PassW0rd – 8 November 2023 (AI Safety Summit)

Speaker A: This program is brought to you by Resonance FM. If you like what you hear, please support our work by making a donation at resonancefm.com/donate.

Speaker B: Hello and welcome to Password on Resonance FM with me, Peter Warren. Possibly the best-informed monthly technology radio program in the UK. In this episode, surprise, surprise, artificial intelligence regulation. Was Prime Minister Rishi Sunak’s quaintly named AI Safety Summit a success? Why did our diminutive leader decide to try to take center stage on the technology? What was it all about? We apologize in advance to the number of professors in the program, but it’s hard to find anyone in charge of an AI company today who is not a professor or has written a book about it. Password has been examining AI for years now. Our first program about it was around 2010, and since then it has featured in every season of the show. Sometimes twice. Last year, even I wrote a book on AI regulation, AI on Trial, and in 2014, we warned both the EU and the French Senate about the technology, traveling to Brussels and Paris to give speeches at the invitation of both bodies. In 2018, the Finnish technology entrepreneur and former head of Nokia, Risto Siilasmaa, became so worried about the technology and the lack of political understanding about it that he taught himself how to program in AI and toured the world trying to raise political awareness. So what’s changed? AI, as it has now been fashionably dubbed, has been around for decades, though often in the past it was called machine learning, a much less glamorous term. So why suddenly is AI or machine learning being taken so seriously? To answer that question, you have to delve into the recent history of technology and the fact that since around 1996, we’ve had the internet. Before that momentous event, there had been two significant AI winters, periods when research funding became disillusioned with the technology and deserted it. One in the late 1970s, showing just how long machine learning has been with us, and the second starting in 1987 and going on until around 2000. The reason was simple. Machine learning needs plentiful amounts of two things: ready access to data and research cash. With the advent of the internet, Both were available in abundance because US big tech, courtesy of the internet, had loads of cash and lots of data because they were offering data storage in the cloud via the internet. For US big tech, it could not get much better because via the terms and conditions policy that no one ever reads, not only were they getting huge amounts of social media data but they’re also getting personal login and location data. The big tech firms not only know who you are, they know what you’re doing, and arguably, courtesy of social media, what you are thinking. But with such massive data gathering came two issues: bias in the data and huge responsibility for what machine learning can do. The headlines have been full of computers taking over the world and taking control from us, but the real issues are that they will do that by taking away mundane tasks from us and by discriminating against us due to bad data in the system. They simply won’t be fair. Here’s AI scientist Simon Bain, chief executive of OmniIndex, a company described as a context-aware data intelligence platform. As OmniIndex’s blurb says, the OmniIndex enables your unstructured data sources to be part of your analytics by combining artificial intelligence for contextual awareness of the content, geodata, AI sentiment analysis. OmniIndex can provide full analytics on the other 80%. Essentially knowing who you are and what you’re doing.

Speaker C: And the issue we have with AI, in my own humble opinion, is one of biasness. So if we look at things like OpenAI, Google Bard, they’re trained on everything that anyone has ever put onto the internet. And generally speaking, stuff that goes onto the internet is done by white middle-class people. Which means there’s an inherent biasness instantly placed on that data, putting aside all of the right and the left-wing gibberish that goes in there as well. So the issue we currently have with AI is one of bias. If someone takes a large language model and uses it to create a document, create a news report, then it’s going to be biased towards white middle-class male people. And I think that’s where we all need to look at more than anywhere else. The ethics of AI taking over the world? Yeah, I don’t think so in my lifetime. We have on-off buttons for that. We just turn the thing off. And if they don’t— powered by any known source that we currently know of, well, when we do create that source, we will know of it and we’ll know how to turn it off. So I’m really not worried about that side of the ethics, which I know a lot of people are. Again, I believe most of that is hype and PR. When someone wants to sell something, they make sure it’s in the news. What better way to put it in the news by telling you it’s going to take over the world and kill off all humanity as we know it?

Speaker B: That’s one of the issues, isn’t it? Because the other point about this too is it’s called AI. This term AI has been spawned, whereas in fact we’re talking about something called machine learning. And machine learning has been with us for quite a long time.

Speaker C: Thank you, thank you, thank you, thank you. I thought I was the only person out there who knew that. So my background is with text manipulation. Text manipulation is now known as a large language model, i.e., getting sentiment from something, getting context from something, creating something from a whole load of stuff. It’s machine learning. Artificial intelligence is not text manipulation. It is just machine learning. Take in a whole load of garbage, learn from it, and spew something out. The AI— there is some really cool AI out there at this moment in time. DeepMind, one of Google’s companies, they do some really good AI with the National Health Service on X-rays and the like, where they do actually do artificial intelligence. But they also do a whole load of machine learning, which has somehow been put into that whole bracket of AI, and it really shouldn’t be. I come back, I was listening to a show, podcast, a couple of nights ago, and they were talking about goo, gray goo made up from nanobots, and how Prince Charles 6 years ago or whatever said that we’re going to be overrun by gray goo from nanobots. It’s never happened. It’s not likely to happen. And currently we’re in the phase of AI going to take over the world. And in 5 years’ time, it’ll be something else. Whoever wants to put themselves into the news at that moment in time. But it is just machine learning. It is just taking what we have put in and learning stuff from it and taking out.

Speaker B: The other issue with this, excuse me for butting in, but it’s also this PR issue because it’s easier to explain things in terms of Terminator or in terms of particular films. And so the media is a bit of a sort of supine victim in all of this. It instantly starts trumpeting this message of robots are going to take over the world or that we’re going to lose our brains into the machine.

Speaker C: I think I’m not sure when I last read the news that said machine learning is going to help us construct a new hospital. It doesn’t make news, does it? Whereas AI is, is going to take over the world and put everybody in hospital, now that’s more of a headline. And I think that is the issue. And then it comes down to why the press is doing that. Well, it’s because us as readers don’t want to read a boring story. We want to read the stories that make us think, however right or wrong they are. And as soon as they go into print whether it be computer print or physical print, then a lot of people actually start believing it. And the more time you peddle it, the more people believe it. And that’s kind of where machine learning comes into its own, because it peddles enormously. I think the issue we have currently is, is one of overhype, is of PR, people trying to sell stuff. Open AI wanting to sell their bits. Microsoft wants to sell their bits. Google wants to sell their bits. What better way of doing that than screaming out a headline that everyone’s going to be killed? Elon Musk doesn’t really do AI now, although having had some of his stuff, he does. But he wants to make sure he keeps in the news.

Speaker B: Data scientist Simon Bain, the remarkably candid CEO of AI firm OmniIndex. Which brings us quite neatly around to the recent AI Safety Summit. Rumors about the event started shortly after an open letter from the Future of Life Institute emerged calling for a 6-month moratorium on AI due to fears over its uncontrolled development. Signed by Elon Musk and Berkeley University of California Professor Stuart Russell, author of Human Compatible, unless lites like myself The letter was the idea of another leading AI critic, Max Tegmark, a man famous for saying of AI that walking closer to the cliff edge to get a better view was not wise. Tegmark, a Swedish-American physicist and cosmologist, a professor of physics at the Massachusetts Institute of Technology and president of the Future of Life Institute, received widespread support. The Future of Life Institute letter also propelled AI development up the political agenda, a process that had already occurred in a very aware European Union, and according to Future Intelligence’s high-level sources, one that was developing global competition. Put bluntly, regulating AI rapidly became an issue of global power, with the US keen to assert its dominance in the area and acutely aware that the EU’s General Data Protection Regulations had stolen a march on it. There were also claims that regulating AI itself represented an economic opportunity. Controlling AI had suddenly become all about status and started a fierce competition amongst politicians and nations to assert supremacy and set a Brexited Britain on a frantic race to claim AI regulation as a UK initiative. Being seen to be both statesmanlike and responsible was compelling for a tattered UK looking to promote an image as a relevant world power. With all significant venues in London already booked, the search settled on Bletchley Park, according to many, the home of modern computing due to the research carried out by Alan Turing. A mathematical genius often called the father of computing. The stage was set. Tegmark, Russell, and others were the warm-up act on the Monday, and the finale was Rishi Sunak interviewing Elon Musk about whether the AI of the press would take over our lives and whether it would leave us out of jobs. It was a soiree that led to some critics stating that Sunak actually ended up looking less than statesmanlike and that his position was demeaning. The Guardian even suggested that Sunak, having seen the success of former Deputy Prime Minister Nick Clegg with Meta, was lobbying for a job in Silicon Valley. And like Banquo’s ghost, always at the back, the spectre of the sentient AI super bean hurrying near. So what’s the truth? Over the years, we’ve carried out a lot of interviews with some of the world’s leading experts in a number of fields, not just in AI, but particularly in business. And we went back to some of these people and asked them. Let’s take jobs. Here’s Massachusetts Institute of Technology’s Professor Yossi Sheffi, author of The Magic Conveyor Belt and director of the MIT Center for Transportation and Logistics.

Speaker D: Well, for several reasons, but let’s just admit that two things. First of all, there are two very extreme sides to the, to the forecasting regarding generative AI. Some serious people like, I would say, Elon Musk, Bill Gates, think that it— or Yuval Harari think that, you know, it’s a danger to democracy, danger to humanity. Other people, not less distinguished, think it will bring a new age of great standard of living and great productivity and all this. There are several reasons why I think the anxiety is overblown. First of all, I live in the United States, Ralph. 3.5% unemployment. With all the brouhaha, there are 2.5 job openings for every job applicant. So we certainly don’t see it. With all, even with all the announcement of high-tech laying off people, all these people are absorbed immediately. So first of all, it’s not happening yet. Second, it usually takes incredibly long time to take place such a change. One example, 1892, Bell Telephone invented the automatic telephone exchange. Until then, there were people, mostly women, plugging, you know, plugs into holes. Well, in 1950, there were still 350,000 exchange operators. Only in 1980, there were no more exchange operators. It took 9 decades. Until the innovation had this impact. People say, well, now it’s of course a lot faster, and it’s true because it goes by software and you don’t need— it’s not like the steam revolution when you had to build new physical plants and it took a long time. But still, even now, job destruction is a very, very slow process. You know, in 2013, University of Oxford had a report that 37% of US jobs will be eliminated by 2023. Well, not quite. I mean, not quite with massive, you know, employment. The number of unemployed people is all-time high and all this. So, okay, I understand. I also understand another thing: the media, bad news create eyeballs and clicks. For some reason, I’m still, I’m not a psychologist. I don’t understand it. People like, people are attracted to bad news. So for the media, the sky is falling, but let me also mention several other things. So not only the slow process, there are three other things that have nothing to do with the speed of the software, the speed of the internet or whatever. First of all, Labor union, labor union, they’re not going to take it lying down. They’re going to elongate the process. Number 2 is regulations. Governments are not, don’t know how to deal with it yet. And regulation in many cases slow the process. Italy now basically outlaws ChatGPT. The Chinese government is regulating, actually very smartly, by regulating the training data, that you cannot use certain training data so it doesn’t go and say things that are problematic. Then there’s the issue of public acceptance. A lot of these things are gonna, you know, involve smart automation. So the question is, how fast will the public accept huge truck with no driver running on the M25, you know, with no driver. We know, for example, today a 777, a 787, an A350 can go gate to gate with no pilot. How many of your listeners would go on an airplane with no pilot on it? My guess is the number is exactly zero. This issue of acceptance. One more example. In 1945, New York had a strike of elevator operators, another job that doesn’t exist anymore. New York came to a grinding halt. People were afraid to get into an elevator, even though they always saw how it’s just on and off, how the operator was operating the elevator. And yet people are afraid to go on an elevator without an operator. New York had to settle with them because the tall office buildings were not working. So there’s an issue of acceptance, which is, which we are not sure how it will work. The fear that we see today may play into it. So I don’t think it will happen. Whatever happens will not happen very fast. And the last thing that I want to say is Since we don’t know really, you know, you ask me as maybe a person who wrote a book about it, so I seem to know something, or at least researched the topic, and there are points of view all over the map. So it’s very hard to say what will happen. So we look at the only thing that we can look, we look at history. And in every industrial revolution, some jobs were eliminated, many jobs were changed. But a lot more jobs were created, and in many cases it is very hard to predict where the new job will be.

Speaker B: MIT’s Professor Yossi Sheffi, author of The Magic Conveyor Belt. So what are the risks? What can machine learning do to us? Why are we calling it AI? One of the big threats Oddly not heavily flagged up in Rishi Sunak’s AI Safety Summit was impersonation. The capability of machine learning systems to impersonate us is terrifying. Not only can the technology copy our voices, our faces, and our bodies, as was graphically illustrated recently when an AI system was used to create nude images of a number of Spanish schoolgirls by taking their body shapes, rendering them naked, and then placing their heads on the models. Even more worryingly, incredibly convincing copies of our voices can be created using the technology. A point behind the recent strikes in Hollywood by film extras, who, like the Hollywood scriptwriters who also walked out due to fears of having their work replicated, struck over fears that their faces and bodies would be rendered into films, thus depriving them of work. It’s a point made by the US West Coast lawyer Shirone Finkelstein.

Speaker E: I think that that’s our biggest problem, that you have all this data, you have unregulated use of AI, and you have people wanting to rightfully protect their name, image, and likeness.

Speaker B: I mean, is this what some of the strikes that are going on in Hollywood are about at the moment? Is this because some people are saying, particularly some of the extras, they’re saying You’re going to film us and then we’re not going to get paid ever again because you’ll just keep on using us forever.

Speaker E: That’s exactly it. That’s one of the big sticking points right now in the SAG-AFTRA strike, where I believe also in the WGA, the Writers Guild strike as well, which is there needs to be control and kind of backstops against the use of AI to misappropriate people’s likeness. This is less of an issue when it comes to you know, the big name celebrities because they’re entering into a contract that is highly negotiated by lawyers on both sides. Every single point is accounted for. But when you’re dealing with extras, with maybe actors who don’t have a line, a speaking role, they’re signing basically union contracts. These are not negotiated things. And so if they sign something where it says that the company, the production company, now has the right to use their images, in AI, either an input or to take it and then create something out of that, they can basically pay an extra for one day on set and then use their image for multiple projects into the future. So what the actors are saying is no, there needs to be some protection and it needs to be not just consent but informed consent. So you’re giving, you know, if we’re going to consent to the use of our name, image, and likeness, it can’t just be in perpetuity, we want to know on a project-by-project basis what that is and for us to be paid accordingly, even if we’re not showing up on set. So it’ll be interesting to see where that lands, but that is a big concern among actors and is one of the big points being discussed in these strike negotiations.

Speaker B: I mean, do you own yourself?

Speaker E: You should. So it depends. I can’t give you a clear yes or no because it depends on the context. If a celebrity is walking around and somebody takes a photo of them and it’s published in a news article, well, there’s First Amendment protections. So there’s not a complete unfettered right to control your image. You are out in the public, and especially as you become more of a public figure, you arguably have somewhat of a less lower protection in the public sphere when it comes to reporting on newsworthy events, etc. So there’s the First Amendment protection considerations, but then there’s the flip side, which is, okay, well, who can monetize my name, image, and likeness? And so in many states, California particularly, there is a strong presumption in favor of a person being able to control the commercial use of their name, image, and likeness.

Speaker B: We are beginning to see these cases that are occurring of, say, revenge porn, for example, where people are taking somebody’s head and putting it onto somebody else doing a particular act. And then they’ve been able to say, no, I’m being impugned, I’m being maligned by this activity. And people have been imprisoned and they’ve been fined for doing this. So do you think that we’re going to have to develop, evolve the law to take into account these technological changes?

Speaker E: I do, because right now What gets tricky is that there’s no clear framework for stopping this, both at the input and the output level. So you have the input data, which might be a copyright infringement. For example, if you’re taking images, you know, you’re taking clips from a movie, putting it into this AI database, that might be a copyright infringement because that’s an unauthorized duplication of a work. Then you have the output, which is an unauthorized derivative work, arguably. So you have different points of infringement, and we don’t really have the tools to be able to address that.

Speaker B: Californian copyright lawyer Shirone Finkelstein on whether you have the right to own your voice, your face, and your body— essentially extremely personal data about you. Something that enraged the Spanish schoolgirls and their parents over the nude copies of their bodies being shared by their male classmates. Which of course raises a question that here on Password we’ve been exploring for years. This is simply another extension of that data that Big Tech already has upon us, the data we have cheerfully handed over via our social media and our mobile phones. It might seem abstract, but the machines are in a position to take all of us Something author and Professor Richard Stainings, an AI cybersecurity consultant for the healthcare company Cylira, says that we should guard against. AI, says Stainings, can be configured to not only take over us but to take over our homes because of the huge amount of technology we naively install in them.

Speaker F: Well, there’s a lot of things, a lot of bad things that could happen, and I think Concern with some of the advances in nefarious artificial intelligence are cause for concern right the way across the population. We are seeing a big rise in the use of good AI, bad AI, and positively very dangerous AI. We are seeing the weaponization of malware using AI to make it enhanced. And to enable it to get past the defenses of the cybersecurity tools that we use to protect our computers, to protect our critical infrastructure, our businesses, our schools, our universities, and of course the government. And this is, this is cause for concern. And it’s not just computers, it’s also the plethora of IoT equipment that we now have in our houses, whether it’s, you know, nanny cams, or camera doorbells or internet-connected thermostats, or even some of the other little appliances that we have running around vacuuming our floors or mopping the floors, for example.

Speaker B: Yeah, I mean, that’s one of the things that people don’t think about, isn’t it? Because they don’t realize there’s a lot of processing power they’ve deployed in their house. A lot of the time it’s not used, and they wouldn’t even think that anybody would possibly be using it for that. Yet the criminals are clever enough to put this web into our houses and turn them into little criminal enterprises.

Speaker F: And all businesses now have some level of IoT. I was talking to an emergency physician, or actually I was talking to a doctor the other day, and she wasn’t aware that her home speaker was listening into her conversations on the phone from home with patients. These are the types of things that we need to make ourselves aware of.

Speaker B: And with AI, this just is going to become compounded, isn’t it? Because not only will these AI systems be able to build together these networks, but also they’ll be thinking about different ways that they can actually exploit this. They’ll be learning from other parts of the network and saying, oh, last time we came across that data, we could do this with that. And so let’s start something happening. And you’re not really going to be able to combat that because it will be happening so fast.

Speaker F: Yeah, AI is obviously a form of intelligence, right? Machine learning, deep learning, various forms of artificial intelligence here. It learns as it goes. So it will observe behavior, your behavior, my behavior, and it will learn to perfectly emulate us so that our spouses might not be able to differentiate between an AI-generated interactive conversation with us compared to the real McCoy. Yeah, absolutely. And we’ve seen recent use of artificial intelligence via a thing called deepfakes, which is obviously the emulation of people either using video or audio or video and audio together. In this case, it was an audio recording that was completely generated by AI that purported to record a conversation between a politician and a journalist about how they fixed the election. And there was another video or audio, another audio bot that was recorded where one politician standing for election was talking about how they manage the level of alcoholism in the country and they were going to increase the price of beer. You can imagine how popular that was amongst the population. But both of these were artificial intelligence generated. They were totally fake. And in this case, it was able to sway the election we believe, in Slovakia to a pro-Russian politician. We haven’t found the causal link between the AI who generated the AI and placed it out there, but there’s a fairly good probability that it was a pro-Russian sympathizer that wanted to sway Slovakia towards a more pro-Putin stance. And this evidently was successful.

Speaker B: Professor Richard Steining alluding to the fake news that has poisoned our lives since the onset of social media. And seen investigations into companies like Cambridge Analytica, which were accused of influencing the US election in favor of Donald Trump on behalf of the Russians. From our data to our homes— another issue brushed on by Rishi Sunak. But will the machines absorb us? It’s been the staple of films like the Kubrick-Spielberg AI and Lucy,. But will the ultimate sci-fi horror film become reality? Two weeks ago, we traveled ahead of the AI Safety Summit to Athens to interview the neuroscientist and Nobel laureate Professor Edward Moser, who was speaking at an event organized by the cybersecurity and technology ethics company ESET. We asked Edward Moser whether we could become lost in an AI. Could we become part of the machine? Moser has been putting probes into the brain to record its activity. Indeed, he used his work to discover how the brain senses time and space, the work he won the Nobel Prize for. So could his probes discover how the brain works and network us? Into an AI system.

Speaker A: First of all, you can record the activity, and yes, as we have shown in our work, you can very easily tell, for example, where an animal is, or even where an animal thinks it is, because a certain group of cells light up, for example, being related to being in a different place than you are. And then we know that the subject then thinks, probably likely thinks, that he or she is there and not here. And you can do the the same with images that you have in the brain and so on. But the first thing I would say here is that people often raise the possibility, “Well, then you can read the mind from this.” And I think the very thoughts we have at any given moment is very complicated because it’s based on the interaction of cells that are all over the brain, very distributed and I think it’s still a very long way to get to the point where you can actually decipher the flow of activity at the millisecond time level that’s going on in the brain. And at least it would require that you have electrodes implanted many places in the brain and that you could pick up many more cells than what we do in the animals. So I think I’m not too worried about that. At the moment, at least because I don’t think most people would allow or want to, or no one perhaps would want those electrodes implanted in the brain, because you have to go inside the brain at the moment to pick up those signals. There’s no way you can get it from outside the brain. So that’s the recording part of it. And then you’re asking about the stimulation. Is there a way we can go into the brain and sort of artificially induce activity in certain cells so that I, for example, could induce a thought in your brain. So, first of all, technically, again, you can’t do that because I think we wouldn’t want to put— or no one would want to or even would put probes into the brain in this way. But the other thing is that you would have to know specifically which of the many billions of neurons to stimulate and when, because— they are not static, their role changes over time too. So, it is still beyond imaginable that you can induce complicated thoughts. I think you can induce much simpler things, and I can take one example. So, what you can do is, for example, you can control a movement of an artificial arm. So, people who have an artificial arm because they have lost their arm, then you can you can record their activity from the brain and then use that to have them think or produce a certain thought, and that thought will then produce an arm movement. So controlling movement and so on is possible, but that is very, very simple because it’s sort of the output part of the brain. If you really go into the brain where the thoughts are produced, then it is far from possible at the moment.

Speaker B: I mean, you talk about input/output. There is talk about connecting up exoskeletons to people to augment them, so that, you know, as you say, with disabilities, but also for soldiers. Do you think that this thing, this research, needs to be regulated? Lots of talk about regulating AI at the moment.

Speaker A: So yeah, I do. So I think I think it reminds me somewhat about when I was young, then it was the biotechnology revolution, right? And suddenly you knew how to clone humans and you can do all of that. And legislation came in place. And same thing now. But the problem is that the development goes in steps. It’s not really linear. So sometimes it goes so fast people who try to work out legislation are far behind and it’s hard to keep up. And I think now AI is developing so fast. So just over the last year with these large language models, it’s amazing how much they— what they can produce and how human-like the products actually can get. And of course, it takes much more than a year to produce that legislation. It has to come and I think it’s a very important task. But the challenge is also that this has to be international. It doesn’t help that one or two countries do it and then the rest does not. I am not an expert on legislation, but I can see that it’s not that easy because of the potential for misuse. There may always be someone who don’t care about the legislation.

Speaker B: Nobel laureate Professor Edvard Moser talking about the possibility of building us into a machine learning system and developing superintelligence, and how far away that development may be. Though for Moser, its use will be in helping us understand the brain. Here he is again on its role in neuroscience research.

Speaker A: Yeah, I think AI can help us both in, in practical ways but also more fundamentally. So the practical thing is, of course, that you can interpret data from the brain, which is now so huge datasets, very complicated, just finding patterns in this among lots of noise, right? But this is something the human eye can’t do any longer. And AI comes in as a very useful tool to see these patterns of activity and the ways things work in the brain, but it’s also at a more fundamental level conceptually, because if you can figure out what AI does to solve a certain problem based on huge amounts of information extracted maybe over long, long training periods, then it is often a clue to what actually the brain does too. So it’s not just a practical tool but also a suggestion of how how the problem could be solved, and then you can go and see, is this really what the brain is doing? Sometimes the answer is yes, sometimes it’s no, and probably there’s something in between. But we need these models or ideas of how could things happen to actually guide our search for brain mechanisms.

Speaker B: That’s fascinating, isn’t it? So the AI can actually teach us something about how we think.

Speaker C: Exactly.

Speaker A: So it comes up, it can come up a lot of suggestions how things could be. And the way AI solves the problem very efficiently is often— there aren’t too many ways to solve a problem in a very efficient manner. And sometimes that’s— if you convert the processes in AI to neurons in the brain, you can actually get ideas you wouldn’t have gotten if you started from scratch.

Speaker B: Should we be scared of all of this though? Is there— because lots of people we’re talking about this potential fusion between us and the machine.

Speaker A: I think there are reasons to be careful, but just as a research tool, I think it is a good thing. I mean, it helps us to explore the possibilities and to do it much more efficiently. Implicit in your question, I think you’re raising the possibility that there is some synthesis between brain and machine that may happen. If we understand them both, they can also talk to each other, and you could even imagine you implant devices in the brain that speak to the rest of the brain and so on. I think, of course, here we need to put the brakes on, but I also think that maybe it will still take a while before we really get there. So we can elaborate on that if you want, but I’m not so scared for the immediate future, but I do think important, like for any new technology, that we have to, um, be aware of the consequences, good and bad, and then try to navigate in that landscape.

Speaker B: Nobel laureate Edward Moser. It’s perhaps a stretch, but the man who came up with the way our brains understand space and what’s in it and time, suggesting that thinking about future regulation is not a bad Something that Russell Neumann, professor of media technology at New York University’s Steinhardt School of Culture, Education and Human Development, and yet another author, claims may be misplaced. Neumann is impatient with the idea that AI should be halted in its tracks. For him, this is akin to the ideas put forward in Aldous Huxley’s book Brave New World, where science is muzzled so the status quo can be maintained. In his book, Evolutionary Intelligence: How Technology Will Make Us Smarter, Neumann makes the case that AI should be seen as a technology that will help us overcome many of the issues that beset us by increasing our ability to think.

Speaker G: Being careful and worrying is probably a good thing, but I think If we follow the advice of some doomsday scenario proponents that say, let’s stop everything, turn them off, unplug them, start all kinds of regulation of these new technologies, I think that’s ill-advised. And my primary argument is that these technologies can be used literally to make us smarter. And I start with the notion of this kind of idea of the Turing test that many of your listeners will be familiar with. This is the notion that if you’re interacting with an intelligent device and you’re persuaded that it is talking to you and you can’t tell the difference between this device and an actual human being, then it has demonstrated intelligence. And I think that’s got it absolutely backwards. How ironic that we want to design this this next level of intelligence on ourselves, our flawed evolutionarily designed selves, and that what we want to focus on is having these new technologies complement our thinking, complement the limited approach we’ve got to intelligence so that after machines made us stronger and telecommunications allowed us to communicate over long distances, these technologies will actually make us smarter.

Speaker B: Now, a lot of people have been saying this. A lot of people have been saying, actually, we need to work towards augmentation by machines. We need to work towards augmenting intelligence. But as you say, we don’t have the most brilliant track record in being intelligent, even though we claim that we do. We don’t have a particularly brilliant track record in being humane. Even though we claim that we’re humane. Why do you think that the machines will do that? Won’t they just be directed by us?

Speaker G: Well, I do think the notion of augmented intelligence is a wonderfully fresh way of thinking about it. Think for a minute about how the human cognitive system evolved in a period of scarcity when we were— for, you know, 99% of our existence on Earth, we were hunting and gathering. And resources were very scarce, and competition with other humans for the scarce resources led us to a way of thinking about our fellow humans and competition and aggression. So when we think that what we want to do is model the next level of intelligence on us, we’ve got it backwards because we’re living in a period of relative abundance, and I think we ought to think about reorienting our assumption that the guy next door is going to steal our lunch and talk about working together to share our lunch.

Speaker B: But that’s going to be the problem, isn’t it? Because we’re in this time when a lot of people are saying, though there’s going to be wholesale unemployment caused by all of these machines, what we’re going to have to do then is we’re going to have to have a universal basic income How are we going to pay for that? I know, we’ll get all of those billionaires to give us the money for the universal basic income. They’ve got to do a bit more sharing in all of this. It’s going to be a difficult thing to achieve, isn’t it?

Speaker G: It is, Peter. I don’t think you woke up with a particularly optimistic mood this morning. Let me respond to this classic issue of whether the next level of technology is going to leave us all unemployed. Each time that’s happened in the past, it’s true that the, the designing of buggy whips and the putting the horseshoes on the horses is no longer necessary. But in each case, a next level of technology has generated a next level of employment. So we can’t be sure, but I think we’ll find that especially if we’re designing these technologies to enhance our capacities. There’ll be a whole new level of jobs. That does mean we need to think very carefully about how universities and other forms of middle and higher education are training the next generation to deal with a very changing technical environment. But I think we should think of it— if you think about the first computer, it occupied an entire room with all kinds of air conditioning and tubes, and then it got to be a small size of a refrigerator, then it became something something in our lap as a laptop, then in our hand. And I think the next level of human-computer interaction is going to be as we wear our glasses and ultimately even maybe contact lenses, the capacity to intelligently enhance our perception and understanding of our environment. So that the notion that it’s going to be kind of a separate physical box sitting next to us, I think, is thinking of of it as we did the horseless carriage. Can you imagine a carriage without a horse? Let’s call them horseless carriages.

Speaker B: New York University’s Professor Russell Neumann and his ideas on how AI, rather than being a threat to us, will actually be a boon. Though, how in this world of automation will we make money if machines that can perform tasks better than us have sewn up what we used to call work. A point debated by UK Prime Minister Rishi Sunak and Elon Musk, and one that economist Guy Standing, co-founder of the Basic Income Earth Network and author of The Precariat: The New Dangerous Class, says he has the answer for: a basic income.

Speaker H: Well, the idea— I never use the term universal basic income, I use the term basic income. The idea is that everybody in society should receive a modest basic amount paid each month, paid individually, paid without conditions as an economic right. And I’ll come to the reasons for it, but what it would entail is that everybody would receive an amount that would depend on the affordability, which is another subject I’ve dealt with in my books. And the crucial thing is it would be unconditional and non-withdrawable. It would be your right as a person. Why I don’t use the term universal is you’d have to apply some sort of rule about people coming into the country to become eligible. And I would say that, that a person who’s a migrant to the country should be a legal migrant for at least 2 years before they become entitled. But that doesn’t mean they shouldn’t be helped. They should be helped by other mechanisms and, and so on. And it’s very important to realize that, that the idea of a basic income is to give everybody basic security. And that means you would have to give people with disabilities a supplement above the basic income because they have higher costs of living to meet, and the disability amount, if you like, would have to take account of that. So the intention is that everybody has an equal basic amount. So that’s the definition. I believe that that can be arrested and reversed. We need to build mechanisms where we control the technology, where the commons is built up in strength, then the ability to say this technology can be positive, but this technology can also be highly negative. And therefore we need ordinary people, you and me, as well as everybody listening, involved in controlling how those technologies are developed and are being used. At the moment, we don’t have that. We don’t have that. We have a combination of the elite working with the technocrats and the technologists. And we could see us drifting into worse and worse situation where even critical comments like I’m making now could be demonized and retribution through technological means could be such that people self-censor their thoughts, their activities. Huxley and Orwell and others painted this potential disgusting dystopia, and we’re much closer to it now than in 1984, and we’re much closer than Brave New World imagined.

Speaker B: Professor Standing. So, was Rishi Sunak’s summit a success? Did it cover the issues, or was it just a political paint job designed to promote the leader of a failing administration onto a world stage one last time, making a despairing attempt to assert the UK as a world power? Here’s Lord Timothy Clement-Jones. Chair of the All-Parliamentary Group on AI, on Sunak’s speech ahead of the summit.

Speaker I: The speech was an extraordinary speech because he identified all the risks, which are considerable, and he by and large reflected what the G7 said both back in last May and also the digital ministers in September. You know, the diagnosis was absolutely correct about the kinds of risk, you know, particularly disinformation and so on. But the actual prescription seems to me to be completely inadequate because he went out of his way to say it would be premature to regulate, whereas nearly every technologist that I know, nearly anybody in Parliament who’s really thought about this at all, indeed internationally, thinks that we have to, A, have a level of horizontal regulation across sectors, which may only deal with things like explainability and transparency and so on, but nevertheless needs to be there alongside the other sort of data-protecting aspects, aspects preventing bias and discrimination, that we need a level of regulation. It may not be as much as what the EU have, but nevertheless Every jurisdiction is going that way, including directives from the White House, for heaven’s sake. I mean, people like to think of the US as being an outlier, but actually in spirit they’re not. It’s just that they have a rather complicated way of making legislation on this.

Speaker B: Is there an agenda to this? Some people have been suggesting that the UK has seen what happened with the General Data Protection Act and thought, Oh, there’s actually a lot of money. There’s a potential industry to be made from regulation. And if we are the leader and we have all of these various institutes, hopefully not just in London, but in Manchester and other cities, that this will be good for us in terms of one, understanding, and two, making money out of it.

Speaker I: No, I don’t think we can do that as a relatively small market. I mean, we are fantastic researchers and developers. And where we could be particularly strong, and indeed we already are, is in terms of the way that we devise and set standards. Now, standards are different from regulation, as you know. Basically, you know, what we’re looking for in all of this— this is what legislation or regulation is all about at the end of the day— having sets of ethical standards which provide guardrails for risk assessment, an impact assessment, continuous monitoring, audit, the way that AI is tested to make sure that’s reliable, indeed even the way that regulators sandbox. So we’re very, very good at that standard setting, and indeed there’s a lot of work going on internationally to get the same converging standards. The OECD is doing a lot, the US are doing a lot on this, And I think I’m very optimistic that the people like the Turing and the British Standards Institute, you know, many others, I mean, even civil society groups like Ada Lovelace have got a very strong input into all of this. But that’s a very different issue from the regulatory aspect. I think at the end of the day, most developers and adopters, if we’re not careful, are going to say, if I’m going to exploit my AI system in Europe, guess what I have to do? I have to make sure that I conform to standards as a matter of obligation. This isn’t just a nice-to-have, you know, the regulator issues guidance and says, yeah, this is what you should do. I think they’re going to look at the AI Act in Europe and say, I have to do it. So I think it’s a bit fruitless for us to be saying that we’re going to be innovation-friendly and context-specific and all the sort language that’s used in our white paper, we’re going to have to come forward with a much greater degree of regulation, which doesn’t have to be as comprehensive as the AI Act, because I actually believe that there’s quite a lot already covered by legislation. But we have to look at some of those principles in terms of particularly, as I said, explainability, transparency, and accountability, where I think regulation needs to start to bite, and it doesn’t need to be complicated.

Speaker B: Lord Clement-Jones, chair of the All-Parliamentary Group on AI, with a widely held view on the AI Safety Summit. School report: good effort but should try harder. As Oxford University put it, it is encouraging to see the AI Safety Summit taking place. AI is a technology that is in some ways unlike any other, and we have seen dramatic progress in it over the past decade. Unfortunately, much of this technical progress has come along a branch of AI that makes it very difficult for us to understand or carefully steer what exactly the AI is doing, or even what the next version of a system will be capable of. As a consequence, the variety of concerns raised by AI across both AI safety and AI ethics is enormous, and the one thing we can be sure of is we do not even understand all the risks yet. Many of these challenges require not just technical understanding but also interdisciplinary expertise of a type that we have traditionally not trained people for. Bringing in leading politicians from the US, China, France, and Germany may have been a coup, but their participation was restrained, as the Financial Times pointed out. Standing alongside US President Joe Biden before he signed Monday’s executive order regulating artificial intelligence, the Vice President Kamala Harris spelt out her country’s intent to remain the world’s technological hegemon and write its own rules of the game. “Let us be clear: when it comes to AI, America is a global leader. It is American companies that lead the world in AI innovation. It is America that can catalyze global action and build global consensus in a way that no other country can,” Harris said pointedly. Then she flew off to the UK government’s Bletchley Park summit on AI safety. In truth, no one knows how machine learning will develop, whether it will become AI, whether it will become the fabled sentient artificial general intelligence. Though, if the summit did prove one thing in spades, a little like this program, for a technology that is meant to have such an impact across society, the much-trumpeted participation of the views of all of society were not canvassed. Something that those behind AI have paid lip service to for the last few years. I asked for opinions in my local pub from the manual workers smoking in the back. Their heads hung down in mute acceptance. You know what it’s going to do, they said. It’s going to take our jobs. You’ve been listening to Password, a monthly deep dive into the technology You Must Be More Interested In, presented and written by me, Peter Warren, and produced by Blue Buffering.

Speaker E: Thanks for listening and goodbye.

Speaker A: This program has been brought to you by Resonance FM. If you like what you heard, please support our work by making a donation at resonancefm.com/donate.