Latest / CXOTalk / Renowned MIT/Stanford Computer Scientist on AI and Collective Intelligence | CXOTalk #909
Transcript
- 0:00Do they just set U the AI and walk away?
- 0:02If they do that, they lose a ton of money.
- 0:05Companies optimize AI for individual productivity, but is
- 0:09that really the right approach? Alex Sandy Pentland is a
- 0:13professor at MIT and Stanford with over 165,000 citations for
- 0:20his work on collective intelligence and trust.
- 0:24AI is in some ways never going to be as smart as people.
- 0:28AI is trained on backward looking data, so it can't really
- 0:32know what's going on in the future.
- 0:34So when it says here's an answer, it's just based on the
- 0:39past. And as we all know, past results
- 0:41do not justify future answers. That's one thing.
- 0:46The other thing is they're not context sensitive.
- 0:48It's like we live in this world and we see how people feel about
- 0:52things and, you know, all this sort of little incidental stuff.
- 0:55How do the young people feel that context is important and
- 0:59figuring out what's going to happen next?
- 1:01And AIS don't have any of that. And for instance, a way to think
- 1:06about this, so it's just sort of a bellwether, is look what
- 1:09people do in finance. Finance is very straightforward,
- 1:13whether you win or you lose. But what they do is they realize
- 1:18that all the data that they have is actually backward looking.
- 1:22It's usually overfit. So it doesn't predict the future
- 1:26very well. And so they have people in there
- 1:30adding their intuition. And in fact, that helps returns
- 1:34absolutely right. It is just not a question people
- 1:38plus AI is better than AI and it's better than people.
- 1:43And a lot of the changes come from sort of when things change
- 1:48or when there's risk, the people become really important.
- 1:51You have systemized thinking about this.
- 1:56Your latest book is called Shared Wisdom.
- 2:00You talk a lot about collective intelligence, or your term is
- 2:05community intelligence. How does that fit into this
- 2:09picture you're describing? We live in a world where we have
- 2:14this thing called culture. There's stuff that everybody
- 2:17knows, right? Some of it's right, some of it's
- 2:20not right, maybe it's not posed quite right.
- 2:23And then there's stuff we've heard about which is sort of
- 2:25broader context. And when you're running a
- 2:29company or an organization, you really want to have a culture
- 2:32that is compatible with what it is you're trying to do.
- 2:36You need to have people aware of all the things that are going
- 2:40around that might come and, you know, sideswipe you.
- 2:43So that is community intelligence.
- 2:45That's awareness of the whole sort of environment that you
- 2:49operate in. And the shared wisdom is the
- 2:51things that everybody believes, some of which might not be true.
- 2:56So you need to continually involve that.
- 2:59And the way you do that is you have you try things out, you see
- 3:03what they work, you have a discussion about it, you see
- 3:06what other people did. And it's this process of
- 3:09refining the the wisdom about what happens.
- 3:12It produces the culture. You might recognize this is
- 3:16science. That's really what science is.
- 3:18People do these experiments. They talk about it with each
- 3:21other. Eventually they figure out how
- 3:23to do things better and predict the future.
- 3:25And, and now you can just build things because you don't have
- 3:28these sort of rules about stuff until the rules change.
- 3:33But you know. So.
- 3:34That's the sort of thing that a company is meant to do, is it
- 3:38has to have a culture and a sort of a community intelligence that
- 3:44fits with what's actually happening.
- 3:46So you can predict where to put your efforts and what to be able
- 3:50to do. And AI can help you with that,
- 3:53but it can't replace that anytime in the near future.
- 3:57How does AI overlay onto this these concepts of shared wisdom,
- 4:05community intelligence and the culture you're describing?
- 4:09Because so far what what you've said is not that different from
- 4:16issues of integrating technology into organizations going back to
- 4:21the Stone Age or at least till the early days of computers and
- 4:24enterprise software. Let me give you an example.
- 4:27So I was on a panel with the CT OS of some very major
- 4:31corporations and we're talking about the real problem is
- 4:35getting the right data, but that's for later in this this
- 4:39discussion. The thing that they sort of said
- 4:43that they had done for everybody, and I've since seen
- 4:45other people doing it, is they created AI buddies for every
- 4:49employee. So what is an AI buddy?
- 4:51An AI buddy is something that's read all those manuals and all
- 4:55the reports and the newsletter and knows what the people down
- 4:58the hall are doing and the people in another.
- 5:00And so you can use it to keep you in the loop, to provide
- 5:05context for your decisions, to know what people do, what works,
- 5:09what doesn't work. And that's really different than
- 5:12just something that, you know, produces a number or some other
- 5:17sort of thing. People don't read those.
- 5:19Manuals. People can't read all the
- 5:21different sort of reports out there, but the AI can and it can
- 5:24pick out the things that might be relevant for you.
- 5:27So it's like having a really brilliant librarian perhaps, or
- 5:32something like that to sort of help you stay.
- 5:34In the loop. That's really critical.
- 5:37A second thing that AI can do, and I talk about this in the
- 5:40book, there's this sort of, you know, personal things, there's
- 5:44the. Group discussions The strength
- 5:46of. Humans is combining views and
- 5:49battle testing them against each other.
- 5:51And you know, everybody hates meetings.
- 5:54Why? Because we get distracted and
- 5:57there's all sorts of weird. Stuff and power dynamics.
- 6:00And so we've built tools using AI and you can look at
- 6:04deliberation IO, all free, all open, where we built little
- 6:09tools that essentially just reflect.
- 6:11They don't get answers. They say, I hear people saying
- 6:14this and that and what that little sneaky sort of reflection
- 6:18does is it keep people focused on the problem and it makes sure
- 6:22that the loud voices don't dominate as much as they tend to
- 6:26do. And so this little tiny thing,
- 6:29it's just like a mediator, except it hasn't got it's not
- 6:33human right. It's just AI makes people's
- 6:37medians more than twice as efficient and effective
- 6:41according to the people than having just humans.
- 6:45Think about that. If you could like raise the
- 6:48efficiency and the accuracy and performance of your meetings by
- 6:52a factor of 2, you'd be golden, right?
- 6:56And it turns out that's a very simple sort of AI thing to do.
- 6:59This results on on their summarization.
- 7:01And then the third thing that people need to do is people need
- 7:05to do what I think of as exploration.
- 7:08You know, it's a big world out there.
- 7:10People are trying lots of things.
- 7:11Some of the things are pretty interesting for what you're
- 7:14doing. Maybe not exactly the same, but
- 7:17you ought to hear about them. And we're always here finding
- 7:20things where, oh really, they're doing that.
- 7:23That's interesting. Maybe if we did it some other
- 7:26slightly different way, we could do that too.
- 7:28That sort of exploration of the world is something AI is just
- 7:32really good at. And so using it for exploration,
- 7:36using it for finding consensus and battle testing ideas, and
- 7:41then using it for people, keeping people in the loop are
- 7:45wonderful places where it doesn't detract from the human
- 7:48intelligence, it augments the human intelligence.
- 7:52Yeah, some tasks will get more productive, That's all fine.
- 7:57But actually what happens is that the tasks will become much
- 8:00more complementary and reinforce each other.
- 8:02Instead of having a bunch of people off doing things
- 8:05separately, you now have a team, a great big team that's like
- 8:09look into the future looking to side to side and making
- 8:13decisions and acting together. And I think everybody knows that
- 8:17that's what a leader has to do, right?
- 8:20To get everybody going in the same direction and pick the
- 8:23right direction. You're really talking about
- 8:26using AI far beyond efficiency. It's not just a matter of
- 8:33executing a, a particular process more quickly, but really
- 8:39into the heart of innovation. We have some questions that are
- 8:44coming in now on LinkedIn and Twitter X.
- 8:48And I just want to remind everybody that right now you can
- 8:53ask your questions and I urge you to do so.
- 8:57Take advantage of this opportunity to ask Sandy
- 9:01Pentland, you know, pretty much whatever you want.
- 9:04So take advantage of it. Let's get some of those
- 9:07questions. And right now we have a question
- 9:11from Anthony Scrifignano, who says, can you talk about how
- 9:18humans are adapting the way they formulate questions to
- 9:22accommodate AI and how that adaptation may be subtly
- 9:28influencing innovation? The way I see the way the data
- 9:33sort of show people using it is, is yes, there using it.
- 9:39The way to to do this sort of like exploration and battle
- 9:45testing of ideas. So normally people are used to
- 9:48looking things up or asking Google or something like that.
- 9:52And now you have a much more interactive medium to do that
- 9:56against. And so that changes things in
- 9:59the experience I've seen that changes things for the better
- 10:03because what you're doing now is you're training the person to be
- 10:07in some sense skeptical. They're asking about questions.
- 10:10They're like prying, like, well, why do you think that?
- 10:13Or why is this true? Or is there anything?
- 10:15So they're being much more engaged with the problem and
- 10:20much more questioning of it because the medium, the AI
- 10:23medium is much more responsive in this sort of way.
- 10:27And that's really the right way to be using it because you don't
- 10:31want to trust the answers. It's not a matter of
- 10:33hallucination. It's a matter of does it have
- 10:36all the data? Does it know what's going to
- 10:38happen the fast? No, it doesn't.
- 10:40All right. So, so you want to be in there
- 10:43and use it as a way to query the world, to query what you know.
- 10:49And if you have that sort of, you know, exploration mindset,
- 10:54that's different than the way people usually do things.
- 10:57We have another really interesting question from
- 11:01Twitter, from X, from Arsalan Khan, who says culture in some
- 11:06ways is a reflection of executive leadership and the
- 11:10incentives and disincentives. If AI fails, should we just
- 11:17blame the executives? Or I think also more to the more
- 11:21to the more to the point, how do we assign blame between the
- 11:26executives and the AI? Culture is actually something
- 11:30that's Co created and leadership has outside outsize influence
- 11:36because they have control of things like incentives and so
- 11:41forth. And plus people assume that they
- 11:44have more data, more vision, broader vision than than they
- 11:48might have. One of the things about AI is
- 11:51then you can have individuals, not just the leaders, have very
- 11:55sort of broad vision, which means the discussion is not as
- 11:58top down as it used to be. I think you need to regard
- 12:03culture as Co created where the leader is sort of the conductor
- 12:09of the orchestra, but actually the sound comes from the
- 12:13orchestra and it's that that is the final sort of product.
- 12:17So I think that notion of of blame is probably the wrong way
- 12:22to conceptualize it. Yes, people will do that, but
- 12:25the way to do it is is are we making good music together?
- 12:28OK. And if the leader is like
- 12:30screwing things up by junking people in different directions
- 12:34and going against what other people think and not explaining
- 12:38so that they can get on board, then the leader needs to change
- 12:42what they're doing or get out of the way.
- 12:45I don't know. It's a matter of blame per SE,
- 12:48but it's a matter of, you know, again, think of the conductor of
- 12:52the orchestra. If their style doesn't match the
- 12:54style of the orchestra, you got a problem.
- 12:57But what happens as AI gains greater quote UN quote
- 13:05intelligence? Does the AI and of course the AI
- 13:09is trained on all kinds of different sets of data.
- 13:14And So what happens to the culture that is then Co created
- 13:19by the human masters and the AI team?
- 13:24You were talking earlier about AI buddies.
- 13:27AI is trained on the past, not the future.
- 13:30It lives in a box, not out in the world, which means it's less
- 13:34sensitive to context and current events.
- 13:36It doesn't really have a notion of causality what causes extra
- 13:41they. It has notions of what people
- 13:43say, but we all are actually are able to predict.
- 13:46The future, at least in the short term, better than the AI,
- 13:51right? And, and particularly in groups,
- 13:54this is just absolutely quantitatively true.
- 13:57So I think, you know, one thing to keep remembering this is what
- 14:01do the quants do? What do the hedge funds do?
- 14:04Do they just like set up the AI and walk away?
- 14:07And the answer is that if they do that, they lose a ton of
- 14:10money. Why is that?
- 14:12Well, because of the back testing back view, because of
- 14:16the lack of context, because of, you know, the notion of knowing
- 14:20what causes what and those are the hottest topics in finance,
- 14:25causality things that the AI can't really do it all.
- 14:28You can't do it without experimentation.
- 14:31So anyway, so, so to go back to the to the question, if
- 14:36everybody has this sort of, OK, we're querying the future, we're
- 14:40querying the facts. If you have new facts, bring
- 14:43them in and contribute it, that'll be great.
- 14:46You know, the the leader can provoke a discussion about
- 14:51longer term tanks. So this is one of the main
- 14:53things that AI allows you to do that we don't do enough of
- 14:57today, which is we often talk about the day-to-day things and
- 15:01what are the bugs and happened. But a leader has to be able to
- 15:04say, yeah, but we're going in four or five years to this place
- 15:09and actually they ought to be thinking about 20 years, 40
- 15:13years. Other parts of the world
- 15:15actually do this. So for instance, if you look at
- 15:17investment funds in China, 40 year time horizons, if you look
- 15:21at investment funds in the Middle East, 40 year time
- 15:25horizons, they're asking we're on this journey, which
- 15:28directions are we headed? And, and people need to have
- 15:32that in the culture. They need to understand that,
- 15:36you know, we're walking in this direction and yeah, we're
- 15:39playing with today's problems and so forth.
- 15:41But the long distance sort of future is over there and and,
- 15:46you know, we're all going to try and make choices that head in,
- 15:49in whatever that direction. It what would be a good example?
- 15:52Say you make products of very sort or have services, but one
- 15:56of the things you know, is that the population is getting older
- 15:59on average, much larger aging. OK.
- 16:03Well, if your services are relative to helping people age
- 16:07in place or people getting more mature, then you ought to think
- 16:11about, OK, so 20 years from now, these are going to be the
- 16:14demographics, the markets going to be huge, maybe plus or minus
- 16:19you what you feel about that. But that's an incredible
- 16:22opportunity, and you know that's going to happen.
- 16:25And the question is, are the directions you're taking in that
- 16:28direction so that you can capture that sort of opportunity
- 16:32or are you marching in another direction?
- 16:34It's not going to really change your day-to-day so much, but it
- 16:39is going to change the direction you step in at least each little
- 16:43decision that you make. And so the long range, that's
- 16:47this exploration, which AI is pretty good at, right?
- 16:51The discussion of what direction to go.
- 16:54Well, that's like this deliberation dot IO thing where
- 16:57it's groups of people, you know, battle testing ideas with each
- 17:00other. And then the AI buddies is
- 17:03what's going on right now and what can I do that's going to be
- 17:07the best decision. What are the problems and the
- 17:10downfalls the potential downfalls of doing so?
- 17:14There's so much we don't know that's going on, particularly in
- 17:18an organization, You know, what is everybody else doing?
- 17:22And I'll, I'll tell you a little story about this later about
- 17:25what some of the world's largest organizations are doing about
- 17:28this. But when you do that, when you
- 17:31queer the AI, you have to have a query mentality, an explorer's
- 17:36mentality. You really shouldn't believe
- 17:39what it says without having some sort of check, either internal
- 17:44or external. It's it's a brainstorming buddy.
- 17:48It's not always correct. It doesn't know about the things
- 17:51that are happening today. It doesn't have an understanding
- 17:54of the human sort of side of the context, but this is the story
- 17:59that about what some large organizations are doing.
- 18:03They're realizing that this sort of AI buddy giving you a sense
- 18:09of context for what everybody in the organization is doing is
- 18:12perfect for work from home. A big problem with work from
- 18:16home is you don't know what the, what the sort of cause of the
- 18:19day is and what other people are doing or distance offices,
- 18:22offices in other countries. So having offices that are in
- 18:26different countries that talk to the actual customers, their
- 18:29person, you know, on a more personal basis is great, except
- 18:33that they don't know how everybody else is doing it.
- 18:36And there's no good way to spread ideas.
- 18:39This is what AI buttons are perfect for.
- 18:41It's keeping everybody in the loop.
- 18:43And what I just said was, for instance, here's one example of
- 18:48a leader of a, a large multinational 350,000 employees,
- 18:54but he says this is that he's going to move down to 150,000
- 18:58employees that are in house, but he's going to double the size of
- 19:03the other sorts of employees that.
- 19:06So he'll actually be hiring 100,000 more people, but they'll
- 19:11be based on projects. They're not in house employees.
- 19:15They're things in different countries, working with
- 19:17different sort of customers and so forth.
- 19:20And So what he's describing is a company that's much more
- 19:23distributed, that's much more relating to its customers, that
- 19:29has these electronic AI sort of aids keep its actions
- 19:33coordinated and keep everybody on the same page.
- 19:36And I think that's sort of the biggest story that's out there.
- 19:40Companies, rather than having these big sort of hierarchical
- 19:43structures in silos, are going to be much flatter.
- 19:48Teams are going to be much more agile.
- 19:49The teams are going to be smaller.
- 19:51From the view of sort of information and context, silos
- 19:55in companies, in layers of management are exactly what you
- 19:58don't want. They're sort of a recipe for A
- 20:01doesn't know what B is doing right?
- 20:04And the AI can help you sort of blend those and make people much
- 20:07more aware. So many of the leaders that I
- 20:11talked to are envisioning the future.
- 20:12Where? Most groups are 5-6 people of
- 20:17different backgrounds so that they can talk to each other
- 20:21about different perspectives using AI and they're relatively
- 20:25autonomous. They're not strict orders from
- 20:28on by the organization is in some ways much more like a
- 20:33venture capital or or actually more accurately like a incubator
- 20:40where yes, the organizations get funding from the top.
- 20:43They have to report back to the top, but the top gives them a
- 20:46great deal of autonomy and they're able to go out and then
- 20:50really act. The thing that makes that
- 20:52possible is that now six people can control a business, set up
- 20:58and control a business that's worth billions of dollars a
- 21:01year. Like you can set up a finance
- 21:03company, a payment company with a small number of people and
- 21:08that's amazing. You never used to be able to do
- 21:10that before, or a logistics company or any of a number of
- 21:14other things where you're getting teams together very
- 21:17quickly, producing things, software becomes very cheap.
- 21:22So I mean, I'll just give you one story is from a friend of
- 21:25mine. This was a doctor in a hospital,
- 21:28no technical background. He saw that the hospital was
- 21:31just, you know, people were not coordinating with each other.
- 21:34And of course, bad health things come.
- 21:36So this doctor with no technical background got a couple of AI
- 21:41tools, a friend or two that knew a little more about it.
- 21:44And in six weeks, they built an operating system for the
- 21:47hospital that the hospital ended up adopting and having much
- 21:52greater results and much better efficiency.
- 21:56So look, if a small group like that can do something as
- 21:59complicated as a hospital, and that's a real example, then
- 22:03think about what's going to happen in the future.
- 22:06AI is about bringing lots of data together, making some sense
- 22:10of it, and then causing things to happen.
- 22:13And if you have a good group of people that are working together
- 22:17and sort of trading perspectives and things like that, you can
- 22:20make magic happen. Folks, I encourage you ask your
- 22:25questions folks on LinkedIn. All these questions are coming
- 22:28from Twitter so far. Where are you, you guys are
- 22:30watching? Ask your questions.
- 22:32And oh, by the way, subscribe to the CXO Talk newsletter so we
- 22:36can notify you about upcoming shows.
- 22:38Just go to cxotalk.com and subscribe.
- 22:42You'll be very happy you did that.
- 22:44Anyway, Sandy, you're, you're describing what we could call an
- 22:50agentic AI future where we have agents and and small tool tools
- 22:58doing hundreds, thousands of little tasks and the efficiency
- 23:03gains that come from that and the rapid adaptation.
- 23:09But along the way, won't there be a tremendous amount of job
- 23:17shifting, relocations and job displacements which will
- 23:22translate to a very painful intermediate period of time?
- 23:27If you get this sort of change to lots of small groups doing
- 23:31things, maybe they have the same brand, they're all price for
- 23:34Waterhouse Cooper or something like that, but it's actually
- 23:37lots of independent things that are funded and and have some
- 23:40cooperation agreements. Then there's going to be a lot
- 23:43of churn. The main thing that will happen
- 23:46from a business point of view is incredible agility and ability
- 23:50to do things and turn on on a dime.
- 23:54From a a worker point of view, an employee point of view, that
- 23:58means you're going to be changing what you do all of the
- 24:00time. And unfortunately, in American
- 24:04Society at least, we've tied things like healthcare and
- 24:07retirement to employment. And we did that back in the
- 24:111950s when lifetime employment was not a joke.
- 24:15And it's a mistake. We need to change it in some
- 24:18way. An idea I have and have been
- 24:22promoting and it's, it's sort of interesting is if when you set
- 24:28up a company, you were required to give 10% of the shares.
- 24:33And remember at the beginning with a company, the shares are
- 24:36worth 0 into a sovereign wealth fund then, and in return you
- 24:42would get guarantees about future taxation and stuff.
- 24:45People would go for it. Now, if we had that in place,
- 24:49say in 1990, we'd have a sovereign wealth fund worth
- 24:53something like $10 trillion today in the United States.
- 24:58And we wouldn't be having a lot of arguments about social safety
- 25:01net because $10 trillion can support a lot of people, right,
- 25:06just from the, the, the interest.
- 25:08So there are ways of distributing the opportunity,
- 25:12the success and the failure in a way where everybody has a good
- 25:18minimum. We're not raising taxes.
- 25:20What we're doing is we're having everybody own a piece of the
- 25:24pie. And I think that's a really
- 25:26interesting, I was just, I'm here in the UK and that idea is
- 25:31really interesting to the government here because I have
- 25:34the inequality problem. They're facing all these sorts
- 25:37of problems with AI and they have the problem, how do you
- 25:41keep people safe in the UK here? They still have this history of
- 25:45guilds. So you could imagine having
- 25:47professional societies and guilds and those are the things
- 25:50that run a fund for the people like you.
- 25:54So if you're a doctor, you join the doctors Guild and you put a
- 25:59bit of your income in there every time.
- 26:01And then all that goes to all the doctors.
- 26:04Many different ways to do this by aggregating opportunity,
- 26:09aggregating risk the way that for instance investment funds do
- 26:13and which we don't do today because we haven't really
- 26:17changed the the social contract since about 19501960.
- 26:23I have to say we started this conversation talking about
- 26:29shared wisdom and community intelligence.
- 26:33We moved and and much of that also improved and driven by AI.
- 26:38We then moved to the impact on the way organizations are
- 26:45constructed, workers created and you've now brought in this, if
- 26:50we can follow the chain, essentially saying AI is going
- 26:54to drive really basic changes in the organization of society,
- 27:01the. Place that is most visible is an
- 27:04education. I mean, we have this education
- 27:07system, I'm part of it. You know, it's based on a
- 27:09template that comes from 1000 years ago and really hasn't
- 27:13substantially changed because knowledge wasn't very dynamic
- 27:17and the world was pretty static in terms of a chosen career.
- 27:21You stuck with it your whole life.
- 27:23Lifetime employment was a real thing up until not too long ago.
- 27:28And So what are we going to do with AI in classrooms?
- 27:31And let me give you an example of something that I was part of.
- 27:37There's a small country in Latin America, Costa Rica.
- 27:40They realized that they were going to be growing coffee beans
- 27:44forever unless they could up their game and become have a
- 27:47population that was technologically sophisticated.
- 27:51And they made an investment. They gave every school kid a
- 27:54computer. This is back in the like 70s,
- 27:57right? And they insisted that they
- 28:00speak the technical language of the world, which was English.
- 28:04So this is Spanish speaking, coffee growing region deciding
- 28:09that every kid is going to be tech savvy and be able to
- 28:12communicate in English. And as a consequence, they got
- 28:15Intel to build a fab plant there.
- 28:17They got software. They're they're doing quite
- 28:20well, thank you. Because they upskilled the
- 28:23population this way. What would be the analogy in the
- 28:26United States for us as well? Why don't we give kids in 6th
- 28:32grade laptops with AI and ask them as groups, not individually
- 28:39to, to build software that does things that they think are
- 28:43important, you know, takes care of old people, makes traffic
- 28:46better. And what they're going to learn
- 28:48is they're going to learn how to work with each other in the
- 28:51presence of these IT tools. And we see that, for instance,
- 28:56I, I helped shape the Media Lab at MIT.
- 29:00That's what it is. It's not an educational program
- 29:03in the traditional way. It's an apprenticeship program.
- 29:06You learn by building things and by sort of pressure testing them
- 29:11against company sponsors, against the real world and
- 29:14various sorts of things. And in that you don't just learn
- 29:18about the tools, you learn about the business, you learn about
- 29:20the context, you learn about the social issues of what you build.
- 29:24And you need to have all of those not siloed into a degree
- 29:28in X or a degree in Y. They have to come together and
- 29:32you have to be able to use the technical tools to be able to,
- 29:36to essentially dance between them and create a new future.
- 29:41And we've done pretty well. I'll brag, humble brag or brag a
- 29:45little bit. We had, I ran a class on
- 29:48entrepreneurship where the challenge was you have to have
- 29:52to a small group that uses these sort of AI tools and these are
- 29:56older AI tools. Now that creates a business that
- 29:59touches a billion people in the world, right?
- 30:02And those billion people have to be outside of the rich
- 30:05countries. They have to be in the poor
- 30:06countries. And guess what?
- 30:08We have a couple people who did it.
- 30:11It's like, that's pretty amazing.
- 30:14So, so these kids can really do things if they learn to, you
- 30:18know, combine the problem area with the tools, with the social
- 30:23organization to work in a team and the passion to go out and
- 30:27make it happen. And in the end, that's what we
- 30:30need more of. I will mention the CEO of Zoho,
- 30:35Sridhar Vembu, who's been a guest on CXO Talk a number of
- 30:39times, has a concept called transnational localism combined
- 30:45with what I've called when I've spoken with him, social
- 30:50capitalism, using Zoho as an agent of change to support
- 30:58technology upskilling in local communities.
- 31:03And I, I will say that they have put their money where their
- 31:07mouth is. I mean, Zoho has invested in
- 31:09this and in a number of communities around the world.
- 31:12It's pretty amazing, actually. It needs to be the shared wisdom
- 31:16that that's the way you do education, not drill and kill.
- 31:20And currently almost everybody does drill and kill.
- 31:24So that sort of notion of having a creative education, an
- 31:28education that actually builds things and brings in lots of
- 31:32disciplines and lots of tools. As I think sort of the key
- 31:35thing, it does get us, if you want to go there into what
- 31:40people are doing outside of the US, because what they're doing
- 31:44is very different than the way the US thinks about it.
- 31:47We have a question that came in early from Chris Peterson on
- 31:52Twitter. And I know there's some
- 31:53questions on LinkedIn, and we'll get to you guys next.
- 31:56And Chris Peterson says, do you see LLMS, GPTS, Gen.
- 32:02AI, fulfilling the promises made by their boosters?
- 32:06Or will AI pivot again to include other, maybe older forms
- 32:12to round out their actual capabilities?
- 32:16All of the sort of modern technology, the LLMS and GPTS
- 32:19and stuff are really pretty good at being speech interfaces,
- 32:25summarizers, librarians. There's a number of things that
- 32:29they do surprisingly well. They're basically looking for
- 32:33long term correlations and relationships that that are so
- 32:37vast in terms of the data and the distances that, that humans
- 32:42just can't see them basically. But they're not reliable in
- 32:46various ways. These are correlations.
- 32:48They're looking backward and so forth.
- 32:49So there's really a, a great need for some of the older tools
- 32:55to be part of the mix. The a key thing is something
- 33:00that that we're running called Loyal agents.org.
- 33:04It's with Consumer Reports, what we're and the law school at
- 33:07Stanford and what we're doing is we're saying, look, if you're
- 33:10going to have an agent buying things for you, it had better
- 33:13represent you and not some other company.
- 33:17All right. I mean, this is a basic for
- 33:19doing handling money and they don't do that.
- 33:23Also, it ought to be auditable. I mean, if somebody says, oh,
- 33:25you're self dealing or you're biased, you have to be able to
- 33:29audit that and answer it. All right, So those are places
- 33:33where more traditional machine learning or expert system type
- 33:37of technologies and other things like that have have a real
- 33:41place. If you take something like law,
- 33:43which is really hot area right now, law is a big expert system,
- 33:46OK. But finding which pieces of law
- 33:49are likely to be interesting is a good LLM problem.
- 33:54So, so it's a mixture of the two things to be able to ask, is
- 33:58what I'm doing legal right now or am I going to get in real
- 34:02trouble? Same thing with payments of
- 34:04various sorts. Is this something that is likely
- 34:08to end up poorly or something that ends up better?
- 34:11And so I think that there's lots of room for growth.
- 34:14But in the technology and evolution, the main thing that
- 34:21is I see as sort of the future of this is, yes, there'll be
- 34:25this evolution and so forth, but all AI is driven on data.
- 34:29All AI is is patterns and data. And so increasingly access to
- 34:35data is going to be the thing that the battle is about.
- 34:39And you said we wouldn't talk about international, but I'll
- 34:43tell you that's what the international people are doing.
- 34:46They want to run all the systems in the world so they have access
- 34:50to all the data in the world. And I think that that's
- 34:54something you really need to think about is if you have the
- 34:56data, the technology will get there to be able to use it in
- 35:00ways that are interesting, but if you don't have the data, you
- 35:03can't get there. An interesting statistic is
- 35:06something that the people ought to pay attention to.
- 35:08There was just a study of all the open source LLMS and what
- 35:13they found is for a given level of performance, the cost of
- 35:17doing that, where the current state-of-the-art LLMS fell by
- 35:2250% every 3 1/2 months and by a factor of 10 every year.
- 35:30So that means in the last three years, the cost of a given
- 35:34amount of intelligence has dropped by order of 1000.
- 35:39So when people talk about all the electricity, when they talk
- 35:42about like all the cost and stuff like that, they're
- 35:45forgetting that this stuff is moving faster than anything that
- 35:49has ever been happened before in terms of technology.
- 35:52And so you're, you're already seeing things that run on big
- 35:56laptops, sort of experimental high end phones, things like
- 36:00that. That's really interesting
- 36:02because that makes it really ubiquitous.
- 36:06It also brings The Who owns the data question really to the
- 36:09fore. But one of the things that's
- 36:12really interesting, I have a, a friend that runs the nation's
- 36:17largest super computer center and of course they do all this
- 36:21sort of AI type stuff. And he says that the moment they
- 36:24were able to get these LLMS to run on super high end laptops,
- 36:29their productivity tripled. And the reason was, is they
- 36:33didn't have to do all the cloud setup and configuration and
- 36:36fight for time. And it's like, you can just like
- 36:39do it. And that's something to think
- 36:41about, right? So what happens when personal AI
- 36:45is available? So control of data that's
- 36:48proprietary or personal, it's on every sort of device, that's
- 36:52going to be a really different world than what we're thinking
- 36:55about with, you know, Claude or something like that.
- 36:58Let's take a pragmatic organizational question from
- 37:06LinkedIn from Kapil Choudhary, who asked an important
- 37:11questions. An important question.
- 37:13He says leaders understand that AI value requires workflow and
- 37:19process redesign, but many hesitate because they lack
- 37:23concrete examples or frameworks. What practical patterns or
- 37:28starting points would you recommend for organizations
- 37:31trying to reimagine work with AI?
- 37:37And I love this question because it's really starting to build a
- 37:40bridge between the vision, Sandy, that you've just been
- 37:44describing about describing. For most organizations, you want
- 37:48to start with things that don't redesign the process.
- 37:51They just make the process better, less error prone, more
- 37:55consistent. It proves the culture about it.
- 37:57But of course, the real gains come from redesigning the
- 38:01process. And I think that a good way to
- 38:04sort of think about the process is where is the data, right?
- 38:08If there's data that aren't talking to each other, then
- 38:12that's an opportunity because the AI can combine the two data
- 38:16sets and find all these things are closely related.
- 38:20If you find that there are silos that have to do with each other,
- 38:24then the right thing to do is to begin to bring talks between the
- 38:28people who represent those salary, those silos, and the AIS
- 38:33and ask what are the opportunities for reorganizing
- 38:36this? And this is where the human
- 38:37dimension comes in is you have to make it clear to all the
- 38:40people in the room that if they can reinvent a great process,
- 38:45then they're going to be rewarded.
- 38:46They're not going to be out of a job.
- 38:48So talking about getting rid of people is probably not the right
- 38:53way to start a conversation like that.
- 38:55You want to say, OK, well, we have these data to sell as these
- 38:58data streams. How can we put them together to
- 39:01build better processes? And what sort of things can we
- 39:04cut out that are like, I've always been a thorn in our our
- 39:08side and move those people into these other sort of processes.
- 39:13It's often sort of difficult. I was on the data board for
- 39:17AT&T. You know, they have this problem
- 39:20that, you know, 100 year old company, everybody was set in
- 39:23their ways and they viewed that they had to now deal with people
- 39:27like Google who are going to be competitors.
- 39:30And so they put in place also upskilling programs that
- 39:35resulted in, you know, people getting applied degrees that
- 39:38gave people early retirement if they didn't want to do that.
- 39:42And they tried different organizations in different parts
- 39:47of the company to see what worked with their culture.
- 39:51And I think all of those are sort of practical things that
- 39:56realistically you're going to have to do.
- 39:57You have to experiment. There's no one answer to these
- 40:00things. It depends on your culture, your
- 40:02business area, where you reside, things like that.
- 40:06This is an interesting question from Lizbeth Shaw, who says
- 40:10social media has a history of magnifying and spreading
- 40:14misinformation and disinformation.
- 40:17Generative AI has history of sketchy information.
- 40:22How do you prevent corrupted shared wisdom and community
- 40:28intelligence in this environment?
- 40:30And I will just mention for people listening that we have
- 40:34coming up, I, I think in April 2 researchers who are going to be
- 40:39discussing the impact of AI swarms on misinformation,
- 40:43disinformation and democracy. So this this exact issue.
- 40:47So what do we do about it, Sandy?
- 40:49There's at least two major sort of things.
- 40:54One is as you need to have a little bit of reform for the way
- 40:58the Internet works, where you can tell that things are bots,
- 41:02all right. And so we run a group that's in
- 41:06the IETF, which are the ones that set the specs for the
- 41:10Internet to make it easier to identify things and, and harder
- 41:16to, to do bad things. And we're feel pretty good about
- 41:19ourselves 'cause we're just finishing a spec that's going to
- 41:22cost North Korea about a billion dollars a year, 'cause they
- 41:26steal things. They're not just disinformation,
- 41:29they're stealing things and little changes about like where
- 41:32did this data come from? So you have provenance make all
- 41:36the difference in terms of, of preventing things like this.
- 41:39So, you know, it's an active area of research.
- 41:42It's also something that people are beginning to realize is, is
- 41:46the highest sort of priority because we're seeing, you
- 41:50mentioned some bad things, worse things are coming.
- 41:52So I was just at a meeting with the heads of some of the very
- 41:57largest financial management companies in the world and
- 42:01they're scared to death. Just imagine that if all of this
- 42:06stuff is about fraud and stealing money and stuff forth.
- 42:09Oh my God. So so one is I'm hopeful that
- 42:13changes will happen there to make it harder for the swarms to
- 42:17attack and easier to defend against them.
- 42:19But that doesn't cure the main thing with social media, which
- 42:24is you get these like eruptions of beliefs that are, are false
- 42:29because it's all based on clicks.
- 42:31You know, more clicks, more money, OK, you're going to say
- 42:35something crazy, You get lots of clicks.
- 42:37OK, So the, the crazy people are the ones that that get all the
- 42:41clicks. And the more crazy you are, the
- 42:43more clicks you get. There are ways of doing this.
- 42:46So for instance, if you look at a thing called Paulus that that
- 42:52Taiwan uses for government conversations, government, how
- 42:56do we run our country? They have ways to sort of
- 42:59visualize what the range of conversation is.
- 43:04And what that does is it lets you identify the crazy people
- 43:08and say, well, that's he's saying that, but he's one of the
- 43:10crazy people. We have a version of that called
- 43:13deliberation IO, all open source.
- 43:16You can look at it studies, blah, blah, blah.
- 43:18It does an amazingly good job at this stuff.
- 43:22But the the big social media companies have begun doing
- 43:27something called community notes.
- 43:29So community notes is a very simple sort of AI system.
- 43:34And when somebody makes a comment that's one of these
- 43:36like, Oh my God types of things, people can begin adding comment,
- 43:41their sort of judgement to it. And when you get comments from a
- 43:45broad enough audience, right, not just the the echo chamber,
- 43:49but a broad audience, you begin to see notes attached to these
- 43:54comments that say, yeah, but nobody else believes.
- 43:58And so it gives you these sort of warning signals and it
- 44:01actually reduces things quite a bit.
- 44:04The part that's missing is that the people who do this all the
- 44:07time don't have a reputation score, You know, in the real
- 44:11world where we talk to people, we know who to listen to and who
- 44:14not to cause that guy always says crazy stuff, right?
- 44:17Or she's always a little off or not whatever.
- 44:20We don't have that in social media.
- 44:21And that's one of the things that is I think a reform that
- 44:24needs to happen. So social media needs to be a
- 44:27little bit more like face to face communication, reputation,
- 44:31community intelligence shared in wisdom about it is what the,
- 44:36the, the community notes does. And that can help a lot, you
- 44:41know, the, the, the, the things like the Taiwan Palace or our
- 44:47deliberation, I know is really sort of the way to do it because
- 44:50it doesn't have this instability where crazy comments get lots of
- 44:55attention, right? That's the thing that you have
- 44:58to watch out for because that's the thing that causes all this
- 45:01sort of BS going on. But of course you're fighting
- 45:05the core business model as you describe, which is clicks and
- 45:09attention from some of the most profitable companies on the face
- 45:14of the planet. Well, the studies that have been
- 45:17done so far is the things like community notes, which is
- 45:19actually deployed, it's actually out there.
- 45:22Facebook actually has that it. It's a minor, tiny little
- 45:27feature. Yeah, yeah, yeah, yeah.
- 45:28So they haven't like fixed at all.
- 45:30But on the other hand, that little thing does have a
- 45:33significant effect and it doesn't reduce their profit.
- 45:37So because people get involved in commenting with each other,
- 45:39right, The right way to do it is to have something that's a
- 45:42different sort of platform that doesn't have that explosive
- 45:45characteristic. I don't know how to do that
- 45:48business problem, but one hopes that some of us, some of the
- 45:55creative sort of people will. We've seen a number of things
- 45:59that are promising there, like for instance, a post processor,
- 46:04This is like an AI agent. So it looks at your stream and
- 46:08what it does is it randomizes it so you don't get lots of you
- 46:12know, the sky is falling sort of comments all at once.
- 46:17Or you can adjust it so you don't hear certain levels of
- 46:22emotion are are down rated. So, so there are things that you
- 46:27can do post doc out of the control of the companies and
- 46:33we'll just keep working at it. I think that one of the things
- 46:37that's really interesting is that a lot of interesting and
- 46:42scary is that there's a belief that having AI agents helping
- 46:48you do searches and find things and, and understand things will
- 46:53reduce the amount of advertising on the web dramatically.
- 46:58And because of legal constraints, if you're actually
- 47:01advising people about purchases and stuff, then some of the
- 47:06legal constraints actually bite where today they don't.
- 47:11And so some of the things that drive that click hunger are
- 47:14going to be changing. Now the danger is, of course, if
- 47:18you have agents doing all this stuff, you could have crazy
- 47:20things, right? And we've already seen a couple
- 47:24of agents that are absolutely crazy.
- 47:26And so we need to be able to, to regulate that sort of thing.
- 47:31This is the same sort of thing that all the big financial
- 47:34institutions are scared about. If you have agents that can do
- 47:38crazy things and the poor user may not know or the company may
- 47:43not know, then you can get huge amounts of of chaos in the
- 47:47financial markets. That's not good.
- 47:50Steve Tout on LinkedIn says that he is doing work.
- 47:57He says we need verifiable data sources and Oasis has a working
- 48:03group on this topic called Data Provenance Standards and he
- 48:07invites anybody to connect with him.
- 48:09If you're interested in this space.
- 48:10And on LinkedIn, he's put some URL.
- 48:12So if you're interested in data provenance, check out the
- 48:15LinkedIn comment from. It's really a key problem and
- 48:20something that people are working on.
- 48:21We're pretty involved in that too.
- 48:24You know, we'll see how all that comes together.
- 48:27We the thing that I think will drive it is when people see
- 48:31financial things happening, you know, $1,000,000 stolen because
- 48:34of data, you didn't know where it was come from, then we're
- 48:38finally going to get provenance. You started talking several
- 48:42times about data and what various countries are doing
- 48:48internationally. Can you give us some overview of
- 48:52that? Tell us what's going on there
- 48:54and why it's so important? I'll just start with a little
- 48:56story. So I was at a dinner where some
- 49:00of the San Francisco's leading intellectual VCs, you know,
- 49:04talking about why they're investing in all this sort of
- 49:06stuff. And I was sitting with the head
- 49:08of Singtel. This is the the biggest sort of
- 49:12telco in the Indo Pacific and Temasek, which is the sovereign
- 49:17wealth fund of Singapore. And if they were having a hard
- 49:20time not laughing out loud because they think that this
- 49:24notion of like, you know, frontier models and intelligence
- 49:27and it's going to run everything is just crazy.
- 49:30What they see is that if you have cheap accurate models that
- 49:36run every process, every business rocess, logistics,
- 49:39ordering, health, et cetera, if you make those agents, then you
- 49:43have all the data in the world. And if you have all the data in
- 49:47the world, you can be able to predict where to invest, where
- 49:51not to invest, what processes to put in.
- 49:54You are in the catbird seat and it doesn't take, you know,
- 49:57genius level intelligence to do that.
- 50:00It is something that is intelligent.
- 50:02It's cheap enough to be deployed throughout all the countries
- 50:06that are engaged in the international trade or have
- 50:10other sorts of business processes like that.
- 50:12So that's what China is primarily focused on through
- 50:17Belton Road and they've been doing very good job at it.
- 50:21You might want to look at versus who are the biggest trading
- 50:24partners in in Africa, but also India's very focused on that.
- 50:30They have a thing called the Citizen stack, which is a
- 50:33generalization of their India stack.
- 50:36It's open, free public software that provides identity
- 50:41contracts, payments, medical information, free, right?
- 50:47Government insisted all the banks and hospitals begin to
- 50:49accept it. Started by a small NGO, but they
- 50:54made a modular version of it that can be used by other
- 50:58countries. Other countries could configure
- 50:59it, right? And so they have at the moment
- 51:0212 countries signed up, 26 all together that are thinking about
- 51:08it and be doing parts. They've signed up 1.4 billion
- 51:12customers to this. Compare that to some of the
- 51:15other country and you probably haven't heard about it, right?
- 51:19But now what they have is they have this data structure that
- 51:23covers most fastest growing part of the world and it covers all
- 51:28the sort of basic parts of the economy.
- 51:30And you can bet people are building AI on top of that.
- 51:34So maybe you want to pay attention.
- 51:35It's not going to be the frontier stuff.
- 51:37It's going to be the stuff that, you know, helps the schools run,
- 51:40make trade happen, keep people healthy, stuff like that.
- 51:44But what they'll end up doing is they'll end up owning in the
- 51:48sense of being able to manage SEA and and take advantage of a
- 51:53data flows from the entire Indo Pacific.
- 51:56China has Belt and Road, which is, you know, sort of the Asian
- 52:00to European sort of thing. Again, they don't have to see
- 52:03the data, they just have to see the metadata.
- 52:06They have to know where things are flowing.
- 52:08And then that tells you where you're supposed to invest.
- 52:11I work as advisor to the UA ES sovereign wealth funds, and Yep,
- 52:19that's where they're headed. You know, this, this is the
- 52:24game. Because as AI gets smarter and
- 52:27smarter and smarter, the scarce resource is data.
- 52:30Do you know what's actually happening that this thing can
- 52:33then reason about? And unless you know have that
- 52:37data, you're just in the dark. Sandy, we're just about to run
- 52:41out of time, but can you tell us the describe the implications of
- 52:46this for us? The thing that people ought to
- 52:50be thinking about is how can they have visibility into larger
- 52:57parts of the economy through their what they're doing?
- 53:00I wouldn't say necessarily owning data, but having metadata
- 53:05about where things are going and who's doing what, that type of
- 53:08stuff. I think also one needs to
- 53:12remember that the thing that will actually drive most of this
- 53:16are the bad guys, the cyberattacks, the people
- 53:20committing fraud, things like that.
- 53:22Those are just beginning to ramp up in a way that's incredibly,
- 53:27it's really sort of hard to imagine.
- 53:29But as we move towards this, this is something that has new
- 53:33attack surfaces. It's handling actual, you know,
- 53:37trade money, things like that. It has real life implications.
- 53:41It runs things like hospitals and it took us a long time to
- 53:47sort of figure out how to semi bulletproof the the Internet.
- 53:52And these new sort of agentic systems are going to take time
- 53:55to bulletproof. And in the meantime, we're going
- 53:58to see a lot of disasters. So be safe, know where your data
- 54:04comes from, monitor things, you know, to know what's happening
- 54:08just the way you would monitor money, right?
- 54:11You know, if $1,000,000 disappeared from your account,
- 54:14you'd know about it in an instant.
- 54:16You need the same thing with these sort of agentic systems
- 54:19because they're actually the controller things that in the
- 54:23near future that are going to be the things that are hit.
- 54:28Sorry to be negative at the end, but you know it's getting it's
- 54:33getting hit that drives progress, sometimes more than
- 54:37having opportunity. Can you just share your view on
- 54:42the open source AI? And I'll just mention folks
- 54:44listening. We have the CTO of Mozilla
- 54:48coming up on a show talking about this.
- 54:51So subscribe to our newsletter, CXO talk.com.
- 54:54Just very quickly your view on open source AI and and its
- 54:58importance really fast please. Well, it's really a key thing
- 55:02because it makes people willing to adopt it.
- 55:05It's free, it's under your own control, you can see it, you
- 55:08would specialize it. And then the question to ask is
- 55:11who controls the data? Always comes down to the data.
- 55:14That's where you get insights, and that's what matters.
- 55:18There's so much to talk about and we barely scratched the
- 55:23surface of all the questions that came in.
- 55:26So I hope that you will come back, Sandy, and do this again
- 55:32another time with us. Would love to, yeah.
- 55:34More to talk about for sure. Things to move forward with.
- 55:39Sandy Pentland, thank you so much for being our guest.
- 55:43Everybody who's watching, you guys are amazing.
- 55:46Thank you For the people who asked questions, check out our
- 55:50website, go to cxotalk.com and subscribe so we can send you
- 55:54updates on new shows and we'll see you again next time.
- 55:57Thank you so much everybody, and I hope that you have a great
- 56:01day.