Latest / CXOTalk / AI Ethics and Trustworthy AI: Navigating Truth and Deception | #864
Transcript
- 0:00Welcome to CXO Talk Episode 864. I'm Michael Krigsman, and today
- 0:07we're exploring how to navigate truth and deception in AI as
- 0:12part of a special series on AI ethics.
- 0:16Our three expert guests are Doctor Anastasia Lauterbach,
- 0:21Doctor David Bray and Doctor Anthony Scribignano.
- 0:25We got a lot of doctors in the house.
- 0:27All of you, welcome to CXO talk. I'm delighted that you're all
- 0:31here. Thank you very much.
- 0:33Thank you for having us. Great to be here with.
- 0:37Anastasia, let's start with you very briefly.
- 0:40Tell us about your work. I spent most of my time with my
- 0:44company. I'm a founder and CEO of AI
- 0:46Entertainment and this company democratizes knowledge of AI
- 0:50with AI muggles, not Wizards. And this means that I try to
- 0:55translate AI, robotics and quantum computing from difficult
- 0:58to easy to understand. For example, I write books like
- 1:02this. This is Rome and Robbie series,
- 1:04and this is a real story about real Cat and his friendship with
- 1:11a robot and everything. What is going on in this book is
- 1:14mirrored in the science behind section.
- 1:16Those are over 100 articles explaining AI in very easy
- 1:22language. I love it a an AI cat.
- 1:25Absolutely. He's an influence.
- 1:27And by the way, did you know that 27% of web traffic is a cat
- 1:33connected? A cat linked content?
- 1:36Cats are addictive. David, tell us about your work.
- 1:39I try to bring rationality to the world in terms of how
- 1:42technology is changing politics, geopolitics, societies and the
- 1:46like. Do so both as a distinguished
- 1:48fellow with the Stimson Center as well as business executives
- 1:50for National security. And then on the side have my own
- 1:53S Corp which tries to help boards and CE OS do the same.
- 1:57Anthony Scrifignano, welcome back.
- 1:59Tell us about your work. I'm a distinguished fellow with
- 2:01the Stimson Center along with David Bray.
- 2:03I. My philosophy is to try to teach
- 2:05and learn every day. No fair cheating on that.
- 2:08You have to do both. I've done a lot of work over 45
- 2:12years with data and data science and AI before it was a thing.
- 2:16I've gotten multiple patents on my name.
- 2:18Lots of invention around things like identity resolution,
- 2:22veracity, adjudication, which we'll talk about today, judging
- 2:25the truthiness of things, and also lots of effort around
- 2:29finding people that are doing new bad things.
- 2:32So novel malfeasance. So these are all the things that
- 2:35are kind of at the edge of computer science and among the
- 2:39hardest of the hard problems to solve.
- 2:41Anastasia, Why is building trustworthy AI versus deceptive
- 2:48AI important? Trust is something tremendously
- 2:52important for human psychologist, one of those
- 2:55constructs which is paramount for human relationships.
- 2:59So no one is going to disclose or I am building something which
- 3:05is based on deception. And they are two economical
- 3:08arguments in favour of paying attention to whether AI is
- 3:13considered trustworthy or not. And the first, I would open up
- 3:18with a quote by Kevin Kelly, who is founder of Wiret.
- 3:21He said that the business of the next 10,000 companies is very
- 3:26easy to predict. You take X and you add AI to it.
- 3:30And obviously this remark which was made I think like around
- 3:342016, 2017, he is now based in figures, in financial figures.
- 3:42For example, around 2013, we expect the EI market to grow to
- 3:491.8, see 5 trillion U.S. dollars globally.
- 3:54And I mean this is tremendous growth.
- 3:55We are talking about compound annual growth rate of 33% from
- 4:0022 to 2030. And there's another caveat to
- 4:04that. When I looked into valuation of
- 4:07businesses around 2012, I could see that the portion of digital
- 4:13assets in this valuation of companies was around 12 to 13%.
- 4:18And after the COVID epidemics, we are talking about 90%, so 90.
- 4:25And digital assets, it's obviously all about data, it's
- 4:29about intellectual property. It's all kind of systems and
- 4:33processes to do something with this data.
- 4:36So this incredible growth has something to do with the
- 4:40progress in AI and we have to pay attention.
- 4:44If I were to summarize what Anas Tasha just said, everybody's
- 4:49leaning into this and everybody's leaning into this,
- 4:51but it's a different this. So if you look at what a lot of
- 4:54organizations want is they want to be able to check the box and
- 4:57say we're doing AI, we're doing Gen.
- 4:59AI, we're doing large language models.
- 5:00Look at us, we're great, right? That's great.
- 5:02What's the problem you're solving and by what right do you
- 5:05think that that thing that you're doing is going to help
- 5:07address that problem? There's in many cases a great
- 5:11argument to be made that you're just accelerating the speed with
- 5:15which you hit the wall. You're helping people understand
- 5:18what you don't know. You're helping people understand
- 5:20more about your questions than about your answers.
- 5:23You're exposing yourself to potentially looking to your
- 5:26customers as if, you know you focused on that and not this
- 5:30other thing that they want you to focus on.
- 5:32So it's, yeah, it's really important to move into this
- 5:35technology. Never would I say don't do that,
- 5:38but I would say, you know, just because you see a new tool at
- 5:41Home Depot doesn't mean you go buy it.
- 5:43You have to have something to address and you have to have a
- 5:46problem that you're trying to go after.
- 5:48And you also have to understand the unintended impact of those
- 5:50actions you're about to take. May I just quote a poet from
- 5:55England from the early 18th century, Alexander Pope?
- 5:58He said trust not yourself, but your defects to know.
- 6:02And this is actually emphasizing that it's not just
- 6:05self-reliance, but it's about humidity.
- 6:08And I think this humidity and thoughtfulness is something
- 6:11which we need to learn more and more and apply more and more
- 6:14when we move into the world with AI.
- 6:17One of the questions I always ask when when someone is, you
- 6:20know, running headlong into, you know, let's AI our way out of
- 6:24this problem is, you know, two questions actually, 1 is what do
- 6:28we have to believe in order to go down that road?
- 6:31And the second thing is, but by what right do you think the
- 6:34answer is contained in what the AI is going to look at to answer
- 6:38that question? Imagine during COVID if you had
- 6:41ChatGPT and you said, you know, what are the most promising
- 6:45approaches to responding to COVID?
- 6:47Well, we didn't know, right? And so you would, you would
- 6:50literally get what we love to call hallucination right now,
- 6:53but on steroids, right? Because the LLM doesn't know
- 6:56that the answer isn't in the corpus of data that it's looking
- 6:59at. And it's going to give you an
- 7:01answer anyway. And it's going to look awfully
- 7:02good. And it's going to really look
- 7:03like a human being wrote it. And it's going to be wonderfully
- 7:07articulate, but just as good as any other rumor you just heard.
- 7:11So can we say that the summary here to make this very
- 7:16simplistic is AI is around us, AI is surrounding us, AI will
- 7:24continue to do so and therefore we need to know what's true and
- 7:28not true. I mean, that seems pretty
- 7:29simple. Perhaps instead of calling, when
- 7:32we say AI, artificial intelligence, we should call it
- 7:36alien interactions because these machines are not at all thinking
- 7:41like you or I. And that's a very real danger
- 7:44when you read articles or for whatever reason begin to
- 7:47anthropomorphize these machines. I mean, just statistically, I
- 7:51mean, the human brain uses between 15 and 20 watts to do
- 7:54everything it does. These models are looking at
- 7:56upwards of 2005 thousand or more.
- 7:59I mean, when we hear like companies thinking about doing
- 8:00nuclear power plants to do their alien interactions as AI, that
- 8:05might tell you that this is simply been designed to look
- 8:08like it's thinking like a human, when in fact it's not.
- 8:11That then also gets to the deeper question I'd like to say,
- 8:13Michael, which is if you remember the Turing test, the
- 8:17Turing test was intentionally trying to have a machine deceive
- 8:21a human into thinking it was human.
- 8:24Guess what? We have been wildly successful
- 8:27at rolling out technologies that will now convince you that the
- 8:30responses are giving whether they're in text, whether than
- 8:32audio, whether in video appear to be human like.
- 8:36But that's a problem because as Anthony was mentioning in his
- 8:40background, I mean, fraudsters, scamsters, they would love to
- 8:43use this technology. And we've all probably seen in
- 8:46the last two years a step change in scam phishing emails where
- 8:49they are now at the point where you can no longer use
- 8:51misspellings as they tell if it's a phishing e-mail or not.
- 8:54And now even better, they're often referencing another family
- 8:57member or friend because they've combed our social media network.
- 9:00And so the are actually intentionally, unfortunately
- 9:04really successful, the Turing test.
- 9:06But the whole Turing test was again a machine trying to
- 9:08convince you it was human. That might not be the best tool
- 9:11exactly. We've been saying here for
- 9:13society writ large. There's a another little nuanced
- 9:16thing happening right now where we now have to prove to the
- 9:18machine that we're a human right.
- 9:20So it's sort of the opposite of that.
- 9:22And if you look at CAPTCHA and the way it was originally
- 9:25designed, it was just OCR was kind of not very good.
- 9:28So if I tilt the letters and I put them together, then OCR
- 9:32doesn't work and the human can read it and we know the answer
- 9:34and that's great, right? Well, then people started to use
- 9:37the, the responses from CAPTCHA to train better OCR.
- 9:42And we'll, lo and behold, all of a sudden the computer gets just
- 9:44as good as we are at reading kind of crappy text.
- 9:47So then they start introducing pictures.
- 9:49And introducing pictures is not a, it's not an insert, it's not
- 9:54an end to that. It's just a way of kicking that
- 9:57ball a little bit further down the road.
- 9:59I was listening to a talk recently with one of the people
- 10:01who actually invented the technology, and he was saying #1
- 10:05he wished he never did. And #2 you know, maybe in the
- 10:08future, you'll have to do things like walk three steps with your
- 10:12phone or jiggle your phone or tilt your phone down or do
- 10:15something. And, and now I can just imagine
- 10:18us OK, do 10 push ups. Like, at what point do I have to
- 10:21stop proving that I'm human? We don't have a good answer to
- 10:24this right now. It's not because the AI is
- 10:27thinking. It's because it's very good at
- 10:29watching us behave. That's so true.
- 10:31And actually we need to establish a language to talk
- 10:34about AIS and machines, and we're still using human terms to
- 10:39describe what we expect. And in social science, for
- 10:42example, tries, trust is described as the belief that
- 10:45another person will do what is expected.
- 10:48And now it's not a person, it's a machine or piece of code and
- 10:52what is really expected. If I use my vacuum cleaner, I
- 10:55expect it to do a certain task. But this is very different from
- 11:01LLM. And obviously because I am an AI
- 11:03and I teach AI and cybersecurity at the university, I know it's
- 11:07has nothing to do with language. It's a statistical stream of
- 11:11tokens and token isn't even a word.
- 11:14But if we take, for example English, it's maybe 2/3 of an
- 11:17English word. Yeah, statistically.
- 11:20But you know, humans expect that it understands and it's simply
- 11:25doesn't, and even not based on the world model.
- 11:29A couple of things to sort of add to the nuances.
- 11:31First, let's go. Let's take the 30,000 foot view
- 11:34before we dive in, which is back in the 1920s when this
- 11:38disruptive technology called radio came out, there were
- 11:41pundits that were saying something similar to what we're
- 11:43seeing with AI nowadays, that this was going to cause World
- 11:46Peace. It was going to have
- 11:47understandings. World leaders would talk to each
- 11:48other on a really regular basis on radio.
- 11:51And then fast words in 1933, about six or seven years later,
- 11:54those same pundits were saying that this was going to be the
- 11:56end of society as we know it. It's going to be the dictator's
- 11:58tool kit. And so I raise that because
- 12:00right now there are a lot of breathless articles that are
- 12:03either saying that AI is going to save us all and the, you
- 12:05know, the good times are upon us or is somehow going to be in the
- 12:09societies. We're probably doing the same
- 12:11thing to the technology that we did to radio and we're missing
- 12:13that. Again, it's just a tool.
- 12:15Now, the second thing I would say is I think we are often
- 12:19using AI as a stalking horse or as scapegoat for things that are
- 12:23deeper in society. Like we ask questions like, how
- 12:26do you know if AI is ethical? Well, how do I know if an
- 12:28organization is ethical? How do I know if ethicalization
- 12:30is not bias? How do I know if an organization
- 12:33is not making bad decisions or hallucinating?
- 12:36And So what this may point to is we need to do a better job of
- 12:39figuring that out and how we actually do it, whether it's a
- 12:41machine, organization, or person.
- 12:44Now I will give a slightly different definition of trust
- 12:46from the background I come from, which is the willingness to be
- 12:49vulnerable to the actions of an actor you cannot directly
- 12:51control. And that actor could be an
- 12:54individual, could be an organization, could be a
- 12:55government, could be a machine. And that what it's been shown is
- 12:59we humans are willing to be vulnerable to those actions of
- 13:02an actor that we cannot control if we perceive benevolence,
- 13:05perceived competence, and perceived integrity.
- 13:07So exactly down to Sasha's point and Anthony's point, there is no
- 13:10way of assessing the benevolence of a LOM.
- 13:13There is no way to assess the competence.
- 13:15I mean, these things are just doing very fancy pattern
- 13:17matching. And in fact, if it's not in the
- 13:19training data, they will give you something that is made-up,
- 13:23but then also an integrity. You say to them that's not
- 13:25right, They will be again, the the prompts will be very
- 13:28cheerful in saying you're right. And then when you say that's not
- 13:30right too, they're right, you're right again.
- 13:31And so integrity, competence and benevolent aren't there.
- 13:35But let's step back and say, how do people assess that?
- 13:37Given the fact that we're now connected digitally, How do we
- 13:40assess that? CXO talk is benevolent competent
- 13:42integrity. How do we select, you know, do
- 13:44that for governments? How do we do that for the world
- 13:46that is large? And maybe this is the challenge
- 13:49hitting societies, free societies in particular, is we
- 13:52lack the ability to adjudicate at speed.
- 13:54A lot of things, the terms that you're using, David, are also
- 13:57epistemologically fuzzy, right? So benevolent might be
- 14:01benevolent from your perspective and not so benevolent from my
- 14:05perspective. And we tend to use terms you
- 14:08didn't, but we tend to use terms like we're good.
- 14:10Well, good for whom, right. And you know, I, I have a, a lot
- 14:16of background in emergency medicine, right?
- 14:18So one of the things they always tell you is expand the circle.
- 14:21You know, if you're going to make a complicated decision,
- 14:23get, get advice from other people and make sure that, you
- 14:26know, those other people will bring in different perspectives.
- 14:28Well, that's great until something's on fire or until
- 14:31somebody's not breathing, and then somebody has to make a
- 14:34decision, and it might be the wrong one, but you have to make
- 14:37that decision with some degree of cut.
- 14:38Now, after the fact, everybody will swoop in and judge you that
- 14:43you should have done this, you should have done that.
- 14:44Why didn't you do this? How didn't you know about this?
- 14:47At some point, AI is going to be making decisions for humans
- 14:51because the argument will be that AI can make a decision
- 14:55before the human decision becomes irrelevant, right?
- 14:58Do I cauterize this vessel or that vessel make a decision
- 15:01right now? I don't.
- 15:03Do you know all the vasculature in the brain and can you let the
- 15:06AI do it right? I don't know.
- 15:09But I could see how we can get there and I could certainly see
- 15:11how we can get there in in a nation state, military
- 15:14industrial kind of context. I could we can get there in, you
- 15:18know, in space. There's, you know, it takes so
- 15:20long for a signal to get from one place to another that you
- 15:22need to have autonomy in order for the thing to be able to land
- 15:26or do whatever it's trying to do.
- 15:27But as we give up this agency, this concept of agency, of
- 15:30giving the right to the machine to make the decision on the
- 15:33behalf of the human, the values of the human have to be imposed
- 15:38into that what is just a bunch of code.
- 15:41And we are not rules based creatures.
- 15:44We are very, very squishy about what we do and we make decisions
- 15:48and we change our opinion based on new facts and new modes and
- 15:51new ways of learning all the time.
- 15:53AI is horrible at that right now.
- 15:56Please subscribe to our newsletter and subscribe to our
- 16:00YouTube channel. Check out cxotalk.com.
- 16:04We have great shows. Coming up.
- 16:06We have a question from Twitter and this is from Arsalan Khan
- 16:13who says we know that AI can be trustworthy or not trustworthy
- 16:18based on many factors. How can normal non-technical
- 16:22people know if AI is affecting their lives, if it is ethical?
- 16:27And is there an opt out for AI? And we know very often there's
- 16:33not. And then let's also talk about
- 16:37the role of software companies. AI is surrounding us.
- 16:41So if if you have a smartphone, you immerse in AI.
- 16:46So we can really discuss all those technologies like, you
- 16:49know, LLMS are on smartphones, but even how we see our calendar
- 16:54invites and how it correlates with tweets or whatever, some
- 16:58notifications from LinkedIn, it's immersed in AI.
- 17:01Some navigation technologies are connected to automation and to
- 17:06AI. And we just need to think how
- 17:09should we behave ourself. And frankly, unfortunately, and
- 17:14this is I guess the difference to radio technologies and
- 17:17everything which proceeded, we must expose ourselves and learn.
- 17:23And I think that EI literacy and technology literacy is something
- 17:27which is a must from an early age.
- 17:30It's like basic financial literacy.
- 17:32Obviously, there will be always some treasury experts and
- 17:36taxation experts and whatever some valuation experts.
- 17:39But everyone must understand that this revenue and this cost
- 17:45and 1 minus another produce some kind of, and I kind of love
- 17:51comparing AI and you know, dealing with AI with the
- 17:55kitchen, right? Maybe because I'm a woman.
- 17:58And obviously you might be a great chef even if you don't
- 18:02work in high cuisine. And you can produce fantastic
- 18:07stuff and be very great and performance some shows.
- 18:10And probably this is the ultimate level of
- 18:13sophistication. But everyone knows how to fry an
- 18:16egg, and I guess with the eye we must learn how to fry an egg.
- 18:20And this is basically why I learn hits and their clever
- 18:24appearance to think about those concepts.
- 18:27What is an LLM? What does it do?
- 18:30Who is Wolfram? Why should we know about him a
- 18:33little bit? What does it mean to have a
- 18:36robot in our house? What is the difference between a
- 18:40robot and let's say, a vacuum cleaner?
- 18:43Right? So I think we just must embrace
- 18:47this knowledge and actually evolve as communities to talk
- 18:52about it, to ask difficult questions, and to understand
- 18:56that there is a purpose behind every single application and
- 19:00service. You can't go through a jungle.
- 19:03You must learn and then you will have some openness to maybe be
- 19:09more precise on what you need. And finally, I love the quote of
- 19:13Pablo Picasso, he said. Machines are quite stupid.
- 19:16They just produce answers. And I think this is up to humans
- 19:20to ask questions, and we need to teach humans how to ask those
- 19:25questions. And this is basically a fabulous
- 19:27opportunity that is opening up in front of all of us to learn
- 19:32more about humanity and what it really means in the world, which
- 19:37is partly dominated by AIS. In Western society, if you apply
- 19:43and your job application has a non western name, you are less
- 19:47likely. A western name application is
- 19:49three times more likely to be selected.
- 19:51And that's not an AI problem whatsoever.
- 19:53Everything we've been talking about here is like, oh, how do
- 19:55we trust the machine? I'm like, no, no, no, no, we
- 19:57have deeper systemic issues. And so almost it's like we're
- 20:01almost pitching AI and saying, how do we ask these questions of
- 20:04AI before we've even paused and say, how do we actually do
- 20:06better as humans? And I really appreciate it.
- 20:10So I would also say that as we go forward from this.
- 20:14It's also thinking about that we've had these challenges
- 20:16before. I mean, most of us are not
- 20:18medical doctors yet. We have to go see a medical
- 20:21doctor. How do you know when the medical
- 20:23doctor is going to prescribe a approach that they really have
- 20:26your best interest in mind? And the way we saw this in the
- 20:29past, and unfortunately, the Internet kind of squashed this,
- 20:32was we did have professional societies that require their
- 20:35members to have knowledge of something, experience in
- 20:37something, and then pledge to an unethical oath.
- 20:40And if any point in time, the member had a concern because
- 20:44again, most people cannot adjudicate, did that doctor act
- 20:47in your best interest or not? It was other medical doctors
- 20:49that would decide to either say, yes, it did the right thing or
- 20:51not, we're going to censor them or going to move their license.
- 20:54Well, unfortunately, what happened with the Internet is it
- 20:56made it so that everybody could be armchair quarterbacks.
- 20:59But I raised that because I worry that we will become so
- 21:01fixated on we must understand AI, we must understand and
- 21:04educate and everything like that without pausing and saying right
- 21:07now there are plenty of organizations where there is
- 21:10issues internally to bad decision making, bias decision
- 21:13making as well as externally and how they're actually interacting
- 21:16with employees or customers. And so I think we actually need
- 21:19to figure out a deeper approach, not just about the machine.
- 21:22Implicit in what you have all been saying is this idea that
- 21:30benevolence is our fundamental goal and we rely on the
- 21:36benevolent human spirit. And we hope that the Anthony's
- 21:39shaking his head. Well, this is my interpretation.
- 21:42So let me ask my question anyways.
- 21:45Before it's even edit, before it's edited, before it comes out
- 21:49of my mouth. So Anthony's trying to edit it
- 21:52in my brain. So my question.
- 21:55So my question is this. If we think about the software
- 21:57companies, over my career, I've consulted with over 100 software
- 22:03companies literally. And as far as I can see, the
- 22:08goal of software companies is innovation with the goal of
- 22:12making money. And yes, software companies talk
- 22:15about we want to change the world.
- 22:17And I'm sure that there's that. That's also true.
- 22:20But the bottom line is it's about money.
- 22:22And so when we talk about trustworthy AI versus deceptive
- 22:28AI, how do we overlay this software company issue on top of
- 22:34it? Anastasia thoughts on this?
- 22:3530 years ago I was working at the Munich reinsurance company
- 22:39as a liability on the right and even now there's no not such a
- 22:43thing as a software liability within the product liability
- 22:47category. So European Union has updated
- 22:51product liability directive, but the rules will be introduced
- 22:55only in December 2000 26 and in the United States, to my
- 23:01knowledge, please correct me if I'm wrong, so far they aren't
- 23:04any kind of, you know, rules which are saying that software
- 23:08vendors are within this liability rules what what is
- 23:14applicable to for example, automotive companies or consumer
- 23:18electronic and whatever. So we need to somehow calibrate
- 23:22what we are talking about. For example, in July this year,
- 23:26there was a bug which was actually played into a software
- 23:32over an overnight update with a cybersecurity vendor.
- 23:37And then all Windows machines went down at the airports and I
- 23:42think around 5000 flights were either cancelled or delayed
- 23:47globally. And I think in the US the figure
- 23:49of financial damage which is known today is for 5.4 billion
- 23:56U.S. dollars. So damage from this update, so
- 23:59there is some legal process which is going on, but so far
- 24:04actually that there will be some damages paid.
- 24:07So this prospect is really, really low because it's very
- 24:11difficult to determine what is the negligence or misconduct
- 24:15here. Then tort law actually excludes
- 24:18a software from this liability things.
- 24:21And then obviously, if there are multiple parties involved, who
- 24:25did what and how, this is really, really complex.
- 24:29And if we are, for example, with food industry or Pharmaceutical
- 24:32industry, the world looks completely different.
- 24:35So the world is being eaten by software.
- 24:38So Marc Andreessen said it in August 2011.
- 24:41But there is no product liability for software vendors.
- 24:46And now we are going into AI. Yeah.
- 24:48So fantastic. So European Union is
- 24:52implementing the European AI Act, so all vendors must comply.
- 24:57It's really killed AI ecosystem in Europe.
- 25:01By the way, I, I represent myself, I don't represent any
- 25:04big grant. So I was really against the
- 25:07European AI Act because it did not solve one single risk in AI.
- 25:13It impose huge amount of costs on AI companies and investors.
- 25:19Venture capitalists are saying we are really our fingers off
- 25:23the European ecosystem because it's all too costly and we need
- 25:28now to balance what do we want. At the same time, 5 European
- 25:32countries, which is Italy, Belgium, Austria, Germany and
- 25:35Netherlands are moving into mass retirement this decade.
- 25:40And if you have something like that, you must think about
- 25:43automation and AI. And there's no policy, there's
- 25:47no thinking how to actually balance the inevitably declining
- 25:51GDP due to this mass retirement. And we have the European AI Act,
- 25:56so Europe shoot itself into the knee.
- 26:00This is my interpretation with all the caveats and that this is
- 26:04complicated, that they are issues, that thoughtfulness is
- 26:08needed, and all of the above. I'm in violent agreement with
- 26:13your position on the dangers of regulations.
- 26:16Sort of. You know, I'm a scuba diver.
- 26:19The regulator's pretty important, right?
- 26:21You die without the regulator. But if if you over regulate,
- 26:24then you can't breathe. And there's a, there's a
- 26:27tendency right now to take whatever pre-existing laws there
- 26:30might be GDPR in this case and write something that looks like
- 26:33that for AI. Well, it's not the same thing.
- 26:36It's not even close to the same thing.
- 26:38And So what you wind up with is this tapestry of complexity that
- 26:42makes it very hard to take a step forward.
- 26:44Now, I'm not saying that that was the right thing or the wrong
- 26:46thing because like you, you know, I'm not, I've, I work with
- 26:49those people all the time, but I'm not in that space and they
- 26:52have to do something. I get it.
- 26:55The, the, the, the software issue that you talked about, the
- 26:59crowd strike thing, you know, was delivered kind of by
- 27:03Microsoft, right? Because most people had
- 27:04Microsoft. But you know, if you play it
- 27:06back and you unpeel the onion and who produced the blank file
- 27:10that got distributed with the update that got processed by the
- 27:14software that was acting like a driver that was allowed by the
- 27:17kernel of the operating system. This is a nightmare to say where
- 27:22where is the smoke detector that should have figured this out.
- 27:25And the world learned about how its own security was working a
- 27:29lot that day, right? So that was a very telling kind
- 27:34of moment for those in cybersecurity.
- 27:36They kind of understood it relatively quickly and it wasn't
- 27:39a big deal to fix. But Oh my gosh, the impact of it
- 27:43was just epic. That also tells a story about
- 27:46how connected the world is right now and how much we have to be
- 27:49careful as we take these steps forward.
- 27:52And so that is a counter argument for some regulation to
- 27:55say you can't just do whatever you want to do and apologize
- 27:58later because it had never happened before.
- 28:01So we're in a very difficult time right now In in
- 28:05epistemology, they're sometimes referred to these as critical
- 28:07incidents or critical moments where there's a point of
- 28:10reflection going on. It's hard to see when you're
- 28:12part of it, but we are definitely part of it and we're
- 28:15making little decisions right now that are going to have very
- 28:18big impact later with imperfect information.
- 28:21The these companies that your question started with, you know,
- 28:25three hours ago, Michael, the question was something about,
- 28:27you know, what companies can do. You know, companies are trying
- 28:31to serve their shareholders, their customers, their employees
- 28:33and their future market. And it's impossible to serve all
- 28:36those at the same time. Absolutely impossible.
- 28:38So if you want to maximize shareholder value, you wind up
- 28:41doing kind of really stupid short term things that often
- 28:44kill your company. If you want to maximize, you
- 28:47know, employee satisfaction and engagement, then you wind up
- 28:50doing things your customers don't care about.
- 28:52If you do everything your customers want, then you don't
- 28:54make any money because your margins go to zero and then your
- 28:56shareholders get upset. So it's just a whole bunch of
- 28:59ways to kill yourself. So now throw AI into this mix
- 29:03and then throw all this regulation into this mix and you
- 29:05get where we are right now in these boardrooms.
- 29:07This is not an easy place to be. Where are we?
- 29:10We are very similar to how when the automobile first came out.
- 29:14It's worth knowing how long did it take before seatbelts arrive?
- 29:18More than half a century now I'm hoping with AI we don't have
- 29:21that. But again, to try and think that
- 29:23we're going to immediately get it right within the 1st 6 to 12
- 29:25months, probably not going to happen.
- 29:27But let me zoom into a very specific sort of, and I think I
- 29:31Anthony sort of said this when we were talking about earlier
- 29:33about how when AI works and doesn't.
- 29:36And what I think we should unpack is not only AI is created
- 29:39equal, you know, we've been talking about generative AI, but
- 29:42there is other forms of AI ranging from experts rule based
- 29:45systems to decision support systems to computer vision.
- 29:48And I think the way you're going to trust, trust in computer
- 29:51vision, which is deterministic, which it is actually much more
- 29:56trustworthy and is not prone to any hallucinations whatsoever,
- 29:59is dramatically different than generative AI.
- 30:01And I raise that because I think, you know, if there is
- 30:04anything that needs to be talked about at the boardroom level,
- 30:07it's first understand the different tools when we talk
- 30:10about AI, that it's not monolithic and you need to
- 30:12understand whether you're reaching for a hammer or a
- 30:14screwdriver. And the trouble is we're writing
- 30:16regulations as if it is monolithic.
- 30:18And then the other thing that's actually to me is how we somehow
- 30:22think there's going to be 1 AI regulation to rule them all when
- 30:26we know when IT systems came online, these things called
- 30:29advanced data processing systems, later IT, you know, you
- 30:32don't write IT regulation for health and think it's just as
- 30:36good for the defense sector or just as good for the commercial
- 30:39recommendation sector. I mean, there are existing laws
- 30:41in the United States, there's HIPAA, there's the Bank Secrecy
- 30:44Act for banks. And so I actually think the more
- 30:46pragmatic approach is to go back to the existing rules that were
- 30:49written for both human and IT systems and say, where does AI
- 30:54fail? Because maybe it's too fast to
- 30:56speed or scope and upgrade those existing laws as opposed to
- 30:59trying to rate policy. And I'll give a very specific
- 31:02example right now in the United States, and it's not just the
- 31:05United States, there's a thing called health level 7.
- 31:06It's the international standard for interoperability across
- 31:09medical systems. There is a standard for tracking
- 31:12the provenance of a decision. You do record in your medical
- 31:15record when a physician made a decision and on what data they
- 31:19did. You don't record in the medical
- 31:21record when an AI does. And the number one usage right
- 31:24now for medical settings is actually physicians love to
- 31:27type, either type or dictate three to four short bullets and
- 31:30then say give me 20, sorry, 2 to 3 pages of physician notes that
- 31:35I will then put into the file. Well, aside from the fact that
- 31:38probably 1, both the company or organization that is delivering
- 31:41your health care should know was that physician notes actually
- 31:45written by the physician or written by an AI?
- 31:47And if so, which one? And the patient should know that
- 31:49too, too. There is a very real risk that
- 31:52if that AI output is later read back in by the same AI, there's
- 31:57a thing called model destabilization.
- 31:59Sometimes it's referred to as jargon as AI self
- 32:01cannibalization, where basically the model essentially it's, it's
- 32:05oversimplifying, but it over regresses and it starts to
- 32:07collapse on on itself and it starts to make bad decisions.
- 32:10And so I actually worry, Michael, that the question about
- 32:12deceptive versus trustworthy AI, again, it gets back to human
- 32:15organizations and what humans do.
- 32:18We may see lawsuits three to five years from now because
- 32:21companies didn't at least take the the first step of at least
- 32:24tagging and labeling when and a decision was made by an AI or an
- 32:29AI plus a human and then later if that was then fed back into a
- 32:32machine. Did they did they do the
- 32:35appropriate actions to avoid model destabilization?
- 32:38Really love this example from medical industry and it might go
- 32:41into further industries like food industry, but I have two
- 32:45further suggestions for regulators and for those who are
- 32:48serving on boards. So one is and by the way, this
- 32:51is not an ultimate solution, but we know that today more money is
- 32:56being spent on capability research than on transparency,
- 32:59safety, you name it, responsibility and it's like 97%
- 33:04of all the papers are published on behalf of those capabilities
- 33:08and only 3% on X AI. So this explainable AI and
- 33:12whatsoever. So we might motivate company to
- 33:15spend more money on that or to support a foundation or in
- 33:18institute university to invest into that.
- 33:22That could be motivational and not just punitive if we go into
- 33:27regulation. And then I'm really, I mean, I
- 33:29absolutely I'm scared about this issue with cybersecurity because
- 33:34AI is a tool, not just in the hands of good people, but
- 33:38criminals are using this really every single minute.
- 33:41And I think we just need to rethink how we approach
- 33:44cybersecurity and how we regulate cybersecurity, things
- 33:47like what is the exposure, cyber exposure, Can you quantify it?
- 33:53Can you name what is this financial figure?
- 33:56So this is tremendously important and very few people
- 33:59are talking about it. You mentioned quantum
- 34:02technologies early on and and certainly we are on the verge of
- 34:06the point where everything that was encrypted will now be
- 34:09readable, even the data that was stolen, right.
- 34:12We are on the verge of, we will have a way to do quantum
- 34:16decryption before we will have a really good way to do better
- 34:19quantum encryption other than scattering all the keys all over
- 34:22the place. So the world is going to get
- 34:25much more complicated. We have to start thinking more.
- 34:28And I'm not going to say start thinking, because there's some
- 34:30fine folks out there thinking about this, about novel cyber
- 34:34malfeasance. What are the types of not just
- 34:37the bad things that happened today with data exfiltration and
- 34:41and Trojans and malware and all this stuff.
- 34:44Yeah, we need to get really even better at that all the time.
- 34:48But that's necessary and not sufficient.
- 34:50We have to, I spend a fair amount of time and David, you
- 34:53know this thinking about what future bad guys might be able to
- 34:56do in in less than a generation with technology that I suspect
- 35:01will be available by that. And it's not hard for me to come
- 35:04up with. As an example, one of the things
- 35:06that we've looked at is using flocking and swarming
- 35:08algorithms. So the way flocks of birds can
- 35:11bifurcate and swarms tend to get bigger and bigger, right?
- 35:14So there are clustering algorithms that behave like
- 35:16that. Those two different behaviors,
- 35:18if you start applying them to botnet attacks and if you start
- 35:21imagining that the botnet attack, the botnet swarm or the
- 35:25botnet flock, depending on what you want to call it, we'll get
- 35:28smarter about how it's failing and therefore better at
- 35:31attacking you. That is a horribly it.
- 35:35It's just a horrific thing to imagine.
- 35:37And I know how to write code that would do that.
- 35:39And who am I right? So if, if, if we can do that
- 35:42sort of thing at an unimaginable scale in the very near future
- 35:48and we're just getting really good at fixing yesterday's
- 35:50malware, we are in for a world of hurt in that world that
- 35:54you're talking about. The good news is there are some
- 35:56fine folks thinking about that. The the even better news is, to
- 36:00your point, we need to incent them to think harder and faster
- 36:02about that. Anastacio made a very
- 36:05interesting point. She said that investment is far
- 36:10higher for technology capability, AI capability than
- 36:15it is for I'll use the term compliance or cybersecurity.
- 36:20And you know, look, human nature simply tells us that innovation
- 36:28is fun and compliance is not. And innovation makes money and
- 36:32compliance costs US money. And yes, there's a cost to
- 36:35society, but it's not to me if I'm a software company directly,
- 36:41unless of course there's there's a data breach.
- 36:44O Part of this is definitely trying to go against the tide of
- 36:49human nature from that standpoint.
- 36:51This is the argument for, you know, smoke detectors and fire
- 36:55alarms and life preservers and all of that is, you know, I
- 36:59have, I have life preservers on my boat because I'm required to
- 37:02have them on my boat. And if I have a boat, I want the
- 37:05best life preserver that's going to preserve my life, right?
- 37:08I want it, but most people will not think that way.
- 37:11And if I'm going to put a ferry out there with, you know, 2000
- 37:14of these things, maybe I start thinking about how much they
- 37:16cost and and you know, what's the most cost effective
- 37:19deployment of, you know, mandatorily require life saving
- 37:23equipment rather than how do I save David Bray's life because
- 37:26we need more people like. Him well, and so, but maybe if I
- 37:29can also jump on that, that question that you asked Michael,
- 37:32I think, you know, we've been talking about cybersecurity, but
- 37:35you can actually create a whole lot of damage without ever
- 37:37breaching a system. And what we're seeing right now
- 37:41and This is why trust versus deception is so important, you
- 37:45know, so using 10 minutes of compute time from worm GPT,
- 37:48which is the dark side cousin of ChatGPT and some plug insurance
- 37:52using data stolen from healthcare data breaches.
- 37:55And many of us may have been affected by a data breach in the
- 37:57last year. That data can then train and
- 38:00create 1,000,000 realistic looking medical records complete
- 38:03with chest X-ray, physician's notes, doctor's notes, or a
- 38:07claim at $250, which is below the fraud detection threshold
- 38:11for several of the services here in the United States because
- 38:14it's more expensive to adjudicate it than to pay that
- 38:16out. And that's 1,000,000 records for
- 38:18upper respiratory infection. Now what the points to, and this
- 38:21is what Anthony's talking about is again, the solution is we've
- 38:25got to incentivize those people who are doing innovation to find
- 38:29innovative solutions to either adjudicate faster, adjudicate
- 38:32with more veracity, whatever it might be at scale.
- 38:35Given that AI is basically doing the same thing that
- 38:38unfortunately, you know, remember when the Xerox copier
- 38:40first came out and we had to upgrade our dollar bills because
- 38:42some people were doing called our copies of dollar bills.
- 38:44So we upgraded that as well. The challenges and the unique
- 38:47challenge with AI is just what's called generative adversarial
- 38:51networks or Gans. And so the moment you create a
- 38:54solution that's really good at filtering, this is valid, this
- 38:57is not valid, then a bad actor will then use that GAN to say,
- 39:00use that GAN to get good at fooling it.
- 39:03And so it's going to be predator prey relationships.
- 39:05But I do think this points to at the speed at which this is
- 39:07happening, we have to figure out ways to incentivize either
- 39:11because we're shining a spotlight on it because
- 39:13companies are realizing that's a whole massive amount of money in
- 39:16terms of payments that they shouldn't be made for fraud
- 39:18purposes or whatever. But there has to be an incentive
- 39:20for market based solutions to this because central planning
- 39:23will not get us out of these solutions.
- 39:25But you do have to figure out a way to shine the spotlight on it
- 39:28because the speed at which this is moving, it's almost like what
- 39:30we're seeing in Ukraine where every every six to nine months
- 39:35there is a generational leap in terms of how they're using drone
- 39:38on drone for conflicts. The only difference is this is
- 39:41going to be behind the scenes in terms of AI spoofing.
- 39:43How do you know if that image is real, that video is real, or
- 39:46that person you're talking to is real?
- 39:48This is a perfect time to subscribe to the CXO Talk
- 39:53newsletter. Do that now so we can notify you
- 39:57of upcoming shows. OK, so questions.
- 40:00So who Wei Wang on LinkedIn? Says that data collection right
- 40:04now is being affected not only by linguistic transliteration or
- 40:09translation nuances, but we're also seeing complex data
- 40:13patterns when the technology, languages, and platforms are
- 40:17contributing to changes in how the data is stored.
- 40:21What do you recommend in navigating and preparing for
- 40:25these increased complexities against the rate of how frequent
- 40:28we are collecting data like this?
- 40:32And before you answer, I have a request to the audience which is
- 40:37you got a dumb moderator so just keep your questions short.
- 40:41I've got to jump on that one. I think Anthony paid that
- 40:43individual to ask that. Question.
- 40:44I did not. I did not but I say it's a
- 40:47brilliant question so I'm going to paraphrase the question.
- 40:50There's lots of languages out there and there's lots of people
- 40:52speaking in those languages and we're trying to teach AI to suck
- 40:56in data spoken in those languages or written in those
- 40:59languages. And a lot of times the the
- 41:02systems that are showing you that language, it's not the way
- 41:06it was originally written. So I'm sure most of us have
- 41:09people in our social network where they write something in
- 41:12their language and then you see what they allegedly said.
- 41:15And then there's a translated button on the bottom.
- 41:17So it's been translated or I'll say transformed.
- 41:20So there's a lot of data out there that's gone through
- 41:22linguistic transformation either intentionally or
- 41:25unintentionally, either at the time of writing or subsequent to
- 41:28the time of writing. And that introduces all kinds of
- 41:31nuance into the language that we consume by our AI systems.
- 41:35That's got to have an impact. And it turns out that it does.
- 41:37And there are two different arguments to, to and there,
- 41:40there's no winning argument here yet.
- 41:42So the Turing argument is that if I get enough and most people
- 41:46you will find addressing this problem, we'll we'll use this
- 41:49argument. As long as I get enough examples
- 41:51of that language, I'll be fine. Just give me millions and
- 41:54billions and billions of articulations and no matter how
- 41:56complicated it is, I will regress around it and I will
- 41:59figure it out. The contrary argument to that is
- 42:03that when we speak a language, we change it and we with the
- 42:06nuance gets introduced into that language.
- 42:08Things like sarcasm and neologism and borrowing words
- 42:12from foreign languages, all of these things confound to our
- 42:14ability to understand. And so right now we're speaking
- 42:17in a certain way that is quasi academic, intelligent.
- 42:22We're trying to speak in complete sentences.
- 42:24We know that there's going to be a transcript later.
- 42:26So we don't want to look like idiots, right?
- 42:27So we're we're trying to say things in a certain way.
- 42:30If we talk maybe over an adult beverage, it might come out a
- 42:33little bit different. So how do when, when you
- 42:36introduce language on top of that, how does it perturb your
- 42:40ability to understand? Is that person upset?
- 42:42Is that person lying? Is that person, and one of these
- 42:45people is an imposter? Which one is it?
- 42:47These are all right. Now you can have an honorary PhD
- 42:50in computational linguistics by tackling any one of those
- 42:52questions really well because I'm sorry to tell you, but the
- 42:56technology hasn't gotten there yet.
- 42:58It's getting better. There are things as Huey
- 43:01mentioned in her question, there are things that you can do to
- 43:04recognize commonly occurring graphemes, to recognize things
- 43:07like David mentioned misspellings before.
- 43:09Some misspelling can be intentional, some misspellings
- 43:12can be accidental. Every time I write something to
- 43:15Anastasia my my autocorrect changes the spelling of her name
- 43:19because there's another person I know that spells the name
- 43:21differently, and if I don't catch it, I send it to her
- 43:23spelled wrong. She knows this now, so hopefully
- 43:25she's not offended. These are things that are
- 43:28happening all the time and it's going to get worse.
- 43:32So what we have to get better at is not just accepting that
- 43:34computational linguistics is give me the dictionary, because
- 43:38there are low context languages where there aren't.
- 43:40There isn't one really good source, or there are languages
- 43:44where the only really good source is the religious text,
- 43:47right? And that's not how people speak.
- 43:48So now you've just consumed the Bible and you're trying to speak
- 43:51English. We don't speak like that
- 43:52anymore. So there's that problem, the low
- 43:54context problem. There's the multilingual
- 43:56disambiguation problem where Peter, people are writing in
- 43:58more than one language. And then there's the language
- 44:00transformation problem where people are changing what was
- 44:03written and you're reading what has been already transformed.
- 44:07All of those are Wild West right now and and you can you can make
- 44:11a lot of impact or working in any one of those fields right
- 44:14now. Greg Walters on LinkedIn says
- 44:17we've lost the EUI think, quote UN quote, alien is the best way
- 44:21to view AI in that vein. Other than that, we have no way.
- 44:26Shouldn't we look at AI through a brand new lens?
- 44:29A lens with no previous anchors or history or KPIs or best
- 44:32practices? AI does not fit only in a
- 44:36vertical or horizontal. It is everything, everywhere,
- 44:39all at once. There are no experts.
- 44:41What is a new perspective? AI as mirror.
- 44:45When 2001 happened, I was responding to 911 in the anthrax
- 44:49events. There were conspiracy theories
- 44:51that have were around, but none of them were able to take root.
- 44:54I mean they were saying like we've done it to ourselves,
- 44:56false flag, US have done it but they didn't take root.
- 44:59Fast forward to 2009, I'm on the ground in Afghanistan and sadly
- 45:02there's an event in western Afghanistan where the Taliban
- 45:06actually took a photo, a valid photo of AUS fighter jet flying
- 45:09overhead and then at the same time briefly took a photo of a
- 45:12propane tank detonating and unfortunately killing innocent
- 45:14Afghans. Both were true photos, but they
- 45:18were clearly out of context because then they went on social
- 45:20media and claimed US air strike kills innocent Afghans.
- 45:24The Department of Defense that we're investigating, and it took
- 45:26more than 4 1/2 weeks to figure out what happened during that
- 45:28time. Of course, the news media was
- 45:30blaming the US and our own ambassador apologized.
- 45:33That took 4 1/2 weeks to actually put that
- 45:36characterization of what really happened versus the deceptive
- 45:38narrative to bed. And then as you know, Michael,
- 45:41back in 2017 when I was at the Federal Communications
- 45:44Commission, there was an event in which we were getting 6007
- 45:48thousand 8000 comments a minute at 4:00 AM, five AM, 6:00 AM.
- 45:52And we were told by our lawyers we couldn't test for bots.
- 45:54Speaking of CAPTCHA, Anthony, we couldn't do invisible means
- 45:57because that would be seen as surveillance.
- 45:59And we couldn't block what was perceived to spam 100 comments
- 46:02from the same IP address per minute.
- 46:04And it took more than 4 1/2 years at that point in time for
- 46:08eventually New York Attorney General to adjudicate and say of
- 46:11the 23 million comments we got, 18,000,000 were politically
- 46:14manufactured, 9 million from one side of the aisle, 9 million
- 46:17from the other aisle. So over the last two decades, we
- 46:20are getting longer and longer tails for the more valid, more
- 46:25authentic, accurate characterization scenarios to
- 46:29get themselves out. We went from they didn't take
- 46:31root in 2001. Now it took 4 1/2 weeks in 2009,
- 46:352017 it took 4 1/2 years. We are unintentionally through
- 46:38no one technology. I would submit Internet,
- 46:41smartphone, generative AI. We are laying the seeds in which
- 46:45it is. You know, it used to be the
- 46:46adage was, you know, a lie can go halfway around the world
- 46:49before truth gets on its sneakers.
- 46:51I would say at this point in time, I'd like and get to the
- 46:53other end of the solar system where the truth gets on its
- 46:56sneakers. And so This is why I would
- 46:57submit the solutions have to be technology neutral and also have
- 47:02to account for humans doing things regardless of anyone
- 47:05machine. I still would say at the end of
- 47:07the day, it's going to be sector specific because the severity of
- 47:10doing this, if you're recommending something to buy on
- 47:13a website is much different than making a decision about your
- 47:16health. It's much different than making
- 47:17a decision on a battlefield. Anastasio there are two
- 47:21questions and I'll combine them and direct them to you.
- 47:24Elizabeth Shaw says that there are strong agendas that power
- 47:31deceptive and malicious use of AI, such as greed, power, ETC.
- 47:38How do you encourage the ethical use of AI as opposed to the
- 47:43deceptive use? And then Arsalan Khan says that
- 47:48there are many underlying issues such as security, data bias,
- 47:53culture, jobs, and and so forth that AI can help or make worse.
- 47:59He's asking fundamentally the same thing.
- 48:02How do we incentivize organizations to take a, an
- 48:09ethical as opposed to a deceptive approach?
- 48:14And I'm, I'm paraphrasing these two questions, but that's
- 48:18fundamentally we spoke about that earlier.
- 48:20The incentive is incentivizing. If you incentivize prospective
- 48:25customer to be more aware and to be more educated on what is out
- 48:30there, because sometimes the issue is not the technology, but
- 48:32the business model and the why behind a certain construct in,
- 48:37for example, how social network is configured.
- 48:40And I think that the reduction of confusion and demystifying of
- 48:44AI is a very noble cause. So in my eyes, there are three
- 48:48fundamental buckets of risks in AI and one bucket is everything
- 48:54which has to do with design. For example, we are talking
- 48:57LLMS, and I mean, for my taste, LLM will always hallucinate.
- 49:03You might reduce certain things, but you know, it will still stay
- 49:07just because I'm not going to dive now into the theory of
- 49:11computation and all of the above, but it will hallucinate,
- 49:15period, because of the design. We are talking about AIS as if
- 49:20AIS are quite new. But actually the technology, the
- 49:23roots of technology is in the 40s, fifties and 60s of the last
- 49:28century. And we must go into a new wave
- 49:31of architectural design to improve.
- 49:34So this either by the construct, by the system, an issue or
- 49:41because there is a human mistake or obviously this is always
- 49:44possible, then there is a malicious intent.
- 49:48So we talked about cybersecurity.
- 49:50Criminals will apply AI for the dark purposes and some companies
- 49:55might want to seduce their customers into certain thinking.
- 49:59So the more educated the customer is, the better.
- 50:03Sometimes we are talking about scalable problems and sometimes
- 50:07we are talking about basic stupidity on behalf of a
- 50:11customer. And when I work with companies
- 50:14and I review their AI portfolio, sometimes I'm like, why are you
- 50:18spending this money on that? And the issue is that the
- 50:22customer believed that the sales personnel of a certain vendor
- 50:26and the vendor consists of 85% of sales and only 15% of
- 50:32engineers. Obviously, this vendor cannot
- 50:35execute. I'm not now going into ethical
- 50:39and trustworthy AI, just the basic configuration of whatever
- 50:43systems. So the more knowledge on behalf
- 50:45of a customer, the better is the outcome.
- 50:48And last but not least, and this will be very specific to an
- 50:52industry or business function. This is so-called human in the
- 50:55loop. This is the third bucket of
- 50:57so-called risk. Now define how much human and in
- 51:01what kind of loop when are we going to introduce the
- 51:05supervision and the human feedback?
- 51:08Once again, this is a balancing act and and this is a play in
- 51:12between of whoever is building and introducing the system and
- 51:15whoever is actually the customer.
- 51:18So it might be an answer which is not very simple.
- 51:23But to my eyes, this education and asking good questions, being
- 51:28capable to review the vendor portfolio and read it to write
- 51:32down what type of problems are we solving, do we need this and
- 51:38at what cost. AI might appear new, but our eye
- 51:43is still our eye. If I have this kind of, you
- 51:46know, rule of the term that 25% of my revenue line will be spent
- 51:51on compute and 15% of my revenue line is being spent on cleansing
- 51:57data. If I am great.
- 51:59Yeah, then you have your economics and you need to keep
- 52:04it in mind before deciding whether you are going into a
- 52:07certain AI implementation or not.
- 52:09What will it cost you? Do you really need to automate
- 52:13here to introduce an AI agent here?
- 52:15Or maybe a human will do just fine.
- 52:18Let me just ask each of you very briefly for a final thought.
- 52:23David, final thoughts on this topic.
- 52:26This is not a tech specific issue.
- 52:28This is a society and company level issue.
- 52:32And it's going to be solutions that look not at the tech
- 52:35specifically, but look broadly. And again, recognizing we're
- 52:38wrapping up data, data, data. We didn't talk at all about data
- 52:41governance. We didn't talk about data
- 52:42cleaning. I know, and Asajj just briefly
- 52:43touched on it. You know, one way where you
- 52:45could do most everything right in AI and completely fail is if
- 52:48your data is bad. And so I would say at the end of
- 52:50the day, the, the, the, the intent of pursuing more
- 52:56trustworthy and less deceptive AI begins first and foremost
- 53:00with putting yourself in the shoes of your different
- 53:02stakeholders and making sure whatever you do, tech or not,
- 53:06data or not, you are thinking about them and how you deploy
- 53:09things. Anthony, final thoughts.
- 53:12Couple of things. One, beware shiny objects,
- 53:15right? There's going to be more and
- 53:16more shiny objects, right? Whatever they're called right
- 53:18now, it's LLMS. That's great.
- 53:21What problem are you solving? Always step back and ask, what
- 53:24do I have to believe? What's the problem I'm solving?
- 53:27Don't get distracted by the shiny object because there's
- 53:29another one coming right behind it.
- 53:31The second thing is, and that's to David's point about cleaning
- 53:35the data. The second point is regulation
- 53:37is going to happen with you or without you.
- 53:39So get involved. Make sure that the regulators
- 53:41understand the unintended impact of some of the things that
- 53:45they're considering. We just, we just wait for these
- 53:47regulations to come out and then it's kind of too late.
- 53:50And then the third thing I would say is, and I often say this, be
- 53:54humble. You can't solve this alone.
- 53:57Get help. There's lots and lots of folks
- 53:59out there that have lots of expertise.
- 54:02We should make new mistakes, right?
- 54:04And we do that by involving other people in our decision
- 54:07process and and being humble enough to realize that I don't
- 54:10care how smart your smartest people are, you should bring in
- 54:13some people that disagree with you.
- 54:15And Anastasia, it looks like you're going to get the last
- 54:18word here. Final thoughts.
- 54:19We might be in technology and AI, but we are all in people
- 54:23business ultimately, and it's really about human leadership
- 54:27and thinking and humility. And I don't encourage people to
- 54:32wait for some decision from let's say Alphabet or from some
- 54:36new regulation out of Washington DC or Brussel.
- 54:40I would really encourage local communities to look into the
- 54:44local ecosystems and see what kind of colleges, universities
- 54:49are out there interested in AI, offering some courses, what
- 54:53schools are interested? What could you do for those
- 54:56kids? What might be the local startups
- 54:59to do, let's say startup breakfast?
- 55:02So they they explain what they're doing and how they're
- 55:05hiring into the company. Maybe some incumbent companies
- 55:09which are adopting one or another type of technology.
- 55:14So I would really encourage this bottom up movement rather than
- 55:18waiting for some big guy somewhere to decide for you
- 55:22because this is how you learn and this is how you create a
- 55:25dialogue and ultimately the progress will happen through the
- 55:28dialogue. We are dealing, as all three
- 55:32have said, with human rather than specifically technology
- 55:38issues. And so the question then becomes
- 55:43how do we harness and focus this human energy for the ethical use
- 55:50of AI? And I think with many other
- 55:53things, it's going to require carrot and a stick.
- 55:56There's no simple solution here. And as complex as the technology
- 56:01is, the human aspects are always harder.
- 56:04So, Carrot and Michael, are you building a snowman is what
- 56:06you're talking about? You know, I'd love to build a
- 56:09snowman. It's actually snowing right now
- 56:11here in Boston. Wait, are you saying do you want
- 56:14to build a snowman? Sorry I'm.
- 56:15Sorry. Sorry, Michael, back to you.
- 56:18Anyway, a huge thank you to our guest, Doctor Anastasia
- 56:24Lauterbach, Doctor David Bray, and Doctor Anthony Scrofignano.
- 56:29Thank you all so much for being here.
- 56:32I'm very grateful to you all and a huge thank you to everybody
- 56:38who watched today and especially you folks who asked such awesome
- 56:41questions. I always say this, you guys in
- 56:44the audience are amazing. Before you go, please subscribe
- 56:50to our newsletter and subscribe to our YouTube channel.
- 56:54Check out cxotalk.com. We have great shows coming up
- 56:59and our next show will be back at the usual time of 1:00
- 57:04Eastern. Thanks so much everybody and I
- 57:05hope you have a great day.