Latest / CXOTalk / Top AI Ethicists Reveal RISKS of AI Failure | CXOTalk #888
Transcript
- 0:00AI systems can go off the rails, run amok, and hallucinate.
- 0:05So on CXO Talk 888, Lord Tim Clement Jones and Doctor David A
- 0:13Bray explain how to anticipate, manage, and prevent these
- 0:18failures before they escalate into enterprise level crises.
- 0:24David, tell us about your work. Currently, I am chair of the
- 0:28accelerator at the Stimson Center, which simply means we
- 0:31are striving to figure out how we do policy and program in
- 0:34parallel because the world is changing so fast.
- 0:37Other than that, I have served in different roles, ranging from
- 0:40countering bioterrorism to Seattle, the FCC senior National
- 0:43Intelligence Service executive. Tim, please tell us about your
- 0:47work. I'm a member of the House of
- 0:50Lords. I speak there on science,
- 0:52innovation and technology, particularly AI, and I Co chair
- 0:57our Cross party group on AI in Parliament and I consult with a
- 1:03number of organizations on AI policy and regulation.
- 1:08When we talk about AI, generative AI in particular,
- 1:14running amok, what exactly does that mean?
- 1:17One of the interesting things about generative AI is that it
- 1:20actually generates its own content.
- 1:22Before then, with different other flavors of AI, we had
- 1:25things that were rule based systems, we had natural language
- 1:28processing. But generative AI is actually
- 1:31able to generate its own content.
- 1:32And if you look at it, whether it's an image generator or a
- 1:34large language model, it's sort of doing prediction and large
- 1:38language models. It's next word prediction and
- 1:39images. It's trying to respond to the
- 1:41prompt you gave it and and and as a result, usually they have
- 1:46these things called random seeds.
- 1:48They also have temperature controls that introduce a little
- 1:50bit of randomness. Now, even if you turn down the
- 1:53temperature control, turn down the seeds and say I don't want
- 1:56any randomness nowadays with these very large models because
- 2:00you're doing compute in parallel, there's still some non
- 2:04determinism because these race conditions that send things to
- 2:06different processors, you could do the same prompt twice and it
- 2:11may not actually result in the same outcome.
- 2:13Plus a lot of these models have what's called observer effects,
- 2:15which if they observe a prompt has been asked, when you ask the
- 2:18same prompt a second time, it's already been altered because it
- 2:21observes that same prompt. So we're dealing with things
- 2:23that are not deterministic, they're not rule based.
- 2:26And so as a result, when enterprises roll this out, they
- 2:28may think it's going to do a certain thing.
- 2:30But as you mentioned, there's hallucinations.
- 2:33We may try to get hallucinations down.
- 2:34Let's say we even get it down to single digits or 1%.
- 2:37You'll still never be able to reliably predict when it may
- 2:40actually do something that's completely not factual.
- 2:43Again, it's doing sort of multi dimensional pattern matching,
- 2:46but not necessarily factual. Tim, you're in the House of
- 2:49Lords and so you're looking at this from a policy and
- 2:54governance perspective. And so from your standpoint,
- 2:58what is the impact of this run amok or go off the rails or
- 3:04hallucination aspect of generative AI?
- 3:07The impact of a hallucination, when it's fake news, or even
- 3:11when it's misinformation or other sorts of misbehaviour, if
- 3:17you like, by the generative AI model can have extraordinary
- 3:22consequences. You know, for instance, if
- 3:25somebody. And of course, humans, quite
- 3:28often linked to a misbehaving AI, are what really matters.
- 3:33And now we see much more retail use of large language model.
- 3:38This isn't just about corporate adoption.
- 3:43This isn't just about a large language model adopted by a big
- 3:47business going amok. This is individuals using AI
- 3:52models for their own purposes, which are then in a sense
- 3:57creating this problem. So, you know, it could be
- 4:00elections, it could be people sending out misinformation,
- 4:06causing public, public unrest, all kinds of areas or indeed
- 4:11discriminatory decisions by an AI system.
- 4:15So it's the very prevalence of AI as well as well as the
- 4:20technical problems with it which caused the problem as well.
- 4:24You know, look, people make mistakes, AI makes mistakes, so
- 4:27what's the big deal? It's when you don't double check
- 4:30the machine. So for example, we saw for
- 4:33example, this was early on in the days of ChatGPT 3.5.
- 4:37There were apparently some, there was one or more lawyers
- 4:40that basically used it to generate some references for a
- 4:43court case. They looked realistic because
- 4:45again, what this is doing is it's doing sort of
- 4:47multidimensional pattern matching.
- 4:49They didn't go and check to see if those court cases actually
- 4:51existed. And then when they got before
- 4:53the judge, lo and behold, the very court cases that they
- 4:56supposedly were citing did not actually exist.
- 4:58And so you need to recognize Jennifer VI will produce things
- 5:00that look very realistic but could be completely fake.
- 5:04But the second example I'll give is there was a competition back
- 5:07in November where they explicitly told the machine
- 5:09under no premises, should you ever transfer funds.
- 5:13And this was made publicly available.
- 5:15Anybody could ask at a prompt, but you had to pay a little bit
- 5:17of money. So it was sort of like a self
- 5:19funded contest. And and by attempt #490 so by
- 5:22prompt #490 even though the machine had been told do not
- 5:26transfer the funds, guess what the machine did?
- 5:28It transferred the funds. And so that's where, again, you
- 5:31have to recognize these are not rule based engines.
- 5:33And as a result, what they give you may be unreliable.
- 5:36Or if you tell it not to do something, you may find
- 5:39ultimately it does. Yes, and we have our own
- 5:42examples in the UK of publicly used machine learning tools in
- 5:50government, you know, having deprived 200,000 people of their
- 5:53benefit on an illegal basis. And of course, our most
- 5:59celebrated case in the worst possible way is all our sub
- 6:03postmasters who were prosecuted by the post Office who were
- 6:09using it was wasn't quite AI in those days.
- 6:12In fact, it was an expert system.
- 6:15And how much worse it would be if it wasn't an expert system, I
- 6:19tell myself, which caused prosecutions people to commit
- 6:24suicide and they're still awaiting compensation from our
- 6:28government. And you know, for the for the
- 6:31wrongs that were done them for being jailed unlawfully.
- 6:36So, you know, this trusting faith that people have in
- 6:41computer based decisions, if you like, putting it very broadly,
- 6:45is something that carries through into AI, particularly
- 6:49when AI has this ability to converse with you and seem
- 6:52rather human. You know, let's face it, they
- 6:55passed the Turing test nowadays, you know, almost certainly, I
- 7:00think that's probably well out of date.
- 7:02And so it has this kind of human look about it which generates
- 7:06public trust, which makes it extremely dangerous in that says
- 7:10in that case. So this trust aspect, this kind
- 7:15of false confidence that leads us to believe that we are
- 7:20receiving quote UN quote, the truth.
- 7:24Is that really one of the the core most dangerous aspects,
- 7:30would you say? If you define trust as the
- 7:32willingness to be vulnerable to an actions of an actor you
- 7:35cannot control, we perceive competence, we perceive maybe
- 7:40integrity, we perceive benevolence in these models.
- 7:43And as a result we are trusting them when in fact maybe we
- 7:45shouldn't. And that's partly because of
- 7:47marketing, that's partly because of what we've been told.
- 7:50But as he said, they appear human and the more they appear
- 7:53human in their conversations with us and interactions with
- 7:55us, we don't step back and have the skepticism.
- 7:59But even before this, there was actually in the 80s where they
- 8:01actually played like the the bargaining game.
- 8:03And the machine was pre programmed to make offers.
- 8:05And so it was never going to deviate from this.
- 8:07But people still penalize the machine for doing unfair offers
- 8:11even though it was pre programmed.
- 8:12And so this is actually more of a reflection of human nature,
- 8:15which is we try to we we try to treat things as if it was
- 8:18ourselves when we really need to step back.
- 8:20And I often tell people AI is not artificial intelligence,
- 8:23it's alien interactions. Do not assume at any point in
- 8:26time is it thinking anything like us.
- 8:28And as a result, when you are looking at different AI methods,
- 8:30including generative AI, know their strengths and know their
- 8:33weaknesses and brief your board, brief your leadership, brief
- 8:36your government, because those different strengths and
- 8:39weaknesses may be your comeuppance if you use the AI in
- 8:43the wrong context. David's so right about that,
- 8:45Michael. And what is so interesting is we
- 8:48defer now to these large language models.
- 8:51Look how polite we are. I mean, David may kick the
- 8:54computer when it gives the wrong answer, but an awful lot of
- 8:58other people say, oh, thank you so much.
- 9:00That's a great answer. And, you know, this polite
- 9:04interaction with a machine is, you know, quite extraordinary.
- 9:08But this is what we are now programming ourselves to do,
- 9:11basically. So again, you know, we are are
- 9:14building up this kind of false platform really and we should be
- 9:18much more skeptical. It does take on the character of
- 9:23a person because it presents the illusion of consciousness, of
- 9:31sentience. Yes, that's I think the biggest
- 9:33thing I want to tell people. At no point in time is it
- 9:35conscious or sentience. This is just fancy
- 9:37multidimensional mathematics. The trouble is you do have some
- 9:40people that are, for whatever reason, either talking about it
- 9:44because it gets clicks or talking about it because it's
- 9:47feeling they they themselves feel anxiousness.
- 9:49But you need to recognize it in a day that the current flavors
- 9:52of AI that we have are really just really advanced
- 9:55multidimensional mathematics. And so when you deploy it, I
- 9:59think this does get to the interesting things.
- 10:00You need to be ready for it to both technically go off the
- 10:03rails because you may have used AI in the wrong context.
- 10:06You need to be ready for the fact that maybe the AI is doing
- 10:09exactly what it it, what the algorithm should be doing, but
- 10:11you trained it poorly on the data, that the data was non
- 10:14representative or there were things that were missing from
- 10:16the data. And as a result, it's going to
- 10:18reach conclusions that do not match what we want it to match.
- 10:21Or finally, those two things could be done right.
- 10:24But what you missed was the human interactions as a result.
- 10:27And so you begin to see exactly as Lord Tim said, people who may
- 10:30be going to an AI model asking it for news and and going down a
- 10:36a increasingly mixture of some real News Plus hallucinated news
- 10:42or news that's being skewed to the person's attention or
- 10:44interest. And as a result, they believe
- 10:47this to be truth without going to other sources to do this sort
- 10:49of skeptical trust but verify. Absolutely.
- 10:52And it's very seductive. And then you get this over
- 10:55dependence that being created. And that's one of the things
- 10:58that I worry it's the deskilling aspects in the future as well.
- 11:03Preeti Narayan on LinkedIn says, do you think we'll reach a point
- 11:08where AI risk officers are as common as chief information
- 11:14security officers? And if so, should they have a
- 11:18Direct Line to the CEO or even the board?
- 11:22That's a that's kind of a pregnant question.
- 11:25There's a lot there. One of the things that I'm
- 11:27increasingly hearing from CIOs and Cisos, chief information
- 11:30security officers both in the private sector and the public
- 11:32sector, is we all know that in cybersecurity, complexity is
- 11:37actually what erodes strong cybersecurity and AI.
- 11:41Guess what? It's one of the most complex
- 11:42things we're bringing into enterprises, into organizations.
- 11:45And so yes, in some respects, there is going to be a
- 11:47convergence between the things you need to do for cybersecurity
- 11:50as well As for AI risk. And one of them that I would say
- 11:53that we need to do more of already is move away from what
- 11:57was called forensic space, sort of like looking for very
- 11:59specific patterns of something. And instead we need to do what's
- 12:03called patterns of life. What does normalcy look like in
- 12:05your organization? What does normalcy look like
- 12:07with this system? And then if you see something
- 12:09odd, that could be a hardware, that could be a software
- 12:12problem, it could be a cybersecurity issue, or
- 12:14increasingly if it's something that has AI behind it, it could
- 12:17be the AI doing something it shouldn't do.
- 12:19And so I actually think there's going to be a convergence
- 12:21between patterns of life based approaches for cybersecurity as
- 12:25well As for what's needed to be done to make sure AI does not go
- 12:28off the rails for companies and for governments.
- 12:31Absolutely agree with that. Because I don't think in the
- 12:34future we're going to be able to separate out that risk role
- 12:37from, if you like, the cybersecurity role.
- 12:40After all, we know that AI can be a cybersecurity risk in
- 12:44itself and therefore the risks of AI start moving into that
- 12:48area. And I, but I do agree with our
- 12:51questioner about the, the, the importance of, of assessing the
- 12:55risk. It needs to be mainstreamed
- 12:58within a corporate structure and it needs to be very, very close
- 13:02to the CEO, the audit and risk committee.
- 13:06It's it should be integral to the way that CEOs that boards
- 13:12adopt artificial intelligence. And if they don't have that
- 13:16approach to risk, then they are very vulnerable when they start
- 13:19adopting it. So, you know, this is very much
- 13:22part of the structure that I think you and I would propose,
- 13:25David, for corporates to adopt. And you know, my worry is that,
- 13:30you know, it's it's adopted to piece meal without that
- 13:33overarching assessment at the very beginning of what kind of
- 13:38structures you need to keep the the corporate safe to make sure
- 13:42that its purpose is being delivered by using these new
- 13:47tools. It's just a sort of random sort
- 13:51of idea that is going to improve productivity.
- 13:54And we're just going to install a few tools here and there.
- 13:57I think we have to be much more holistic about the way that it's
- 14:00delivered. Is there a greater danger from
- 14:04AI risks than traditional security risks?
- 14:09Security security is now a board level topic because of the the
- 14:17potential reputational, financial, legal damage that can
- 14:20ensue. Where does AI stand from a risk
- 14:25standpoint and the prominence of the the risks?
- 14:29My assessment of the current flavor of AI, the current
- 14:33generation we face, it's a lot like insider threat, just like
- 14:36you need to be worried about insider threat where you don't
- 14:39know is that human might do something that's maybe not
- 14:42malicious, but they did something wrong.
- 14:43They just sent 250,000 names publicly that should not have
- 14:47been named publicly. That's just an example.
- 14:49Or maybe they are malicious. Maybe they're actually
- 14:51disgruntled employee or they're actually getting paid to do
- 14:54something, or maybe they're getting coerced.
- 14:56And so you need to treat almost when you bring AI into
- 14:59enterprise, it actually needs to be a way to upgrade.
- 15:02And you'd have the conversation with your board about we're
- 15:04going to treat AI just like we need to upscale our own insider
- 15:07threat issues. Because again, this AI, you
- 15:10know, if haven't, you know, I would not recommend using
- 15:13DeepSeek. But if you're using DeepSeek,
- 15:14you should be aware that it's also sending metadata and
- 15:17information overseas. And are you comfortable with
- 15:19that? So I would say it, it, it is, it
- 15:22is a risk. It is a significant risk at the
- 15:24speed and scale it can do. However, I think we have tools
- 15:28very similar to how we actually approach insider threats within
- 15:31companies. The same thing can actually be
- 15:33done for AI too. And I think the shrewdness of
- 15:35the question, Michael, is that I don't think many boards are
- 15:40there yet. I think that Dave is absolutely
- 15:43right about what should happen, but I don't think that is
- 15:46currently happening. There's a great focus on or what
- 15:49the opportunities are, but it's not really looking about the
- 15:53risk, how it fits within the risk appetite because I don't
- 15:56think the risk is necessarily appreciated or the level of risk
- 16:02is not yet appreciated. So I think boards, I mean, my
- 16:06worry about boards universally is they don't necessarily have
- 16:11the skills yet to really assess at that level the kinds of AI
- 16:16that they should be adopting, how AI will fit with their
- 16:19values and their objectives and so on.
- 16:22But you know, this is these are quite big issues across the
- 16:26board. Let's jump to Twitter and take
- 16:29some questions there. I've this audience is so smart,
- 16:33so intelligent. I love these questions.
- 16:36Here's one from our friend Anthony Scruffignano, who of
- 16:40course has been a guest many times on CXO Talk.
- 16:43And he asks this as a new generation is emerging that has
- 16:48learned to let the machine do the work, how do we get the
- 16:52workforce to use their excess capacity to think for themselves
- 16:57and focus on new unmet challenges?
- 17:02There's two buckets. First is as they are using the
- 17:05machine and working with the machine as Lord Tim said, and I
- 17:08agree enthusiastically, still be skeptical.
- 17:11You know, just cause the machine has given you an output, you
- 17:14almost need to sort of double check to work.
- 17:16And it doesn't mean you have to check everything, but you need
- 17:18to, you need to gain some expertise.
- 17:20The second thing is, is we now need to figure out how do we
- 17:25actually sort of do a combination of of not just one
- 17:29individual to an AI, but teaming across multiple people and
- 17:33machines because it may very well be.
- 17:35The future is especially for an entry level employee, you may
- 17:39actually have either as a deputy boss or even as a boss, your
- 17:43initial boss might actually be giving you tasks through an AI
- 17:46system because what you're initially doing is the more, you
- 17:49know, sort of learning entry level work and you have to be
- 17:52comfortable with that. But then what's your recourse if
- 17:54you think the the AI is giving a wrong task or the task is no
- 17:57longer fit for purpose because it might do that.
- 18:00And so I think we actually need to actually begin very early on,
- 18:04maybe as early as middle school, exploring what does it look like
- 18:07to do combination hybrid human AI teaming and then figure out
- 18:12what what should be the actions supposed to be skeptical of
- 18:15what's coming from the machine. So you triangulate it.
- 18:17And then two, what do you do if you think whatever the AI is
- 18:20asking to be done or whatever task is being given to the team
- 18:23doesn't make sense? When does the human raise a hand
- 18:25and say, I don't think that's right, I think something's wrong
- 18:28with the machine? Absolutely agree with David.
- 18:30It starts really further down the pipeline at school in terms
- 18:35of the kind of critical thinking skills that we need.
- 18:38The trouble is we're starting from here and therefore
- 18:42corporates, you know, they need to assess the risk of, if you
- 18:46like, over dependence on these tools where there is no proper
- 18:51human in the loop where there probably should be, there is an
- 18:54over dependence on the tools and there is no real human
- 19:00intervention which actually could get the same kind of
- 19:03result. So the skills of individuals
- 19:07using these tools are going to be absolutely critical.
- 19:12And you know, we no longer need now to amass information.
- 19:16We don't need to in a sense, gather the data and, and or
- 19:21analyse the data in many ways, but drawing conclusions from it
- 19:26and critically assessing it is going to be absolutely, you
- 19:30know, right front and centre for the kind of skills we need in
- 19:33our businesses. And you know, again, it requires
- 19:37an understanding at the top level of our businesses as to
- 19:40what, you know, can usefully be done by humans versus machines.
- 19:45And I don't think that understanding is quite there
- 19:47yet. And I think, you know, David,
- 19:50you do a lot in this area and it's and it's really a question
- 19:53of bringing that right front and centre.
- 19:56I think this idea that we're all going to have a great deal more
- 19:58thinking, I'm sitting there in business and this is going to
- 20:02allow us to dream up new creative thoughts.
- 20:06I think might be slightly optimistic, but I think it is
- 20:09going to be a partnership because you know what they talk
- 20:13about AI in the legal industry now is the sparring partner.
- 20:18And I think that's a lovely analogy, basically where you're
- 20:21testing propositions and you're kicking the machine and saying,
- 20:25well, what about this, what about that?
- 20:27Can we do it better this way or that way?
- 20:29And you, you have somebody that you can argue with.
- 20:32So I quite like that. I think Lord Tim hit it spot on,
- 20:35which is there's, there's actually some articles that say
- 20:38the reward for your AI partner is more work, you know, so it's
- 20:42not necessarily going to free things up.
- 20:44I, I love the sparring partner example partly.
- 20:48Well, so in January of this year, the US administration
- 20:52pushed out some export controls and they, they rushed them out
- 20:54the door before there was a change in leadership.
- 20:57I kind of felt like they were rushed out the door.
- 20:58And so I actually went to an AI model and GPT and I said,
- 21:02pretend you want to get around these export controls and you
- 21:05not only want to get around them, but you want to make
- 21:07money. And sure enough, guess what the
- 21:08AI did? It told me five ways to do that.
- 21:11And so exactly that as we look at different policy things, that
- 21:14we look at different decisions, we should at least red team
- 21:17them. Red teaming coming from the
- 21:18world of what could possibly go wrong or how might they be
- 21:21misused or abused. I think this is increasingly
- 21:23where again, but it's not that you turn to the AI and just turn
- 21:26it all over. It gives you thoughts almost as
- 21:28its sparring partner, as Lord Tim said, and then you as the
- 21:31human adjudicate yes, that makes sense, or no, that couldn't
- 21:34possibly happen. This would be an excellent time,
- 21:37folks. To subscribe to the CXO Talk
- 21:41newsletter, go to cxotalk.com. Subscribe to our newsletter so
- 21:46we'll notify you of discussions like this so you can
- 21:50participate. Do it now.
- 21:52So here's a really interesting question from Arsalan Khan when
- 21:56he's talking about the US, but I think this is broader.
- 21:59When the US, he says, removes AI guardrails, is the US population
- 22:05or the world really ready? Tim, I think this one's tailor
- 22:09made for you. Although the USI mean individual
- 22:13states are different. I mean, obviously we've got some
- 22:16regulation in California and Colorado and some other states.
- 22:19But by and large, we don't think of the states as having federal
- 22:23legislation on this. Although I see that even the
- 22:26president has expressed some frustration about that.
- 22:30And I wonder whether federal legislation isn't coming down
- 22:34the track. But yes, I mean, I think that
- 22:36the that that they should be a desire on, if you like the
- 22:41ordinary U.S. citizen to have standards that are universally
- 22:45applicable to the kind of large language models that we've been
- 22:48talking about. And there are those
- 22:50international standards. And interestingly enough, you
- 22:53know, US institutions like the National Institute for Standards
- 22:57and Technology have developed some very sophisticated
- 23:00standards as well, which could be adopted.
- 23:03So, you know, if the, the if the willingness was there to
- 23:07legislate or regulate, it wouldn't have to be very
- 23:11complicated. And I think it would provide a
- 23:13much greater degree of public trust, which is the trade off.
- 23:17You know, I'm a great believer in having a level of regulation
- 23:21which doesn't impact on innovation because it creates
- 23:25public trust. But you know, I think there's
- 23:28still quite a long way to go. And and, you know, the trouble
- 23:31is that we risk having different regulatory regimes across the
- 23:37world. So my view is that what we need
- 23:39to try and do is agree international standards, which
- 23:42mean that we can adopt these tools universally across the
- 23:46board and then we don't have to sort of think jurisdiction by
- 23:50jurisdiction in all of this. But if we don't have regulation,
- 23:54then of course that puts the owners back on business to make
- 23:58sure that we are kept safe. And we are we can trust the AI
- 24:04tools that are being used. And you know, the the more, the
- 24:07less regulation you have, the more responsibility business has
- 24:11sense. It's not so much that the US has
- 24:13removed guardrails, it is that they have not enacted anything
- 24:16at the federal level that actually is a legislative
- 24:18guardrail. Now that said, just this week on
- 24:22Tuesday, we did see in a new executive order and some
- 24:26additional strategies come out on AI.
- 24:28And if you read them, they actually call for NIST to be
- 24:31involved. And so they're actually is
- 24:33movement this way. I think it is interesting though
- 24:35because it could end up be that the US right now for for various
- 24:38reasons is going to learn more through executive orders and
- 24:41through executive policy versus congressional.
- 24:44The other thing that I also would raise, at least from AUS
- 24:46perspective, and this is just one way to go about doing this,
- 24:50is that we do have existing laws in the healthcare sector,
- 24:53including HIPAA. We do have existing laws in the
- 24:55banking sector including the Bank Secrecy Act.
- 24:57It could be that we actually see Congress, instead of trying to
- 25:00do 1 monolithic AI legislation, which might prove difficult for
- 25:04various reasons, may have the different committees go for an
- 25:07upgrade. What does AI mean for in the
- 25:10healthcare space? What does AI mean in the defense
- 25:13space? What does it mean in the banking
- 25:15space? And so the US may have a
- 25:17slightly different approach than what Europe has done or other
- 25:19countries. But it is the case that, yes, we
- 25:22have not done anything at the federal level legislative yet.
- 25:25But as Lord Tim said, stay tuned.
- 25:27And this is from Kurt Milne, who says will it take a material
- 25:32public failure for C level teams to shift focus from AI
- 25:37opportunity to also focus on risk?
- 25:42Sadly, that is probably true. I wish it wasn't.
- 25:45I wish we didn't have to have, you know, a big data breach for
- 25:49people to get excited about cybersecurity or, you know, some
- 25:53of the other crises that have caused corporates to change
- 25:57their behaviour. We I don't think we've yet had a
- 26:02major scandal involving AII mean, you know, I suppose it
- 26:06would be where, you know, a company's marketing material was
- 26:11entirely created by AI and it turned out to be, you know,
- 26:15completely eating some creative lunch or something.
- 26:20I mean, I, I, the moment I can't quite envisage the kind of
- 26:24corporate earthquake type of disaster that that would that
- 26:31would happen, but I, I, that may well be right.
- 26:34But I very much hope that the work of people like David
- 26:37penetrate the corporate world so successfully, Michael, that we
- 26:41don't have to have of a crisis that causes behaviour to change.
- 26:45I'm a believer in voluntary corporate governance if that can
- 26:49be achieved. It's A tag team effort because
- 26:53what you're doing too, I remember we met in 2017 and you
- 26:55were chairing the UK strategy for AI.
- 26:58So you are well ahead of the curve as opposed to something
- 27:02that I could say more in the US where we're still behind the
- 27:04curve. But the other thing I would say,
- 27:06Michael, is there already are some examples where people have
- 27:09used AI for helping to write code.
- 27:13Some of the code that the AI suggests compiles and some of it
- 27:16does not compile at all. And so that should be warning
- 27:18indications. Do not blindly trust the
- 27:20machine. But even more interestingly
- 27:22enough, and I think this was in the last week or so,
- 27:26unfortunately one of the AI agents that help with writing
- 27:30code completely deleted a company's code library.
- 27:33So that should tell you there are things that are warning
- 27:37clouds on the horizon that if you do not think about a risk
- 27:39based assessment and you put this in charge of something
- 27:42that's very tied to your intellectual property or tied to
- 27:46your finances, bad things can possibly happen.
- 27:49I should mention the whole area of copyright infringement,
- 27:53Michael, has become incredibly controversial in Europe.
- 27:58And you could say that, you know, for reputational purposes,
- 28:02it's the AI model developers which are at risk of very, very
- 28:07strong reputational damage if we don't find a proper solution to
- 28:12the, you know, training on copyright material and so on.
- 28:15And then of course, earlier on, David mentioned the lawyers who,
- 28:19you know, had false citations. I mean, that, you know, I mean,
- 28:23the legal profession now they've wised up pretty quickly to the
- 28:27fact that they need to have closed systems which can be
- 28:30relied upon, not just asking ChatGPT to give you a, you know,
- 28:35a legal pleading that is highly dangerous.
- 28:38So there's already been quite a bit of wising.
- 28:41Let me just also mention that it's interesting.
- 28:47If we look at traditional security and data breaches,
- 28:51there have been massive data breaches.
- 28:54I think all of us have had our Social Security numbers and
- 28:57personally identifiable information, you know, spread
- 29:01out there on the web. There's no escaping it.
- 29:04And still these breaches happen all the time.
- 29:07One of our major retailers, Marks and Spencer, you know,
- 29:10they, they were unable to stock, unable to deliver, I mean, you
- 29:15know, for months and months. And I think they're still
- 29:18suffering from the after effects.
- 29:20So, yeah, I mean, risk management, risk assessment is,
- 29:24you know, so crucial in all of this.
- 29:27And, you know, AI needs to be added to all of this.
- 29:31And our first questioner rightly raised that it needs to be very
- 29:35much integrated into the whole risk management process.
- 29:40Here we have a question from LinkedIn.
- 29:42It's from Mohamud Gibrel and he says this is to David.
- 29:46He says. Do you see the need for new
- 29:50agent management tools and platforms, For example, managing
- 29:55agentic AI sprawl with security, compliance, cost and so forth?
- 30:01Short answer is yes. The slightly longer one is it
- 30:04may not look like some of the tools of the past because this
- 30:08is not steady state. Because we just talked about how
- 30:11generative AI and other AI approaches are, are possibly non
- 30:15deterministic and that they are generating their own content.
- 30:19If you do what has been done in the past with forensics and you
- 30:21assume steady state, you will always be behind.
- 30:24And so that's where I'm really excited about sort of patterns
- 30:27of life where CIOs and chief information security officers
- 30:30can go to their boards and say, look, we can actually get a, a
- 30:33win win where we upgrade the cybersecurity posture of our
- 30:36organization while simultaneously also getting it
- 30:39better, in better shape for managing AI risk and actually
- 30:42detecting AI risk. Because we're looking at what
- 30:44the normal patterns of life are for our information systems and
- 30:47our AI systems, and then if we see something odd, we're
- 30:50stopping it and then doing a pause and trying to figure out
- 30:53what happened here. So I think that's necessary.
- 30:55The other thing I would also give a nod to is for certain
- 30:58flavors of AI, especially the ones that are precompute.
- 31:02I triple E after about five years, released just about a
- 31:05month ago called the Spatial Web Protocols.
- 31:07And just like how everything on the web that we use HTTP to find
- 31:10things on the web and we use HTML to actually describe a web
- 31:13page. Now it's actually you can
- 31:15actually use what's called HSTP, which actually allows you to
- 31:19basically address anything in space and time almost if it was
- 31:22like a web domain or web URL. And you can use HSML to actually
- 31:27write things up and bound things by space and time.
- 31:30And So what does that mean for AI?
- 31:32We can actually write descriptions and say in general,
- 31:35autonomous AI systems that are flying plane should not plow
- 31:39into the ground and general cars should not actually like drive
- 31:42into other cars. You can actually begin to have
- 31:44space and time constraints. And even more interestingly,
- 31:46there's there's some inner early indications that you can
- 31:48actually take that and actually use it for policy as well,
- 31:51because policy often has boundaries.
- 31:53JPL Jet Propulsion Lab with NASA has been using this as
- 31:56experiments on NASA on the moon to actually because of the lag
- 32:00effect, giving these sort of like the commander's intent to
- 32:03the possible moon Rover or whatever is on the moon, but
- 32:06then allowing the AI that's on board actually on something that
- 32:09would be on the moon try to navigate within the space-time
- 32:12dimensions. And so this gives me hope that
- 32:15for some of the future of what we do with AI, especially if it
- 32:19is precompute, we can also bound it by space and time
- 32:22constraints. The other element of this is
- 32:25absolutely right at the corporate level and the adoption
- 32:28level at by business. But of course, you know, agentic
- 32:32AI, some of it's snake oil, some of it not has that autonomy, but
- 32:36it also is now becoming retail as well.
- 32:40So, you know, again, we have to make sure that, you know, if a
- 32:43business has 1000 employees, they're not all thinking, yeah,
- 32:47I'm going to adopt this agentic AI.
- 32:49I'm going to put all these tools together in it in a semi
- 32:55autonomous way because that is going to be a threat to the
- 32:58business as well. It's not just cyber security or
- 33:02data risk. It's, you know, decision risk.
- 33:04There are so many aspects to this.
- 33:07And so it double s the, the, the, the risk as far as I'm
- 33:10concerned, where you have that degree of autonomy in an AI
- 33:14agent that you know, could unravel an awful lot of things
- 33:18further down the line. So, you know, again, that means
- 33:22that boards and senior executives need to be very, very
- 33:26mindful. Arsalan Khan comes back and he
- 33:30says this is really intriguing. He says why can't we have AI
- 33:34write its own policies? What are the pros and cons of
- 33:38the AI deciding the policies and implementing the policies?
- 33:43There's no reason why you shouldn't, as long as you've
- 33:46trained the AI in the right kind of a way.
- 33:49You can be a monitor. You can, you know, I think you
- 33:52have to have some sort of generative adversarial kind of
- 33:56approach to it. You can't have, I don't think
- 33:58the same, you know, AI should be looking at itself, but I think
- 34:02what you need is 1 AI looking at another and that is going to
- 34:05become extremely prevalent in my view, where you have a mixture
- 34:09of tools, one is monitoring another one is risk assessment,
- 34:13seeing another. You know, after all, we know how
- 34:15quite a lot of work has been created.
- 34:18I remember when there was a, a painting in Paris that was then
- 34:22sold in created in Paris, sold in New York for half $1,000,000
- 34:29called I think it was some Eduardo de Bono.
- 34:32And it was created by this generative adversarial Network,
- 34:36and it was created by one AI criticizing the work of another
- 34:40until eventually it produced this rather splendid portrait.
- 34:45Well, you know, let's take that forward a few years and say, how
- 34:49do we make sure that A, we create the policy and then make
- 34:54sure that it's complied with. So, yeah, I mean, I think the
- 34:57chief compliance officer has already got tools of that kind
- 35:02in the box. This is from Chris Peterson, who
- 35:06is maybe taking a slightly contrarian view here.
- 35:10And Chris Peterson says, is there a balance we can strike
- 35:15between AI regulation and AI innovation?
- 35:18And here's the point, especially given the velocity of chains and
- 35:22the billions and trillions of dollars riding on the success of
- 35:27big AI businesses. In other words, sure, guys, we
- 35:31have to manage risks, but let's also not throw out the magic as
- 35:38we manage ourselves risk into the ground.
- 35:41Thoughts on that? If you use whether it's human
- 35:45produced or you use another AI to produce constraints for what
- 35:49an AI should and shouldn't do, then you allow innovation to
- 35:53happen within those constraints. And so if you have a separate
- 35:56system that is monitoring that harm is not coming either
- 35:58mentally or physically to a human being, then yes, that
- 36:02gives you an additional degree of of of confidence that you
- 36:05wouldn't have if you had just the AI system by itself.
- 36:07You may tell the AI system don't harm humans, don't harm them
- 36:11mentally or physically. But as we we know,
- 36:12unfortunately, whether a wrong prompt is asked or the data was
- 36:16incomplete or the non determinism, sometimes things
- 36:18can happen. So I think when I, when I brief
- 36:21companies and boards and even governments, I say, you know,
- 36:24what you really should be doing is not necessarily setting five
- 36:27year goals, but setting five year visions for the how you
- 36:30want the technology and society or technology in your company to
- 36:34perform and then use constraints as your friend.
- 36:37And so just as what turns him was saying that you could use AI
- 36:40to actually sort of bound in other AIS.
- 36:43There are other approaches too. And we've seen already, I've
- 36:47already seen there's some certain companies that are now
- 36:48using AI to help you coach to do better prompts to an AI.
- 36:52And so that's already a proof point that this is possible.
- 36:55So I think at the end of the day, you definitely can have
- 36:57both. And given that, and as Lord Tim
- 37:00said, you know, regulation will always be playing catch up with
- 37:03the technology by using constraints as opposed to
- 37:06setting very fixed things, which actually might be either out of
- 37:09date or or or or restrict innovation.
- 37:12If you do constraints, that allows you to still protect the
- 37:15things that need to be protected, but allow people to
- 37:17explore that space as well. I've never believed that good
- 37:20regulation is the enemy of innovation.
- 37:23I've always thought of it. It's actually the friend because
- 37:26we want the right kind of innovation that benefits humans,
- 37:30that doesn't substitute machines for everything that we do that
- 37:34is for our benefit and doesn't create some of the harms that
- 37:39we've been talking about today. And I think the way you do that
- 37:42is the way you set the principles out of the kind of AI
- 37:46that we want to see. And then we try and adopt
- 37:49standards, mandate particular standards.
- 37:52And we've got to be agile in the way that we do that.
- 37:54And the great thing about standards is they can change
- 37:57over time. You know, and we talked about
- 38:00NIST earlier, but there are many other organizations, IEEE, which
- 38:04David also mentioned earlier, ISO, the international standards
- 38:09organization. And all of them are trying to
- 38:11move towards interoperability internationally.
- 38:14And for me, that is I think the the the goal, if we can get that
- 38:18and we can mandate that that sort of quality into large
- 38:24language models, into other forms of AI, then I think we've
- 38:28made some real progress. But I think the idea is that the
- 38:31idea that no regulation means we can just have a complete free
- 38:36for all and produce any kind of innovation, whether it's good or
- 38:40bad for humanity, I think is what we need to push up against.
- 38:45Is there a distinction between this balance of a of regulation
- 38:51versus innovation when it comes to generative AI versus any
- 38:56other new technologies? Tim, any any thoughts on that?
- 38:59I always use the example of the automobile.
- 39:02I mean you look back at the automobile, what created the
- 39:05growth? I mean, it was the rules of the
- 39:08road. It was manufacturing standards,
- 39:11you know, and I I mean, and and it was 10 years when you look at
- 39:155th Ave. photographs, nineteen O 3 to 1913, that came in pretty
- 39:20fast. You know, it was a horse and
- 39:22carriage in 19 O3 automobiles except for one horse and
- 39:26carriage in 1913 going up 5th Ave.
- 39:29So we've been here before and we didn't just allow, you know, the
- 39:33automobile to run everybody down on the roads.
- 39:37Of course we started with a red flag which was probably, you
- 39:40know, may be safe in the circumstances.
- 39:42But very soon we got sensible regulation and look how
- 39:48innovative has the automobile industry been over the last 100
- 39:52years? Enormously.
- 39:53So I'm actually optimistic that we can get the right form of
- 39:57regulation without stifling progress and innovation.
- 40:02This is from Chris Davidson and David He says on LinkedIn.
- 40:06He says most of the conversation around AI centers on innovation
- 40:11from big companies, Open AI, Anthropic, Microsoft, so on.
- 40:16As an equity partner in a new AI integration startup, he's
- 40:21curious what you have to what you think are the biggest
- 40:25opportunities for the little guy in the marketplace.
- 40:29In other words, what are the areas of value on which smaller
- 40:32companies can stake a claim with minimal risk that they will get
- 40:37crushed by these big players in the next three to six months?
- 40:41So I'll give 3 real quick. One would be while these big
- 40:45players are focusing on generative AI and generative AI
- 40:48is actually using algorithms that, you know, deep neural
- 40:50networks that goes back to late 80s, early 90s.
- 40:53So it's a technology only possible today, but it's not
- 40:56necessarily new. I would say look at additional
- 41:00approaches that are coming out work of Carl Frisson on active
- 41:03inference. I mentioned that, that that's
- 41:05actually something that can be pre compute and be bounded.
- 41:08And I think the future is not going to be 1 AI method to rule
- 41:11them all. It's going to be mixed model.
- 41:12I mean, we know that's actually what what happens with us is
- 41:15that you use different things. And we may find that generative
- 41:17AI is really great for replacing national language processing.
- 41:21But as you see with the papers are coming out, these models are
- 41:23not reasoning. They're only doing pattern
- 41:26matching and into the degree that pattern matching and
- 41:27reasoning matches are doing that.
- 41:29So first is there are other AI approaches that are not getting
- 41:32anywhere near the oxygen that they need, partly because
- 41:34generative AI has, has has consumed so much.
- 41:37The second thing, so is we need AI at the edge, including
- 41:41generative AI at the edge. And, and so while these
- 41:44companies are selling you sort of the, the monolithic platform,
- 41:47you got to subscribe to their services.
- 41:49I want something that I can run on my phone.
- 41:51I want something that I can run on my laptop without it calling
- 41:53back. And so it's got to be able to
- 41:55operate in low, low, low bandwidth environments and low
- 41:58processor environments. And you've seen, you know,
- 42:01DeepSeek, that was example of distilling a model.
- 42:04I'm not necessarily saying use DeepSeek, but you also see that
- 42:07both Berkeley and Stanford took ChatGPT 3.5 and they went to
- 42:11Llama, the open weight model and and found ways to actually ask
- 42:15ChatGPT how to upgrade Llama to be almost as good as ChatGPT
- 42:193.5. So we need more innovation about
- 42:21AI that can run at the edge in disconnected environments or or
- 42:25low processing environments. Because at the end of the day,
- 42:27some people may not want to connect back to the cloud.
- 42:30And then finally, this is actually more the human
- 42:32dimension that right now, some of these companies, companies,
- 42:34I, I really try to have empathy for the fact that there's some
- 42:37companies out there saying like it's going to displace 50% of
- 42:40jobs. Well, actually, I don't think
- 42:42that's the case. I think while jobs may be lost,
- 42:44people will retrain. And so it's actually an either
- 42:47or. And So what I want to see is AI
- 42:49companies and, and AI startups that are actually helping people
- 42:52navigate the chasm between you used to do this, now you're
- 42:55going to be doing this with AI. How do I accelerate your
- 42:58learning journey so that you're, you're, yes, your job has been
- 43:02let go, but you're now doing a different type of job and you're
- 43:05actually much more productive. And so navigating that in a way
- 43:08that gives people agency as well as reducing anxiety.
- 43:12I think there's going to be a whole industry there.
- 43:14And I think right now anyone who's just saying it's just
- 43:16going to be a loss of all jobs period without saying.
- 43:19And at the same time, new jobs will be created.
- 43:21I'm really interested in the new jobs that will be created and
- 43:24how we actually have private industry help people skill up
- 43:27for that. Lord Tim, thoughts on very
- 43:30quickly on this issue of opportunity.
- 43:32How can small players survive? Yes, very quickly, I thought
- 43:37David's point was very shrewd. My point would be actually data
- 43:41is still going to be king in all of this.
- 43:44The large language models are a commodity.
- 43:46I don't think they've made any money from the large language
- 43:51models. So my view would be I would go
- 43:53to open source or distill in the way that David discussed and
- 43:57find a proprietary database that I had an exclusive license to
- 44:03which had a particular use maybe in healthcare or education or
- 44:07agriculture, something like that.
- 44:09And I was the only AI developer who had the access to that data
- 44:14and was able to use my AI model. And I think then I would, and I
- 44:19would get an open source model to help help me with all that.
- 44:23And I think then I would be in business.
- 44:24But you know, I'm not an AI developer.
- 44:27What do I know? We need to spend a few minutes
- 44:31before we're done here talking about jobs and the impact of AI,
- 44:38generative AI on jobs. David, you want to take some
- 44:43crack at this first? When I met Lord Tim back in
- 44:462017, what really resonated at the time with the UK government,
- 44:49they were proposing data trust. Here in the United States, we
- 44:52call them data cooperatives. But this idea that businesses or
- 44:56people could come together, like maybe we're all musicians or
- 44:59we're all artists and we actually say we are willing to
- 45:02let our data be used for the following purposes.
- 45:04Well, here we are now in in 2025, looking at 2026.
- 45:08We need that in the United States.
- 45:10I mean, because I actually say that right now we have a few
- 45:12companies that seem to be repeating the lessons of Napster
- 45:15where they have acquired data, but not necessarily respected
- 45:18intellectual property, not necessarily respected the equity
- 45:20of the people that produced it. And it also makes it harder
- 45:23whether or not I can trust it, because when the machine gives
- 45:25me an answer, where did it come from?
- 45:27So I think this does put the premium on exactly what Laura
- 45:30Tim said, which is the future of jobs is whether it's a
- 45:33interesting data set or it's people that are musicians,
- 45:36artists, entertainers, or maybe it's also we just care about
- 45:39finding a cure for Parkinson's. On top of it, though, we no
- 45:42longer have to shift the data. One, that's a cybersecurity
- 45:45risk, but then two, what's called Federated learning, where
- 45:47the algorithms actually come and learn in situ on the data
- 45:50itself. And we can actually monitor
- 45:52what's being learned. And then when it leaves, we
- 45:54actually record that. And then there's some either
- 45:56financial exchange of value or some non financial, some other
- 45:58benefit that it's equity and whatever cure for Parkinson's or
- 46:02whatever research was done. I think the future of jobs is
- 46:05both rethinking data so that we can actually make sure people
- 46:08have some sense of stakeholderism in their data.
- 46:11And then again, it's it's it's it's, it's, it's getting out
- 46:15there and saying this whole narrative of it's going to kill
- 46:17jobs period, as opposed to it's going to displace jobs.
- 46:20Much like how when the automobile showed up, well, yes,
- 46:23horse drawn characters and those sort of things, they went away.
- 46:26But then there were taxicab drivers or truck drivers or
- 46:28people that maintain cars. New jobs were created.
- 46:31And so the faster we can actually help people be aware
- 46:35that new jobs will be created and then help them with that
- 46:38journey, I think will actually will actually reduce some of the
- 46:42anxiety that's present in societies at the moment.
- 46:44I take that data communities aspect extremely seriously.
- 46:50I think valuing our data and making sure that collectively in
- 46:54the way that David describes it, we can make use of that and
- 46:58value it and monetize it is going to be a very important.
- 47:02I chair an author's licensing collecting society, and I can
- 47:06see the equivalent happening in all kinds of areas, not just the
- 47:10creative industries, but, you know, farmers collectively, for
- 47:14instance, with their data and being able to use that and in a
- 47:18sense parlay, you know, license it, make make make it valuable
- 47:23for, for, for use by AI developers.
- 47:26So I think entirely right. The second thing I would say is
- 47:29that I think that human empathy will still be a requirement.
- 47:33I think those human creative qualities are still going to be
- 47:38important in the future. So, but and then of course what
- 47:42we've already mentioned is that critical thinking that's needed
- 47:47because if we, if we don't think critically, then frankly AI is
- 47:50definitely going to take over and we're not going to have
- 47:54really much of A function as particularly if AGI comes along
- 47:58anytime soon. Do you have thoughts on what
- 48:02kinds of jobs will get displaced by AI?
- 48:06When what? What kinds of jobs will not and
- 48:09and very quickly please? If you can stick into Claude or
- 48:12ChatGPT, you know, tell me, give me a business plan for the next
- 48:165 years, you know, and this is my vision and this is these,
- 48:21these are the annual reports of my competitors.
- 48:24Then, you know, I think you're and that's exactly what you can
- 48:28do nowadays. Then you're going to find that
- 48:30kind of job going. So I think this is a white
- 48:34collar revolution basically. And I don't think we've been
- 48:37here before in anything like the same way.
- 48:39And it's not a blue collar revolution that David was
- 48:42talking about drivers and so on. I mean, you know, we have a gig
- 48:46economy, but that doesn't and they're driven by algorithms in
- 48:49terms of their performance. But that doesn't mean to say
- 48:52that those jobs are going. People are still going to be
- 48:55riding around on scooters and motorbikes in the future.
- 49:00We're going to be the ones, the professionals are the ones who
- 49:03are at risk. David literally in one sentence,
- 49:06what jobs will leave and what jobs will remain just being
- 49:11displaced or not? So much like the printing press
- 49:13where scribes went out of work but typesetters arrived.
- 49:16If you were in the business of web development, interface
- 49:19development, that'll increasingly be done by AI.
- 49:22Anything that that the past is like the present, that'll be
- 49:25done by AI. The new jobs will be the jobs
- 49:28that are novel. Where the past is, you know,
- 49:30past does not inform the presence, where you have to
- 49:33think on your feet, where each day is different.
- 49:35And the people that are then when they look at what comes out
- 49:38of the AI are being those skeptics, being those critical
- 49:41analysis. Arsalan Khan says something
- 49:43interesting. Whoever defines ethical
- 49:45boundaries defines the direction of AI.
- 49:48If the world doesn't have consistent ethical boundaries,
- 49:51then AI is just a reflection of our own biases.
- 49:57Who wants to just very quickly take a crack?
- 50:00That's an interesting statement, but let's think back to the
- 50:021700s, Arsalan. There were plenty of things that
- 50:04people thought in the 1700s that are ethical that we would not
- 50:07think are ethical now. Same thing in 1800s I often say
- 50:10to my British counterparts in War One, the British thought
- 50:14that queue boats were ethical and submarines were not ethical
- 50:17and switched to World War 2 that had flipped.
- 50:20Ethicals are socially and temporally defined.
- 50:23What I think he's saying though is what are the principles we
- 50:25want to hold true to? And I think that is an important
- 50:28conversation to have. But recognize that ethics, you
- 50:31know, there are plenty of things we thought 200 years ago were
- 50:33ethical that are not ethical now.
- 50:34Let's make sure that as we look at ethics now and we spend time
- 50:38on it, maybe what we really be should be saying is what are the
- 50:40principles and the constraints that we care about as opposed to
- 50:43just ethics? There are principles that have
- 50:46been agreed internationally. They may not be agreed in terms
- 50:50of regulation, but the OECD principles, China signed up to
- 50:54those, you know, G7G20. So I, I think, I think I'd be
- 50:59more optimistic than Ashland. Looks like Wei Wang from
- 51:03LinkedIn is going to get the last word here and I'll just ask
- 51:05you each for the one quick sounds sound bite sentence.
- 51:11She says this when there is a gap in popular perception and
- 51:15the technical reality of implementing reliable reliant AI
- 51:19solutions. What advice do you have to for
- 51:25teaching and explaining clearly the misconceptions that and to
- 51:32build the rigor into under to helping convey the complexity?
- 51:36So it sounds like the individual was trying to make the case to
- 51:39either like a board or C-Suite or whatever.
- 51:41And I would say bring examples because there are plenty of
- 51:45examples, but also let you know. And I wouldn't necessarily do
- 51:49this for the board because they may not have the bandwidth, but
- 51:50for your employees and your staff, let them play with the
- 51:53technology and see when it works and when it's fallible, because
- 51:55then it actually makes it much more real and it's, it's
- 51:57visible. However, I think this is a case
- 52:00where, again, whether it's case studies or experiential itself,
- 52:03you almost have to sort of show and make it real because
- 52:06otherwise people will not see the value.
- 52:09And again, that that that fact that we see the machine and we
- 52:11think it must be like us because it's talking to us or it's
- 52:14giving us text. We need to recognise that our
- 52:16own human emotions will run the risk of missing the fact that we
- 52:20have to be sceptical of these alien interactions.
- 52:22We are going to be the victims of our own trust if we're not
- 52:26careful. And you know, I, I'm a great
- 52:29believer in encouraging a culture of curiosity, but I
- 52:32think we have to encourage a culture of caution as well.
- 52:35I think we just have to get the balance right between the two.
- 52:39And if we can encourage that, and it's not just business, it's
- 52:42government, it's families, you know, and where way behind the
- 52:46curve, unfortunately, because technology is moving so fast.
- 52:50Well, with that, we're out of time and clearly there's a lot
- 52:55more to discuss, so I hope you'll both come back and be
- 52:59guests on CXO Talk in the future so we can continue this
- 53:02conversation. Thank you very much, Michael.
- 53:05It was great to be here. Truly appreciate it, Michael.
- 53:09And a huge thank you to Lord Tim Clement Jones and Doctor David A
- 53:14Bray, and to the amazing audience who asked such
- 53:18excellent questions. Thank you all.
- 53:21We have incredible shows coming up.
- 53:23Check out cxotalk.com, subscribe to the newsletter, and we'll see
- 53:27you again next time. Take care, everybody.