it's a starting point but I would be very careful   when we consider that to be an ending point. Share final thoughts on advice that you would give   to business and technology leaders who want to  be more effective using data, using AI. Inderpal,  

do you want to jump in with that one first? I've been doing this for the last 20,   25 years, starting from the days when I  did that program for the NBA to now. I've   always felt that whenever I was doing it, I  felt, "Oh, it can't get better than this,"   but it always seems to get better than that. I think we are now in one of those moments where   there is the potential and the opportunity to have  a tremendous impact not just on business but also   on society. I think, because of that implication,  that there are these major societal considerations   as well. We absolutely have to get involved, and  that would be my biggest advice to people either   on the business side or on the technology side. You need to really get involved with what's  

happening here. There's just tremendous,  tremendous potential. There's never been a   better time to be involved in data and AI. Anthony, it looks like you're going to   get the last word here. Number one, I would say,   ask "Why?" a lot. Why are we doing this?  What do we have to believe? Why this data?  Make sure that you understand before you jump into  that method with that data. Make sure that method  

and that data are, in some way, justifiable (not  only against what you intend to do but against   what you are not doing by doing that instead). Then the second thing is to make sure that   you pay very close attention to how  the environment is changing so that   you don't get caught by the change that  makes what made sense no longer sensible.  Then the last thing is something I always  advise, which is to be humble. It is extremely   rare when you know everything you need to  know and have all the information you need   without widening that circle and bringing in  others that have some sort of expertise or   some sort of perspective that you don't have. Inviting that expertise and that perspective   is not a sign of weakness.  It's a sign of great strength.  With that, unfortunately, we are out of time.  I just want to say a huge thank you to Anthony  

Scriffignano and Inderpal Bhandari. Anthony, thank  you. It's wonderful that you've bene here again,   and I hope you'll come back another time. Absolutely. Thank you so much, Michael.  Inderpal, I'm so honored that you joined us.  Again, I hope you as well will come back and   be a guest on CXOTalk again at another date. Delighted to do that, Michael. Thank you  

for having me. For those with  unanswered questions, please   link in and we can continue the conversation. Everybody, thank you for watching, especially   those folks who just ask such great questions.  You are such a smart and bright audience.  

We love your questions. Keep watching CXOTalk. Go to CXOTalk.com. Be sure to subscribe to our   YouTube channel and hit the subscribe button at  the bottom of our webpage. You can subscribe to   our newsletter, and we'll tell you and  notify you about our excellent upcoming   shows and guests. We have lots of them. Everybody, thank you so much. Hope you   have a great day, and we'll see  you again next time. Bye-bye.

2023-07-08 12:10

Show Video

Other news