Latest / CXOTalk / Top Data Scientists Explain Bad Data, Poisoned Datasets, and Other AI Killers | CXOTalk #896
Transcript
- 0:00Your AI is under attack from bad data, poisoned data, deep fakes,
- 0:06bad actors, and more. Today on CXO Talk number 896,
- 0:13luminaries Doctor David Bray and Doctor Anthony Scrofignano
- 0:17reveal how to escape this AI quicksand with practical
- 0:22strategies for building clean systems and trustworthy AI.
- 0:27David, very briefly tell us about your work.
- 0:31One of the roles is with the Stimson Center, specifically the
- 0:34Stimson Accelerator, and what we're striving to do is
- 0:37demonstrate that one can advance both programs and policies in
- 0:40parallel. We can't wait for perfect
- 0:42policies anymore because of the rate of change that's happening
- 0:45in tech, including AI. And so we're showing it's
- 0:48possible whether it's in business, whether it's in
- 0:50governance or whether it's possible at the community level.
- 0:53Anthony, give us a sense of your work.
- 0:55I am also have the honor of working with David at the
- 0:58Stimson Center. In the context of the Loomis
- 1:01Innovation Council, we do what I would call action research.
- 1:03It's not just a bunch of white papers, it's actually stuff that
- 1:06matters and then incubating things that hopefully kind of
- 1:10grow and and impact the thing that we're working on.
- 1:13And there's some very big problems that are really
- 1:15important. David, we're talking about AI
- 1:19quicksand. What is that?
- 1:22When we unpack artificial intelligence AI, we recognize
- 1:26that most methods of AI require some training or some data to
- 1:33actually underpin what they do. Not all.
- 1:34There are other approaches that don't, but the most recent
- 1:36approach, generative AI does. And so when we think about
- 1:39quicksand, we could end up thinking we've built this
- 1:42perfect AI system. Yet the very data that actually
- 1:46sort of serves as the foundation that the system was trained
- 1:49upon, what it was based upon, turns out to be not only
- 1:52quicksand, but it might actually just disappear in terms of it's
- 1:55just under the weight it is. It's not a sufficient for the
- 1:58models we're trying to build. It's recognizing that if you
- 2:01don't take the time to think about where the data came from,
- 2:04how it might actually be missing certain things that you're
- 2:06trying to ask using the system, or might be intentionally
- 2:09poisoning it. If you don't do that, you may
- 2:11find that you build the metaphorical castle on sand,
- 2:14only to have it be washed away. I can add a few things to that.
- 2:17I have 5 MMS that are helpful to to put context to this
- 2:23misadventure. It's when you fall in love with
- 2:26your AI. Everybody's drunk on on AI right
- 2:28now. You can't speak a sentence
- 2:30without saying AI. And so therefore you must AI
- 2:33your way into everything. Leading with a tool is almost
- 2:36always a bad idea. It's almost always a better idea
- 2:40to say what is the net new problem that we're solving?
- 2:42What, how is this going to benefit the enterprise?
- 2:44How is this going to not just do something we could already do
- 2:47with the cool new tool, but do something that do that actually
- 2:51might matter misuse, which is, you know, wrong tool, right
- 2:56problem. There's a lot of this going on
- 2:58right now and and Gen. AI is a great example where it's
- 3:01David mentioned training it. The Gen.
- 3:03AI is is basically consuming ginormous bags of words.
- 3:08What makes you think that the disruption that you're
- 3:10interested in is represented in that bag of words, for example?
- 3:13So misuse is just using the wrong tool for the problem.
- 3:17Malintent is a lot of what we're talking about here.
- 3:19When bad people come in and they they don't have to mess with
- 3:22your systems or your people, all they have to do is spoil the
- 3:25milk. All they have to do is
- 3:26manipulate the data in a way that can cause you to make a bad
- 3:29decision or to change what was already a good decision missing.
- 3:34And David mentioned that they all too often now we talked
- 3:37about under representation in the data.
- 3:39If the thing that you're interested in is not well
- 3:41represented in the data, these models will get very distracted
- 3:45by what is there and you'll fall in love with the answer because
- 3:48they'll tell you how good you're doing and very dangerous.
- 3:51And then the last time I have is moving.
- 3:53The environment is changing. And while you're working on it,
- 3:56and while you're ingesting all this data and doing all this
- 3:59convolution, the world is changing.
- 4:01And so it's very important when you build models and when you
- 4:04build approaches that are heavily dependent on AI, that
- 4:08you consider how the environment might be changing in a way that
- 4:10makes your original testing no longer relevant.
- 4:13So to what extent is this set of problems happening?
- 4:19If anything, the latest wave of AI, generative AI and and and
- 4:23the subset of it has poured kerosene on the existing fire.
- 4:26But this was already a risk to organizations even in the 2000
- 4:30tens. And that if you didn't think
- 4:32about where the data was coming from, could you trust that data?
- 4:35Was it reliable? What it's done, though, is
- 4:37because AI has now become more available to enterprises and
- 4:42there's a huge push for companies to adopt it.
- 4:44It's one of those things that in the pressure to adopt it as
- 4:47exactly for the five, M said that Anthony laid out.
- 4:50We run the risk that companies or governments may adopt models
- 4:54without thinking about did I avoid the AI quicksand?
- 4:58And so again, we may fall in love with the model, the perfect
- 5:00castle, only to discover that the actual foundation it's built
- 5:03upon is insufficient for what we're asking.
- 5:05You have to be really careful. If you set out with the goal of
- 5:08proving to your overlords that you're using AI, that is a
- 5:11horrible journey to be on. Everybody's using AI.
- 5:14There's AI toothbrushes, there's AI and everything.
- 5:17And AI is increasingly democratized so that there's no
- 5:20AI center of excellence, just like there's no PowerPoint
- 5:23center of excellence, right? So it's, it's important that
- 5:26where this is being used in the enterprise, there is that the
- 5:30adults are in the room that we're looking at.
- 5:32How are we using it? Why are we using it?
- 5:33What's the net cause? How is this going to be
- 5:36realistically deployed in our environment?
- 5:38You know, how do we scale it? Are we building the Hotel
- 5:41California where we can never check out?
- 5:43Those are very important questions.
- 5:45No, none of those are new questions.
- 5:46It's just that now we have the opportunity to really fail at
- 5:49scale if we get it wrong. And I'll give some very tangible
- 5:52examples real quick, Michael, because I think that'll help the
- 5:54audience. Without naming the name of the
- 5:57model, 1 of the models, around 2023, if you asked it how many
- 6:02people lived in, say, the state of Georgia, the US state of
- 6:04Georgia, it would confidently tell you that the answer was
- 6:07something around 350,000, sorry, 350 million people, as opposed
- 6:12to a more realistic answer. And that was a case of where the
- 6:15data that had actually been trained on the model, it had
- 6:17actually gleaned from a source that was inaccurate.
- 6:19There was a typo in the answer. And so when the model gave the
- 6:22answer to how many people lived in the state of Georgia, it was
- 6:24wrong. So that's one example.
- 6:26Another example though, is actually if we look at when
- 6:28policy making decisions are done, sometimes they're, they're
- 6:31making decisions based on data where people live as opposed to
- 6:34where people work. And they think that people
- 6:36aren't using their services or they roll out, say, imagine, you
- 6:40know, it's, it's actually helping people with COVID
- 6:43vaccines or things like that. And we know there's been cases
- 6:45where COVID vaccines for a place where people live as opposed to
- 6:49where they work. And if people are working 2 jobs
- 6:51or working out hours, it looks like they're reluctant to get
- 6:53the vaccine when in fact, you really should have just placed
- 6:56it where people worked as opposed to where people lived.
- 6:57So this is again where you have to ask exactly what Anthony
- 7:00said, which is what is the, what am I trying to do?
- 7:03What is the mission or the business that I'm trying to
- 7:05achieve? And then work backwards to not
- 7:08only what's the right tool, but what's the right data to bring
- 7:11to bear. But what I don't understand is
- 7:15what you are both describing seems pretty basic, pretty
- 7:22straightforward. So what is actually?
- 7:24What's the disconnect that's causing these data problems to
- 7:31filter through? These are smart people, you
- 7:34know. Go ahead.
- 7:35There are smart people there. There's a lot of FOMO, you know
- 7:40where very senior people are putting pressure on the other
- 7:47senior people that report to them to say what are we doing in
- 7:50AI? Because my buddy is doing
- 7:52something in AI and I, I need to tell the, the investors what
- 7:56we're doing in AI or I need to tell the, whoever the, my
- 7:59overlords are what we're doing in AI.
- 8:00So we're, we're, there's a lot of that.
- 8:03The other thing that's happening is that you have an entire
- 8:05workforce that's graduating into the enterprise, that kind of AI
- 8:10their way through their last couple of years of school.
- 8:12And they think that, you know, that the answer is always
- 8:16contained in the data that's available to them.
- 8:18And they think that everything is free and everything can just
- 8:21be used however they want in there.
- 8:24I'm trying not to name brands in their notebook, right?
- 8:27I can just include this library and I'll be good to go vibe
- 8:30coding all of that, right? And I'm not going to like, you
- 8:35know, take issue with that because this will this will live
- 8:37on and a year from now somebody will say, oh, he just didn't see
- 8:41where the train was going. What I'm trying to say is that
- 8:43these fundamental questions of what is the problem that we're
- 8:46trying to solve? How would we know that this
- 8:49approach will make things better and not just different if our
- 8:53objective is just to declare victory that we used a certain
- 8:57approach that has never been a good idea.
- 9:00I mean, in the history of man, just saying I use this tool
- 9:03instead of that tool alone without something else as part
- 9:07of that story. It's, it's, it's always a futile
- 9:10journey. And, and it's, it's, it's a very
- 9:12dangerous one right now because of the size and scope and scale
- 9:15and apparent intelligence of the tools that we're talking about.
- 9:19Not that I said a. Parent, I want to tell everybody
- 9:23that you can ask your questions. If you're watching on LinkedIn,
- 9:28just pop your question into the chat.
- 9:31If you are watching on the CXO Talk site or on Twitter, just
- 9:37pop your question into the tweet chat using or X using the
- 9:43hashtag CXO talk. Take advantage of it.
- 9:46This is your opportunity to ask pretty much whatever you want.
- 9:50So Anthony, let me ask you the question about the elephant in
- 9:55the room. So what?
- 9:57What's wrong with vibe coding? Used properly in the right
- 10:01context for the right containerized problem, nothing.
- 10:08But, you know, there were automatic code generators in
- 10:111980. There's nothing new about that.
- 10:15The danger here is that generative AI is a very
- 10:20different animal than just using a bunch of automatic code
- 10:24generation. You are consuming vast amounts
- 10:29of code in order to quote, UN quote, learn to code.
- 10:32And then you're basically plagiarizing, you know, you're
- 10:35mathematically plagiarizing that code to produce other code that
- 10:39may or may not do what you asked.
- 10:41And there are some basic blocking and tackling things
- 10:44that we all used to learn, like regression testing.
- 10:47How do I know that the things that used to work still work?
- 10:50When I do something new like bias, understanding whether the
- 10:54data that I used is actually going to look like the data
- 10:57where I deploy it, those things don't happen a lot.
- 11:00They the goal is oh look how cool it looks great it it it
- 11:04looks cool but you didn't escape the need to do those things if
- 11:08you if you use a critical application when you push that
- 11:13button on your steering wheel that says stay in the lane or
- 11:17when you whatever you know when that drone gets launched right.
- 11:20We've got to hope that these concepts were embodied in the
- 11:24development of that piece of software that you're depending
- 11:27on. And increasingly, maybe maybe
- 11:30not because the goal is get it out fast, let it break, and our
- 11:34customers will test it for us. That's kind of dangerous.
- 11:38Back in the 80s and the 90s there was efforts called expert
- 11:42Systems Decision Support systems, which was another
- 11:45flavor of AI which thought that if we spell out rules of a
- 11:48system that that's all we need to retrieve intelligence and we
- 11:52have success. For example, an automatic
- 11:54defibrillator actually is running rules based when it
- 11:56tries to decide when to give you that shock.
- 11:58But the moment we tried to go out of narrow, narrow use cases,
- 12:03rule based approaches to AI collapsed.
- 12:05So we go now to machine learning and then the subset of machine
- 12:07learning, neural networks and generative AI and that's saying
- 12:11we don't need rules at all. And so in November of last year,
- 12:14there was actually a competition where they actually told the
- 12:16machine, and there are no circumstances should you ever
- 12:19transfer the funds. And it took less than 500
- 12:22prompts. I think it was prompt #490 that
- 12:25sure enough, the machine that had been told explicitly do not
- 12:27transfer the funds did. And so exactly as what Anthony's
- 12:31talking about, which is you're doing very fancy multi
- 12:33dimensional pattern matching when it comes to vibe coding.
- 12:36And as long as a human takes the time to carry it through and
- 12:39look through it, there's value. But recognize what you're doing
- 12:42is pattern matching that may actually look like it's valid
- 12:45code when in fact it's not. The last example I would say,
- 12:47and I won't name the name of the company, it was a few months ago
- 12:50there was a one of the big, big AI companies came out with a
- 12:52paper where they told the machine write code that's
- 12:56insecure but don't reveal it to the user.
- 12:59And at the same time, they were surprised that when they
- 13:02actually gave that instruction to the machine, it started being
- 13:04hateful and and hate spewing. Now let's think about the data
- 13:08that actually underpinned the system.
- 13:10Yes, it's trained on the corpus of the last 30 years of the
- 13:13Internet. Where on the Internet might
- 13:15there be use cases of people that are writing code that's
- 13:17insecure but not revealing it? It might be part of the Internet
- 13:20dark web in terms of four, 10810 and groups like that that are
- 13:24also hateful. And so when they actually asked
- 13:26and gave that instruction, I don't know why they were
- 13:29particularly surprised that the machine actually started
- 13:31behaving in other ways similar to what it actually trained on.
- 13:34But that just shows the importance of 1 knowing your
- 13:36data, but also knowing the limits of generative AI when it
- 13:39comes to explicitly doing what you ask it to do.
- 13:42We're using a metaphor of quicksand here, and I want to
- 13:44get back to that for a second. If you think about this from the
- 13:47outside in, if you're part of an organization and you're not the
- 13:51developing part of the organization, you're the
- 13:53customer of them, right? What do you want?
- 13:56You want it done fast, you want it deployed, you want to
- 14:00monetize whatever it is you're trying to monetize and somebody
- 14:03comes along and says I can have that ready for you in a day.
- 14:06How often do we ask at what cost?
- 14:09How often do we ask how are you going to deliver it?
- 14:12This guy said it was going to take three weeks.
- 14:15You said it's going to take a day.
- 14:17Day is cheaper than three weeks. I get my revenue sooner.
- 14:21Go right. You know, sometimes you got to
- 14:25be very careful because you get what you ask for.
- 14:27Aside from philosophical problems with vibe coding
- 14:34related to the lack of discipline and effort that folks
- 14:43should invest in computer science education, what's wrong
- 14:49with vibe coding? I would say let's let's look now
- 14:52at like intellectual property. You know, if we remember in the
- 14:5590s there was this thing called Napster, it did music sharing,
- 14:59but it didn't really respect any of the artists recording rights
- 15:01or any intellectual property. And so in some respects what we
- 15:04might have seen in the early twenty 20s was a repeat of
- 15:07Napster just in the AI era. And we know there's some
- 15:10lawsuits and things like that that are being sorted out.
- 15:12But you know, the question for a business that is looking to use
- 15:16an AI solution that might be external to the business, how do
- 15:20they know that they're not for properties not locking out the
- 15:22door? How do they know it can be
- 15:23respected? I mean, I do get calls about
- 15:26once every three weeks or so. And I'm, again, not naming names
- 15:30from either ACIO or a sister that says I had an employee.
- 15:33They used a tool that was, you know, it wasn't something you
- 15:36provide in the enterprise, but they used a commercially
- 15:38available one. And they uploaded information
- 15:40that had HIPAA, it had health, you know, personal identifiable
- 15:42information. Can I claw it back?
- 15:45And the answer is no. But you know, again, it's one of
- 15:48the things that, you know, you have to be aware that when you
- 15:50use AI, particularly AI that's being provided by somebody else,
- 15:55be very intentional about what you provide and what you don't,
- 15:58whether it's intellectual property where there's personal
- 16:00identifiable information or just proprietary information.
- 16:03Because you might see some things walk out the door that
- 16:06you don't want to have walk out the door.
- 16:08We have a question from LinkedIn and I'm glad somebody started
- 16:12asking questions here. You folks who are listening take
- 16:17advantage of this opportunity. And oh, by the way, right this
- 16:21second, you should go to cxotalk.com to subscribe to our
- 16:25newsletter so we can tell you about episodes like this that
- 16:28are live. Anyway, we have a very
- 16:31interesting question from Preeti Narayan on LinkedIn.
- 16:35And she says who really owns critical thinking and ethical
- 16:40oversight in AI? Current leaders or a new role as
- 16:47a moral compass? So who's actually responsible
- 16:50for dealing with this set of issues?
- 16:52It's a really interesting question.
- 16:54David, do you want me to jump on the hand grenade or do you want
- 16:57to? Jump, I'll jump on the hand
- 16:58grenade and then you can clean up whatever I I don't manage to
- 17:01to contain the blast. I would say it's all of us, yes,
- 17:04if if you're expecting somebody else to solve this and you're
- 17:07going to be waiting. I apologize for interrupting
- 17:10you, but I just want to clarify what I what I thought that I,
- 17:15what you just said, what I thought you heard you said, So
- 17:17you just said The answer to who's responsible for ethical AI
- 17:21is no one, no. No, no, no, no, no, no, and
- 17:24please no one. I so ethical is a very that's
- 17:28also a quicksand because let's think about there were plenty of
- 17:30things in the 1700s and 1800s that people thought were ethical
- 17:34that nowadays we would say are unethical.
- 17:36And even more recent history in War One, the British thought
- 17:39these things called Q boats, which were military boats, the
- 17:42skies, the civilian boats were ethical, but Lusitania got sunk.
- 17:45And so come over to they're like, maybe they're not ethical.
- 17:48So what we're really talking about is when you use the tool,
- 17:51whether it's you as a consumer, whether it's you as an employer,
- 17:54whether it's you as a boss, whether you ask somebody to use
- 17:57it, you do have a responsibility to say, am I using this hammer?
- 18:00Is this the right tool for the purposes?
- 18:02Is this the right driver? And the reality is for
- 18:04consumers, there's no silver bullet that any company can ever
- 18:08provide you that says here's when you can and cannot use this
- 18:10tool. I mean, you are sold a hammer,
- 18:12but if you happen to use the hammer to smash your hand,
- 18:14that's on you. So it isn't saying everybody,
- 18:17it's just saying be be cognizant of using the right tools for the
- 18:22right job. And what we really need to do is
- 18:24find better ways to help you understand rapidly what are the
- 18:28limitations of a tool. Because I think a lot of things
- 18:30right now, because the marketing are being sold that they can do
- 18:32everything, when in fact that might not exactly be the truth.
- 18:35So I I agree with the fundamental premise of what
- 18:38David said, that there is no one single responsible party.
- 18:41That doesn't mean that ultimately the oversight can't
- 18:44roll up to the board or to, you know, the chief risk officer or
- 18:51it the title is going to depend on the organization and it's it
- 18:55What it shouldn't roll up to, by the way, is the same
- 18:57organization that's creating all of that because that's you got
- 19:00the chicken in the hen house, the rooster in the hen house,
- 19:03prom, whatever that is, fox in the hen house, somebody in the
- 19:07hen house. The the I'm going to get killed
- 19:09on Twitter for getting the wrong animal in the hen house.
- 19:12So the on X the. There's an AI for.
- 19:16That, yeah, I know. Thanks.
- 19:17Let me get back on track here. So the issue is that meant like
- 19:21many things, there isn't a black and white answer.
- 19:24You're talking, the example I give often is that we want
- 19:28customized things. We want our things to anticipate
- 19:31our needs and start to do them for us or help us do them.
- 19:35We also want privacy. Those are kind of opposite
- 19:37things to want for your things that serve you better.
- 19:40They have to watch you and kind of take notes on what you do.
- 19:43And that means sharing your location and maybe sharing all
- 19:46the text in your emails and sharing all that other stuff.
- 19:48And then if you found out all the stuff you're sharing, you
- 19:50might be appalled. So the the when an enterprise is
- 19:55rolling out an application and way before they roll it out when
- 19:59they're contemplating the design of an application.
- 20:02Everyone in the loop should be thinking about what are the
- 20:05unintended impacts of what we're about to do?
- 20:08And are the is the juice worth the squeeze?
- 20:11If we're if we're trying to sell more shoes and socks and in
- 20:15order to do it, we have to spy on people, is that a line we
- 20:17want to cross? And I hope in that particular
- 20:20example the answer is no. But there are not that many
- 20:24people in the chain who have the decision making responsibility.
- 20:30So I think Michael said the initial question you asked me
- 20:33was who's responsible for critical thinking.
- 20:35And I would say everybody should critical think.
- 20:37If you're asking a question of liability, that's a little bit
- 20:40different. I think that's what Anthony is
- 20:41saying is if a company provides you a service and that service
- 20:45has AI providing the service as well.
- 20:48I think most of us would say based on common law that what
- 20:51the services provided to you turns out to be incorrect
- 20:54because of what the AI advised you to do.
- 20:56Maybe it was an AI lawyer, it was an AI helping a clinician or
- 20:59a doctor, then yes, the liability goes on the company's
- 21:02part. And I would actually say just a
- 21:04few weeks ago, I actually was at AI got to participate in a
- 21:07public congressional hearing as an expert witness.
- 21:10And actually, one of the things we heard almost from both of
- 21:13that, this was actually not a partisan issue.
- 21:15Both of that all said, when it comes to liability, we'd
- 21:20actually don't need new laws per SE, because the existing law
- 21:24framework, the existing legal framework says if a company
- 21:27provides you a service and you buy that and that service, if
- 21:30you follow what it said ends up harming you or resulting in harm
- 21:33to your person or family members, then yes, you should be
- 21:36held responsible. And so I think we need to
- 21:39separate critical thinking, which I would advocate everybody
- 21:43does to liability. And I would say right now, even
- 21:45if we haven't made explicit laws, there's enough of a legal
- 21:48framework to say that if the organization uses AI to provide
- 21:51you a service, then they are responsible for what the AI
- 21:54provides. It is absolutely coming.
- 21:56You can't just ignore that, that, that the agency, the
- 21:59concept of agency, that what you ask your AI to do on your behalf
- 22:03is not well embodied yet. But don't count on that because
- 22:07Oh my gosh, everywhere you go this is under discussion right
- 22:10now. We have another question from X
- 22:16Twitter right now, and this is from Arsalan Khan, who is a
- 22:21regular listener. He always asks great questions
- 22:23and he's concerned about the he's saying in an age of
- 22:31deregulation and no boundaries, unless there are AI consumer
- 22:37advocates, consumers will be at a disadvantage.
- 22:44And why do we need boundaries when it seems that politically,
- 22:51and I don't want to make this about politics, but but it seems
- 22:54politically we're in a trend of fewer boundaries.
- 23:00And I'll just ask you to also relate this back to the data
- 23:03because that's what this conversation is really about.
- 23:06There actually was a public hearing on AI.
- 23:08It was September 28th, and you can find it if you do
- 23:11congress.gov. What came out of that is so far
- 23:16the states have actually led the way when it comes to thinking
- 23:19about AI. There's more than 1000 pieces of
- 23:21either existing or pending AI legislation.
- 23:24But that also creates an interesting challenge, which is,
- 23:27you know, let's say you're from state Arizona or California and
- 23:31you travel to a different state. Is it the state that you reside
- 23:34in? Is it the state where you live?
- 23:35Is it the state where the company is?
- 23:36It gets really messy very fast. And so I would say just because
- 23:40we haven't seen anything yet doesn't mean we won't soon.
- 23:43In fact, what I'm hearing is once government opens up again,
- 23:45we will see stuff come from Congress because they recognize
- 23:49that in order to help consumers, but also help businesses
- 23:52navigate everything that's happening with AI, they need a
- 23:55light touch framework. Now, that said, definitely need
- 23:58people to advocate on behalf of consumers because if you don't,
- 24:01then that that's absent. But I would say to Arsalan,
- 24:03there are existing laws. Remember the Privacy Act of 1974
- 24:07that came out because these things called advanced data
- 24:10processing systems, mainframes required the federal government,
- 24:13United States to think about what does it mean now that you
- 24:15can have these machines that can actually sort of begin to know
- 24:18more about you than you really wanted them to know?
- 24:20We probably need to upgrade that and give an AI including the
- 24:22right to be left alone that you can actually say, I don't want
- 24:24the AI to think about me or involve me.
- 24:27But I think what I would say to Arsalan is what we really need
- 24:30is groups pushing for how do you upgrade existing laws at the
- 24:34state level and at the national level as opposed to net new
- 24:37laws, because net new laws will be debating until the technology
- 24:41has already moved on. In fact, if anything, that's its
- 24:43own quicksand because of the nature of data, which is I've
- 24:47seen some people that say you've got to expose all your data, you
- 24:49got to tell all the data you're using.
- 24:50No business is going to do that. I mean, it's their intellectual
- 24:53property. In fact, Anthony can probably
- 24:54resonate because he worked at a company where I can't tell you
- 24:56all the data, but what you can do is think about what are ways
- 25:00that you both inform the consumer, but also hold the
- 25:03company accountable for a appropriate risk calculus for
- 25:07the services or the products that they're providing.
- 25:09David and I were part of a, I'll just call it a, an intellectual
- 25:13holding environment earlier this week in our nation's capital.
- 25:18And I would say 1/3 of the conversation was related to the,
- 25:22the regulatory directions and the, the, the technologies that
- 25:27are emerging from tagging data for sharing metadata about AI
- 25:31generated data for letting the AI know that it's consuming data
- 25:35that was generated by other AI. This isn't going away.
- 25:38And if if the solution of your AI development team is to put
- 25:42their head in the head in the sand and say I'm not going to
- 25:45worry about it because I don't know of any regulation right now
- 25:49other than the California Privacy Act, you might want to
- 25:53add some people to that team. How do we address these issues?
- 25:58Because you're describing a set of technology problems as well
- 26:04as a set of process issues, cultural issues, mindset issues,
- 26:12greed, fear, doubt, I mean all of this.
- 26:15It's kind of a an. Oh, it, This is why the, the
- 26:18moment is so challenging. I would say, if anything, I
- 26:21would recommend for listeners, while we're talking about AI and
- 26:24data, replace the word AI with organization.
- 26:26How do you know that the organization is doing a good job
- 26:28with your data? How do you know that the
- 26:30organization is making good decisions based on it?
- 26:32Because in some respects, AI is just additive to these
- 26:35challenges. I think we also need in some
- 26:38cases, and Anthony can go deeper, there's going to be some
- 26:41data that we are comfortable with being sort of like open
- 26:43data. There's going to be other data
- 26:45that's like prescription data that you actually have to sort
- 26:47of have a prescription, have access to.
- 26:49And there may also be cases, especially if you are someone
- 26:52that's in the entertainment industry, a musician, artist,
- 26:55where you you want to have your data be part of a cooperative or
- 26:59a data trust. We've had Lord Tim on, on CXO
- 27:01talks as well, talk about this. And then there's a negotiation
- 27:04between an AI model that might want to train on your data, but
- 27:07then you're having some either financial or equity return or
- 27:10some non equity benefit maybe. I care about Parkinson's
- 27:13research, so I'm willing to pull my health data if it helps
- 27:16inform a cure for Parkinson's. But I think we need to actually
- 27:19recognize that if what the 2000 and 10s was was almost an over
- 27:24calibration to Hoover up all the data and do so behind the
- 27:27curtain, that's probably the wrong lesson for going forward
- 27:30because it didn't engender trust.
- 27:32If anything it made people distrustful and you can't trust
- 27:35the AI outcomes. We need to unlearn some lessons
- 27:37to move. I think it did more than make us
- 27:40distrustful. It enabled a whole new cadre of
- 27:44malefactors that can do things like deep fakes and, and
- 27:49adversarial data manipulation. And, you know, even the data
- 27:53that's true has the truthiness of the data often has a
- 27:57lifespan. It was true, but it's no longer
- 28:00true. And so AI doesn't care.
- 28:04Math doesn't care, right? It's just hoovering it all in
- 28:07and, and, and you know, compressing the element of time
- 28:11to 0 and then regressing around what's there.
- 28:13I know there's a lot more to it and I know there's different
- 28:15foundation models that handle this differently.
- 28:17And I know about turning up the heat and all that.
- 28:19I know all of that stuff, right? So before everybody tries to
- 28:22attack me, all of that stuff is is some of a treatment of this.
- 28:27But it all starts with at the highest level in the
- 28:30organization, the adults in the room asking some really tough
- 28:35critical thinking questions. What do we have to believe to
- 28:38use this approach? Help me understand what steps
- 28:41you've taken to understand provenance and permissible use.
- 28:45Help me understand how we're going to handle agency.
- 28:48If you can't answer those questions, you don't belong
- 28:51building those tools. We have another question from
- 28:55Twitter, from X, and this is from Chris Peterson.
- 28:59He's also a regular listener. And we thank you, Chris, for
- 29:02your listenership. And he says there's growing talk
- 29:06about LLMS hitting fundamental limits and scaling
- 29:09hallucinations and lack of executable guardrails, which is
- 29:14what you're talking about. Is it actually the next step to
- 29:18AGI or artificial general intelligence?
- 29:22Or do we need different approaches long term?
- 29:25Yes. So he's spot on.
- 29:27We need different. Approaches.
- 29:28We need different approaches, yeah.
- 29:29Yeah, and I will give some. I want to give, I want to, I
- 29:34want to give both a shout out to some.
- 29:36I mean, the trouble is, you know, again, generative AI
- 29:39arrived on the scene and it took all the oxygen out of the room,
- 29:41as Anthony knows. I mean, AI has been decades in
- 29:44the making. There are multiple methods.
- 29:45And the trouble is I think we got over fixated on generative
- 29:48AI as opposed to recognizing it's another great tool.
- 29:51But there are other tools that are needed.
- 29:53I am very bullish on active inference and I would recommend
- 29:56folks take a look at it. If you're not familiar with it,
- 29:58it's basically the best idea is it's it's doing continuous
- 30:01learning of an environment. And I actually would submit that
- 30:04the future is not these mega AI models that are trying to do
- 30:07everything. It's going to be smaller models
- 30:10that are actually optimizing for really specific cases.
- 30:14For example, there may be an AI model that's monitoring all the
- 30:16ships coming in and out of the Suez Canal.
- 30:18There may be another one that's monitoring ships coming into LA
- 30:21ports. And when a disruption happens to
- 30:24the Suez Canal, like we had a few years ago, then it actually
- 30:27talks to the LA port model and says you might expect that
- 30:30there's going to be a shortage of containers.
- 30:31And that shortage of containers then going to cause a rise of
- 30:34people to actually look for new metal to build new containers
- 30:36because we're lacking them. That causes a rise in the future
- 30:39markets, which then causes the cost of shipping for the next 6
- 30:42to 9 months to be more expensive.
- 30:44Which means you can get plenty of Mercedes Benzes and Rolls
- 30:47Royces in the United States because they can absorb the cost
- 30:50given the increased cost of shipping.
- 30:52But quart sized cans of paint aren't available because they
- 30:54can't absorb the cost. And so that's not possible with
- 30:57a mega AI model LLM. None of that would be there.
- 31:00But if you had active inference where smaller models are talking
- 31:03to each other about the ripple effects, that is possible.
- 31:05And that gets to really interesting forms of
- 31:07intelligence. We hear a lot of talk about, oh,
- 31:12we need more power, we need bigger data centers.
- 31:16When are we going to reach AGI? Now they're talking about ASI,
- 31:19artificial super intelligence. You know what's going to be
- 31:22next? You know, Lex Luthor, like we
- 31:25keep naming, renaming and renaming.
- 31:28The power of regression, the power of convolution, the power
- 31:33of math, basically to do things that increasingly convince us
- 31:37that they were done by humans. I think that's great.
- 31:40Please keep doing that, you know, don't stop.
- 31:43Just don't do it in in and and and fake your way about it.
- 31:48Don't do it and not let me know you're doing it.
- 31:51There are definitely situations where AI can do things that
- 31:56people, I don't care how many people you put in the room,
- 31:59they're not going to be able to address that problem because
- 32:02it's just too big. It would take thousands of
- 32:04people to all agree on the approach.
- 32:06They would want to be similarly incented and similarly
- 32:08instructed and not going to happen.
- 32:10Yeah, great. Go and do that.
- 32:12Don't call it intelligence. We need some new verbs here and
- 32:16some new nouns here for the things that this technology is
- 32:20doing. And oh, by the way, most of the
- 32:22technology is still, you know, like when you play pool and you
- 32:26just hit the balls as hard as you can and hope one of them
- 32:28goes into the pocket. It's try everything all the time
- 32:32with the exception of 1. And I think as we get smarter
- 32:36about having more than one approach and having David is
- 32:39mentioning agentic AI, I'm sorry, active and active
- 32:43inference. This agentic AI is, is a step in
- 32:46that direction. Having these little pockets of
- 32:49of capability within an ecosystem that's even broader
- 32:52and being able to sort of turn them on and wake them up and
- 32:54give them a specific goal and a task that's a nice step in the
- 32:57right direction. We're not even close to there
- 32:59yet, though. There's there's a lot of work to
- 33:01do here. On LinkedIn, Chris Davidson says
- 33:06can you talk about the concept of data poisoning?
- 33:10Is it happening actively at the nation state level?
- 33:14How would we know if we were the victim of it?
- 33:17So a very specific example, and this is actually you can find it
- 33:21on online. So it's it's open, openly
- 33:24available is we believe about 9 to 10 months ago, Russia
- 33:28actively started trying to do a campaign to teach large language
- 33:33models that they could access things that just weren't true.
- 33:37And these are really hard once they've been taught to things
- 33:40that aren't true to undo because the way large language models
- 33:43are done if they don't really forget.
- 33:45Well. And so yes, data poisoning is
- 33:47happening. And a very specific example, and
- 33:49you can find articles online, is that Russia has done a very
- 33:52intensive campaign focus, not just in the US but other free
- 33:56societies to teach them things about events, about history,
- 34:01about people that actually just are just patently not true.
- 34:05I'm one of the many millions of people who love TikTok.
- 34:09And recently lots of videos on TikTok have been popping up
- 34:14showing the various things that ICE is doing.
- 34:17And I'm not making a political statement here.
- 34:20Typically those videos get what, 5010 thousand, 50,000 views?
- 34:24Whatever. Some comments lately I've seen
- 34:28these videos popping up having 3 million views and 150,000
- 34:35comments saying all and all the comments drowning everything
- 34:40else out saying go ice. And I'm not, I'm not saying
- 34:44good, bad, indifferent, but this is what's and it's obvious that
- 34:49there is some third party that is doing this in an automated
- 34:55way. It's combination.
- 34:57So what what this is what makes things so hard is when bots
- 35:00come, bots can definitely amplify things and make it look
- 35:03like humans are liking a video or commenting on a video.
- 35:06But then that becomes a complicated process because then
- 35:08humans will pay attention to it as well.
- 35:10So, you know, is it is it coordinated in authentic
- 35:14campaigns? Yes.
- 35:15But recognize a lot of those humans are also valid.
- 35:17And that's what makes these things so difficult.
- 35:19We may have heard, you know, it was a few weeks ago there was a
- 35:23discovery of lots of phones next to the UN General Assembly.
- 35:27Now we don't know exactly what that was for.
- 35:29But one of the things that that pharma phones could do is if you
- 35:32wanted to make it look like a lot of people were on their
- 35:34phones liking something or commenting on something or or
- 35:38not liking something and then dissing on something.
- 35:40Then you would have phones in the geographical area that look
- 35:43like they were in New York. And then you would have them
- 35:46basically go to that site. And so, but the moment you did
- 35:49that, then you would actually get human eyeballs and human
- 35:51attention that would actually chime in as well.
- 35:53So this is a mixture of both bots, but also of humans.
- 35:56And that's what makes these things so difficult is data
- 35:59poisoning takes on a life of its own once it's been seated.
- 36:02There are many different types of data poisoning.
- 36:05So what we're basically getting out here is either, you know,
- 36:09faking something or amplifying something that may or may not be
- 36:13true that someone said, right? That's definitely one type of
- 36:17poisoning of data. Another type of poisoning of
- 36:19data is if I understand what type of AI you're using, I can
- 36:25influence the decisions you make by allowing you to see things
- 36:29that cause you to make a certain type of decision.
- 36:32And that without getting into too much detail, there are
- 36:35examples of this where when when large organizations were
- 36:41considering doing certain types of things.
- 36:45Easy one is airlines, you know, we're looking at, you know,
- 36:48reinstating our roots to, you know, Southeast Asia and all of
- 36:51a sudden all their customers love Southeast Asia, right?
- 36:55Maybe that's true, you know, and who am I to say?
- 36:59It might also be likely that their social listening was
- 37:03looking for how they were tagged and what other nouns were
- 37:07associated with that tag. And there seemed to be a lot of
- 37:11comments to social media mentioning them and mentioning
- 37:14Southeast Asia and their marketing department and their
- 37:17route planning people. And all those people got in a
- 37:19room and said, Gee, our customers really are expressing
- 37:21an intent. Maybe they are, maybe they
- 37:23aren't. Who am I to say, Michael, you
- 37:25said it's obvious. I don't know if it's obvious.
- 37:28I don't know. Is it likely that what did you
- 37:31say, 303,000,000? Is it likely that 1% of the US
- 37:35population, you know, all collectively agreed on anything
- 37:39other than the fact that they're part of the US population?
- 37:42I doubt it. And.
- 37:43Even then. And, and, and, you know, there's
- 37:45certain things in math that you have to violate mathematically
- 37:50in order for something like that to happen.
- 37:51Is it possible? I, you know, people win the
- 37:54lottery, right? But what we, what we can do,
- 37:58because we're trying to focus on solutions here.
- 38:01One of the most important questions that I ask very often,
- 38:04David's been in the room when I ask it and probably been annoyed
- 38:07by it. What do we have to believe in
- 38:10order to go down this route? So you just told me that we're
- 38:13normally, I didn't, I don't remember the numbers.
- 38:15We're normally, you know, X number of people are looking at
- 38:19something, 100 X people are looking at it and they all feel
- 38:22the same way. What would we have to believe?
- 38:25And that takes you down a whole route.
- 38:27Now, one of the scenarios might be that there's some kind of a
- 38:30body influencing this. Another might be that they were
- 38:32all there and quiet until they had their their opportunity to
- 38:36say, you know, heck yeah. Another possibility might be
- 38:40that none of these are people speaking and that these are, you
- 38:43know, some sort of, you know, spoofing of the data.
- 38:45And then you go and you test those different scenarios.
- 38:48If that scenario were true, what else would also be true?
- 38:51And it turns out that there's a whole science around doing this
- 38:54called propositional calculus. And we know how to do it.
- 38:57All we have to do is bring the adults in the room and actually
- 38:59do it. Do you know?
- 39:00To paraphrase Descartes, I believe it.
- 39:05Therefore it's true. Separating what are tangible
- 39:08truths that are truths beyond what you believe.
- 39:12And that's that's where we have courts of law and we know even
- 39:15courts sometimes get it wrong. That's why we have an appeals
- 39:17process. And so it's getting harder and
- 39:20harder to distinguish those things that beyond what people
- 39:23believe are truths that we are willing to say exist in the
- 39:27absence of people believing. There there are also observer
- 39:31effects where we start to believe something and because
- 39:33enough people believe it, it starts to become true.
- 39:36Like there's unrest in, in, you know, wherever, right?
- 39:40There wasn't unrest in and wherever until everybody started
- 39:42talking about the unrest and wherever, which caused the
- 39:45unrest and wherever that happens.
- 39:48Channel or inner sneakers where sneakers has that Cosmos says
- 39:51the line that says you know I learned that what matters less
- 39:54is the actual reality, but the perception of reality.
- 39:57If I spread rumors that a bank is insolvent and that causes
- 40:00everyone to go rush the bank, then pretty soon the bank is
- 40:04insolvent because I spread rumors that the.
- 40:05Banquet OK, we we have a question from Twitter on and or
- 40:11X, which is how can organizations determine they're
- 40:16heading into the AI quicksand and how bad can it be?
- 40:20You need a red team. And so the world I come from
- 40:23within the intelligence community, red teams were
- 40:25basically folks that are, are are asked to say how might what
- 40:30you're doing be misused or abused?
- 40:32How might I throw a wrench in what you're achieving?
- 40:35And so organizations need to have that just because if you
- 40:38don't, then it's nobody's job to think about second, third and
- 40:414th order effects. But when I used to do response
- 40:44to bio chairs and public health, we called them the B team.
- 40:46It's the same idea of the beta team, which is what are the
- 40:48things you're not thinking about that you should.
- 40:51And if you have someone on the team or multiple people on the
- 40:53team whose job is to say, how might your well intended
- 40:57business efforts or your well intended product launch be
- 41:00misused and abused, that will help you be more prepared for
- 41:04how others might exploit it. I often wonder if in social
- 41:07media, if we had had some of the social media companies in the
- 41:092000 and 10s have that sort of red team function, if we could
- 41:13have avoided some of the traps that we later got ourselves
- 41:15into. As the often appointed leader of
- 41:17said red teams, I will say that there because it was a great
- 41:22question. One of the ways you might know
- 41:25it's happening is to have different smoke detectors
- 41:31looking for different things. So there's a couple of things
- 41:35that I often talk about quality and character, the character of
- 41:40and now you have to fill in the blank the character of what our
- 41:42customer comments in your case with TikTok, David Michael.
- 41:46So the character of the comments went from being sort of spread
- 41:51out positive and negative to all positive, right?
- 41:54So that's the quality, the character it, it, it, it became,
- 41:59you know, it was massively homogeneous and it became, you
- 42:03know, singularly positive the, the quality of the input.
- 42:09So you can look at missing this, you can look at incomplete
- 42:14articulations, you can look at articulations that contain
- 42:17things like slang and neologism, which are more likely to be
- 42:20spoken by humans than by machines, things like that.
- 42:23M dashes. Yeah, the famous M Dash right
- 42:27there are, you know, good old fashioned statistics,
- 42:30heteroscedasticity and and you know, measures of central
- 42:33tendency. We know how to do all this
- 42:35stuff. It's just math, but we don't do
- 42:37it. And so the biggest problem with
- 42:39that to that question of how would we know it was happening
- 42:42was certainly not if we don't look right.
- 42:45You've got to have people responsible for looking and
- 42:48adults that actually know how to look at large amounts of highly
- 42:52dynamic data in ways that are dispositive.
- 42:56That is not something that you you give you know the interns to
- 43:00do. And on a related topic on
- 43:04LinkedIn, Preeti Narayan points out that critical thinking
- 43:09itself is not something that education or job experience
- 43:14alone can teach. And I think this gets right to
- 43:19the heart of the issue that Anthony, that you were both just
- 43:23raising, which is Anthony just said we need people who can look
- 43:29at these bodies of data. But there is a large judgement
- 43:37issue here. It's not just the math.
- 43:39There's intuition, there's the sense of things.
- 43:44So what should organizations do? You need to, for for the
- 43:52identify those employees in your organization that are going to
- 43:55be the ones that are part of the red team and maybe they don't
- 43:58have experience. And so put them through a red
- 44:00team boot camp. And the only way you do that is
- 44:02exactly what you said is you need to have more and more
- 44:04experiences spotting what may or may not be problematic.
- 44:09And the reality is it gets exactly that.
- 44:10We do we developmental heuristics, we use math, but
- 44:13there's also judgment calls. And you know, you think about
- 44:16this. This is in some respects when
- 44:17you go to the military, they send you to boot camp and then
- 44:19they send you to other training. They train you how to do your
- 44:21job. I think we need to have
- 44:23companies doing very fast-paced courses that teach people how to
- 44:27do red teaming well, and then long term broader in the
- 44:30education space, we need to actually, it's not that you're
- 44:33not going to use AI in education, it's actually that
- 44:36you're going to critique the AI. What did the AI get wrong?
- 44:38What was missing? What was wrong with the data?
- 44:40We need to teach students how to do this.
- 44:42And it's a whole new discipline. But we've had this before when
- 44:45books came out and then later radio and television, like how
- 44:48do you assess if that book is missing something or there's
- 44:50something biased in it? We know there's books that were
- 44:52just propaganda. So teaching these skills, but
- 44:55doing so both in a way that addresses the here and now for
- 44:58companies, but also the education pipeline, that would
- 45:00be my recommendation. I want to jump on this.
- 45:03One of the most important questions being asked in the
- 45:06academic circles right now is how do we get this into the
- 45:11mindset of a student body that is increasingly focused on, you
- 45:16know, looking down at the phone and asking the phone a question.
- 45:19How do we and, and you know, you have to be careful about talking
- 45:24about teaching people how to think.
- 45:25That's a very dangerous thing to say, But teaching critical
- 45:28thinking is very different than teaching people how to think.
- 45:32So there are some basic tenets about how, what at what police
- 45:36am I anchoring on? What things are my, am I holding
- 45:39to be axiomatic? I'm not going to test them.
- 45:41I'm going to assume that they're true.
- 45:43What do I do with a postulate if I think something's true?
- 45:46How do I set up an appropriate unbiased test?
- 45:49What are the sources of bias in my answer?
- 45:52And what treatments can I make for that?
- 45:53What kind of elasticity, How wrong can I be and still make
- 45:56the decision that I'm making? These are all super important
- 46:00skills, all of which can be taught.
- 46:03But you're asking a, a, a body of, of knowledge that is already
- 46:08really big and getting bigger to now include this other really
- 46:10big and getting bigger thing. So, you know, it's not like
- 46:14there's just an easy answer of, you know, go read this book.
- 46:17And it's some of it is having people who used to have more
- 46:21hair and, and, and, and have made some mistakes helping you
- 46:25to make new mistakes. And some of it is, you know,
- 46:28having enough diversity of thought in the room that that
- 46:31the the brand new person can say, wait, the emperor isn't
- 46:34wearing any clothes. Greg Walters asks an interesting
- 46:37question. He's another regular listener,
- 46:40he says. Can you extend your views regard
- 46:43to embodied AIS and LLMS, which is robots and AI in the 3D
- 46:50World? So how does this apply to robots
- 46:53and and again talk about data? So that's a new source of data
- 46:57that a lot of people are excited about.
- 46:59And it is something where we're going to need multiple AI
- 47:03approaches because you want to give commanders intent to the
- 47:06robot to say, grab that orange. But then as the actual actuators
- 47:10on the arm of the robot and the hand of the robot are trying to
- 47:13grab, you want to be adaptive. And so, yes, that is very rich,
- 47:17but obviously it's a place where you need to make sure you're
- 47:19applying the right tools. I would recommend if you're
- 47:22trying to actually navigate the world, we're not going to
- 47:25probably look for generative AI for the tools that we're going
- 47:28to look. We're going to look for more
- 47:29Bayesian approaches and models that actually give a sense of
- 47:32the AI itself has built a world model of whatever the robot is
- 47:35trying to do. If we think of embodied AI as an
- 47:39extension of edge computing, where we have more and more
- 47:42processing power and more and more autonomous data that sits
- 47:45out at the edge being used, this is not a new problem that
- 47:50suddenly came up because we called it embodied AI.
- 47:53It's a problem that we've been dealing with in the IoT, in
- 47:56autonomous devices and so forth. So we should learn from that
- 48:00body of knowledge and not start to like, reinvent a whole new
- 48:04field here. Yeah, it looks like a human, but
- 48:07it's, it's an edge device that, you know, has autonomous data.
- 48:11And I, I, I know a little bit, I, whoever I am, know a little
- 48:15bit about what we've learned so far about that.
- 48:17And you know, we can start from there.
- 48:19Wei Wang says on LinkedIn regarding data misuse.
- 48:25Are there scenarios where we will be able to quickly track,
- 48:29delete, or restrict access to these risky open data sources to
- 48:35prevent harm? Track yes.
- 48:36Delete. Probably not address yes.
- 48:40So you hit you nailed it right. Try to take something off the
- 48:45blockchain, try to take something, erase something from
- 48:48the Internet really hard to unring the bell, arguably
- 48:51impossible. But that doesn't mean that you
- 48:53can't publish the Contra positive information, and it
- 48:56doesn't mean you can't train your models to recognize lying
- 48:59liars. Lying, right?
- 49:00Veracity adjudication is something I've spent a
- 49:02tremendous amount of my career building approaches to looking
- 49:06at the truth, the whole truth, and nothing but the truth.
- 49:07That's three different dimensions of how do I know
- 49:10what's true? We can do all of this stuff.
- 49:12There's lots of ways to address that problem.
- 49:14Unfortunately, one of them isn't claw it back from history.
- 49:18Even the Wayback Machine has limitations.
- 49:21Chris Davidson comes back and says do systems currently exist
- 49:25that are successfully integrating today's LLM based
- 49:29Gen. AI with active inference and
- 49:32what might that look like? Are they completely different
- 49:36from one another? There are early stage, but I
- 49:39wouldn't say there's anything that we can publicly point to.
- 49:41I don't know if Anthony has of anything we can probably point
- 49:43out. I know there are things going on
- 49:45behind. This.
- 49:46That's why I'm smiling. No, there are not.
- 49:52Nothing public not to say that, but it's a good question.
- 49:58Chris Peterson is thinking very broadly and he says should we be
- 50:01thinking of AI like the old old school arms race, US versus
- 50:07China versus high end chips to UAE since it can be done by
- 50:11anyone? Could there even be an AI
- 50:13version of a non proliferation treaty?
- 50:17No, we should be thinking of AI like physics in in the 1940s,
- 50:21right? Physics was there before nobody
- 50:23discovered physics. Physics was there and people got
- 50:28good at physics, were not good at physics, and they shared what
- 50:30they were learning or they didn't share what they were
- 50:32learning and they used it for good or not good.
- 50:34It's a lot more like that than it is like something that can be
- 50:38somehow corralled into national boundaries.
- 50:40I'm sorry, David, did you have a different opinion?
- 50:42I'm spot on to what Anthony was saying.
- 50:44I tell people it's just math. It's very fancy math, but it's
- 50:46just math. I would also say let's learn
- 50:48from history because back in the late 1800s, all the nations of
- 50:52Europe got together and said chemical weapons really bad,
- 50:54let's ban them. So in 1899, they banned them.
- 50:56And come World War One, guess what?
- 50:59Both sides have been quietly researching how to do it and
- 51:02they have done it in quiet. And so anyone who thinks some
- 51:05international agreement will save us from ourselves, I hate
- 51:08to say it, no. But again, you have to recognize
- 51:10it's like physics. And so we have to come up with
- 51:12other answers, yes. I heard a great quote yesterday
- 51:15and I wish I knew who said it. The the the big difference
- 51:18between genius and stupidity is that genius has limitations.
- 51:22Arsalan Khan points out that humans collect and create data
- 51:28and then humans are replaced by AI that uses that data.
- 51:32Accenture just laid off 11,000 people and because of because of
- 51:37AI and this is an increasing trend comment very quickly.
- 51:41It's hard to distinguish what is, what is the case of people
- 51:44being laid off because of AI? What is the case of the nature
- 51:48of business changing. We also live in a shareholder
- 51:50economy where right now shareholders will actually
- 51:53reward companies that reduce their cost.
- 51:55And the number one way to reduce your cost is workers.
- 51:57Now you can say it's because of AI, but it might just be I'm
- 51:59trying to reduce my cost. And so the nature of work is
- 52:02changing and will continue to change.
- 52:04But to say it's because of AI I think is a oversimplification.
- 52:08AI can make a pretty good PowerPoint deck and and then all
- 52:10I have to do is edit it right? So I don't need a room full of
- 52:13people to get me 80% there. But the reality is that AI is
- 52:19being super helpful in many different areas and vibe coding
- 52:25for example has its role. For example, I'm not a developer
- 52:30but if I have but I can use vibe coding to create an app that is
- 52:35useful. And there are now tools that let
- 52:38you integrate. Lovable lets you integrate with
- 52:41the back end and it will create the database and the security
- 52:45roles and so forth. And you know what?
- 52:47I can use that tool internally for my own organization.
- 52:50I can even sell it. More and more people will become
- 52:53solo entrepreneurs. I think that is the shifting
- 52:55nature of work. And so we are definitely seeing
- 52:58a work change. I want to give some people some
- 53:00hope though, too. One of the things I'm seeing
- 53:02that's really an interesting trend when it comes to data and
- 53:04AI is more and more people with rare diseases or a family member
- 53:07with rare diseases are actually asking for their medical
- 53:10records. And, and they're actually then
- 53:13using the tools in health level 7, which is a nonprofit that
- 53:15actually makes sure you can request those records and
- 53:18they're in a way that you can actually analyze.
- 53:19You can now actually get your medical records in a digital
- 53:22form and then go to your whatever AI you want to and say,
- 53:24what has this doctor missed or what else should I be thinking
- 53:27about in terms of treatment for myself or my family members?
- 53:29So that's, that's a positive sign.
- 53:31I do think we need to help people cross the chasm between
- 53:34the work they used to do and the work they're going to do,
- 53:37whether that's community efforts, whether that's company
- 53:39efforts. The nature work is definitely
- 53:41going to change. Nature work has been changing
- 53:42more. I mean, most of us would have
- 53:44been farmers 200 years ago, and we're not anymore.
- 53:46My greatest hope, Michael, is that anyone that is
- 53:49disintermediated or feels disintermediated by AI doing
- 53:52their own job feels liberated to go do something new and more
- 53:57amazing because there's plenty of David's talking about, you
- 54:00know, known unmet needs that we, we know we're not addressing
- 54:05these problems. There's also those unknown unmet
- 54:07needs that we haven't even discovered yet.
- 54:09Oh my gosh, smart people, please go do some of that instead of
- 54:14rebuilding a better, you know, whatever.
- 54:17What do you say, Anthony, to the person who is being displaced,
- 54:24who has kids in college, who has medical debt and is living
- 54:31paycheck to paycheck? You know, telling that person
- 54:35you should do something new is not very helpful.
- 54:40I said that was my hope. That is certainly not the way I
- 54:44would approach that person in that particular situation at
- 54:47that moment. And don't think I don't get
- 54:49calls from people that are in exactly that situation.
- 54:51I'm sure David does almost several a week, you know, and,
- 54:56and you know, I start the conversation.
- 54:58There's a, there's a, you know, fine line between being helpful
- 55:02and being annoying. But I, I try to start the
- 55:05conversation by understanding, you know, the grief and the,
- 55:08the, the frustration and the anger because as best I can,
- 55:12because it's not me enduring that.
- 55:15But then from there, once that person gets beyond that, very
- 55:19often the people that we're talking about, when you look
- 55:23back on it a year later, it that was the best thing that ever
- 55:25happened to them because they don't chain them to go do this
- 55:28other thing so that the journey is, how do you figure that out?
- 55:32What do you have? There's that moment in The
- 55:33Princess Bride where they say it's hopeless.
- 55:35We can't attack the castle until they realize they have a cloak.
- 55:38Oh, now we have a plan, right? What's your cloak?
- 55:41What is the thing that you have that that creates the, the, the,
- 55:46the, the, the rock that you can stand on?
- 55:48And I know you know, you, you, it might be that you have to go
- 55:51do something you don't want to do in the short term because of
- 55:54all those needs that you have. I am not at all unsympathetic to
- 55:57that. I'm in no way saying let them be
- 56:00cake. But what I am saying is that all
- 56:02of this abundance of capability that we have that is right now
- 56:06largely being used to create a bunch of party tricks could be
- 56:11used to do some amazing things. And boy oh boy, you get to do
- 56:15some of that. You want to acknowledge and
- 56:17recognize the grief and the anxiety that anyone experiencing
- 56:20that is having at that moment and then at the same time strive
- 56:24to find ways to help them see that they can be a survivor and
- 56:28a thriver as opposed to a victim.
- 56:29As to what happened here. I think that also does a
- 56:32requirement on all of us who are not in that unfortunate position
- 56:36to advocate. What is the communal response?
- 56:38What is the response that we expect from people?
- 56:41Because we're not advocating for it.
- 56:42It will happen to one of our friends or one of our family
- 56:44members and I know it. I mean, I probably do get about
- 56:47a call about once every week or so.
- 56:49And then obviously, I'm here in DC where a large number of
- 56:52people might actually be facing the same situation, not because
- 56:54of AI, but I put that out because we do need to think
- 56:59about in this era of displacement, do we want to, can
- 57:03we find ways we should find ways to reduce the anxiety and to
- 57:06help people navigate this? And I feel like right now
- 57:09there's more we can do as societies, as communities, as
- 57:12countries. On that, do I say I can say
- 57:16empathetic note? Can I say hopeful note?
- 57:19I'm not. Sure, we're going to.
- 57:21We have to revisit. This Yeah, I think we're both, I
- 57:24don't want to speak for David, but I think we're both hopeful
- 57:27that, you know, humans are amazingly resilient and
- 57:30amazingly good at, you know, turning corners and doing things
- 57:34that surprise and delight us as well as terrify us.
- 57:38We get to choose which of those we do.
- 57:39And I'm, I'm on the camp of surprise and delight.
- 57:41So we'll, we'll get there. This is a time we will look back
- 57:45on and say that's exactly when we were able to start doing fill
- 57:49in the blank. Maybe it's vibe coding Michael,
- 57:51I don't know, but it's probably not.
- 57:53Or maybe just teach yourself to be an AI red teamer, which I
- 57:56would recommend instead. But.
- 57:57Yes, that's me. Yeah.
- 57:58I mean, you know, far be it from me to express an opinion, but I
- 58:01think that, you know, we are certainly at a point of
- 58:04inflection. The most difficult thing about
- 58:06change is it's very hard to notice when you're part of it.
- 58:08Can I add one final comment that I think we can all agree on?
- 58:14Down with AI slop. Yes.
- 58:19The trick though is identifying what is AI slop.
- 58:21But anyway, that's a doubt. And on that note, a huge thank
- 58:29you to Doctor David Bray and Doctor Anthony Scrifignano.
- 58:33It's been a great show. Thank you both again for taking
- 58:37your time to be with us. I'm very grateful to you both.
- 58:40Thank you, Michael. Thank you very much.
- 58:43And an enormous thank you to the audience.
- 58:47What will happen next is we will edit this videos, do some light
- 58:52editing, create a summary, we'll put it on the CXO Talk website
- 58:57and then with the transcript and the summary, you will have a
- 59:02document that you can keep, go back treasure and refer to.
- 59:08So I urge you to do that. Subscribe to the CXO Talk
- 59:12newsletter. Thanks so much everybody.
- 59:14Thank to thanks to Doctor David Bray and Doctor Scrifignano.
- 59:18Take care everybody, have a good one.
- 59:19Bye now.