Please join me in inviting Professor Matt Jones.

Thank you very much. Thank you, Michelle, and thank you again to the Worshipable Company of IT for making this lecture and the series possible. And let me add my thanks to you in the room. Not only Rain, not only a tube strike, but then you had to find your way round a maze. So you must really want to come to this lecture, or you're really lost. Regardless of which of those apply to you, I promise I'll make your time worthwhile this evening. And then those of you at home, this lecture is also for you. Now we're going to do something quite unusual for a lecture or the start of a lecture. I'm not going to begin with a scholarly argument, nor am I going to give you some detailed algorithmic descriptions. Rather, we're going to listen for about 20 seconds to this marvellous singer, Bryn Tervil, another Welshman. That wasn't the reason I chose him. And he's singing, of course, at the King's Coronation. What I'd like you to do while you're listening is to pay particular attention to these four clerics, if you can, and watch their expressions and movements. So are you ready? Let's listen. They were being surrounded by that music. They were being moved, they were being perhaps reminded, they were feeling emotions. Music is, I think, one of the most powerful examples of an environment that you can sit within and be shaped. And what I'd like us to do tonight is to reorientate the way we look at artificial intelligence. What I'm hoping we can do together is to start thinking about AI as an environment. As an environment that we already and will increasingly think inside of. Some of you have been to many of the lectures, thank you. Some of you, this will be the first time you've come to one of these lectures. If this is your first time here, this is a great time to join the series. You are very welcome. At the start of the lecture series, we worried about AI being better on us, uh being faster, being smarter, beating at us everything. We worried about AI making us become more like it, that we would become assimilated and having to adapt ourselves to AI. And we also considered perhaps if it were just kind of domesticate us, it would take away our wildness. Now we need to put away those perspectives. That's not the way to think about AI. We also considered more, it seemed more jolly, more friendly ways you might think about AI. Humanoid, um uh embodied artificial intelligences, that, oh, you can even now buy one. You can rent one in the States from a company called NEO for $200 a month. This robot will come and help you do your chores in your house. So a more friendly perspective on AI. And of course, many of us have been told in our workplaces that AI will be our co-worker, will sit amongst us and work in our teams. Some of you might remember this little um AI. I brought it along, I think, to the third lecture. This is Mofflin. It's a Cassio, AI-infused pet. And this pet can give you companionship and comfort. And it makes a little noise. I was going to bring it again, but last time it distracted everyone and we we lost all of the point of the lecture. But you know, AI, artificial intelligence, is not like us. We need to ground this moment with that perspective. I've just come back from um Darvi in Mumbai. Uh, as Michelle said, we do lots of research work, working with um very diverse communities to understand the future of artificial intelligence. So here I am in Darvi, it's Asia's largest informal settlement. Each day, when I stepped out of my workplace there, I just couldn't be confronted by the embodied diverse intelligence around me that could weave entrepreneurship, could weave social and cultural events into very compressed and full spaces. You might have seen last week the Pope wrote an encyclica about AI, uh, roughly translated. The title of that encyclica is magnificent humanity. Now, I don't regardless of what you think about the Pope or about religion, that word, really important to remind us. All of you, we are magnificent humanity, very different from the artificial intelligences that we're seeing emerging. While I was in uh Mumbai, after the busy days, I would read uh uh The New Yorker, and in uh I think it's April the 27th edition, there was this amazing poem. And I want to share a little bit of that poem with you tonight because it really gave me a jolt and reminded me of how different we are to large language models and emerging AI. So I'm gonna read um two portions of this poem, uh, a bit of the top, first of all, and then a bit at the bottom. So last night, after months away from home, a lap wing piercing the still, dark, still with its warnings. Language, perhaps, began like this: a campfire, one person, to keep watch. You would have needed something to sound danger. A word like tiger, airstrike, run. And once the words entered, they were unstoppable, dragging dirt into the hallways of our brains. And then as we go down to the bottom, the ancestors of lap wings, they had feathers for a million years before ever using them to fly. Our tongues, monarchs in the cradles of our mouths, now scream, now sing. I didn't know that lap wings had feathers for a long time before they flew, but I did know that we, humanity, had tongues and vocal tracks long time before we created language and meaning. But our ability to communicate, our ability to have language, came through embodied long connections with each other, with the land, and with activity. Artificial intelligence is not like us. The origin of language for artificial intelligence is simply a probabilistic machine that churns through all of our previous utterances. Very, very different. So, this is where we got to. Um, your best relationship with AI is one where you definitely don't see it as a coworker or a friend or even someone down the pub. However, tempting that is. Let's all move away from it being a presence and consider it as a power. There are different types of power, aren't there? So if you look at fire and elephants, these are you could call them agentive powers. Uh you can sort of start them off, but they have their own animus. You might want to point them in the right direction, but they have agentive power. Another type of power is the material and instrumental powers in things like violins, in things like paint, or in things like clay. Now, material and instrumental powers, they don't just start on their own, do they? And if I put a lump of clay here, I'd be very surprised if it started to work by itself. We work with those materials, those instruments, and they also push back at us and shape us. Famously, that was seen, I think, by um some of the works of Michelangelo. So Michelangelo described how he created the marvelous statue of David. He said he looked at this big um chunk of marble and he worked with the cracks and the imperfections of that stone to release the angel that was David. He had to work with that material, understand that material. And of course, famously, when he created the Sistine Chapel, lying on his back, it wasn't really like us typing a prompt, was it, into a AI. Make me a Sistine Chapel fresco. No, he worked hard and struggled, had difficulty to create that wonderful masterpiece. A third type of power, which we will focus on or spend about half our time in this lecture thinking about, is the power of the environments around us. Now I wonder if any of you have noticed when you walk in like this city, do you walk faster than when you walk in the country? Probably, particularly when it's raining. Now, the city itself isn't giving you instructions to that, but the environment has shaped you to move in that way. Or I wonder if any of you have been to buffets and you've decided to pick up a large plate rather than a small plate. You might find that that environmental structure gives you the opportunity to eat more than you were expecting. And of course, this is a big topic, isn't it, right now, for governments around the world and indeed our own government to think about the power that the digital environment has through our everyday exposure to it. So, agenda power, instrumental power, and environmental power. Have a look at that table, and I wonder which one of these would you naturally reach towards when you're thinking about your use of AI at the moment? I think a lot of people might be thinking like this, and certainly a lot of the sort of sales discussions are around that sense that AI is most powerfully seen as an agent. That we will point it in a direction and it will go off and do stuff for us, thereby reducing the amount of work we have to do or making us more efficient. Of course, that has merit. But I want us to move down the table tonight and think about what we get if we take different relationships with AI. And we're going to start by thinking about in more detail AI as a material power, as an instrument. Hands up if you uh ever had the fortune of either learning the violin yourself or having someone in your family learn the violin. Okay, quite a few of us. Now, when, and I learnt it too, when someone picks up an violin to start with, uh, the noise is usually something like this. Not very pleasant. She didn't look very happy over, did she? That was AI generated, by the way. Now, a lot of people seem to think they can pick up an AI instrument and just start playing it. Well, if you do that, or if you have that relationship with AI, I think you're gonna end up with this what's called AI slop. Basically, the rubbish that you see churning out. It's similar to getting a violin and thinking, I can just play that violin. But if you practice a long time with violin, you can become like this. Now, this virtuoso, and if you were a virtuoso, the instrument he's playing is not simply about him making music. He's also being deeply affected by his engagement and interaction with that instrument, as research has shown. Our brains rewire, we hear differently, we think about the music differently. That instrument isn't a passive object. Here's another instrument. This is Skye Brown, she's a famous uh British Olympian skateboarder. Incredible. And what she's doing is using her instrument, which is a skateboard, to do something that Andy Clark calls extended cognition. The skateboard is becoming part of the way that she thinks, part of the way that she senses that environment and moves around it. Ooh, a bit more scary example here of an instrument. Oh, the scalpel. Scalpels in the hands of very experienced practitioners are things, they're instruments which are almost not separate to the hand that holds them. They allow the surgeon to get a sense much more than he could and she could from looking at the body or even just feeling with their hands. Now I told you I went to Mumbai ago often, and about a year ago, um, it's monsoon, uh, and I fell down a sewer. Uh I've been to India many times and just forgot that during monsoon, when there's lots of floods, you need to remember which part is the road and which part is the pavement. So I ended up having to go to a hospital in Pauai in Mumbai, uh, and after some rehabilitation, they gave me this instrument, um, a crutch. This is a simulation, this isn't real me. And of course, that crutch was very useful. It allowed me to get around, um, it allowed me to get back to my normal working style. But if I had leant too long onto that crutch, then what I would have found probably is that my natural ability to walk would have been thinned, eroded. My relationship with the instrument would have flipped. Instead of me being the prime agent, it would become the thing that was driving my interactions. Now we're going to try and experiment now in this room because you can see this thinning, not just physically, but mentally. I want you to really try to remember this phrase. Want to put all of your effort now into remembering that phrase. The river remembers the stone. Perhaps you could say it out loud with me, it will help. One, two, three. The river remembers the stone. And now, in your own heads, just go round and round remembering that. Now, I'm gonna put up another phrase, and I don't want you to remember it at all, because I'm gonna leave it on the screen. Okay, you need to put no effort into remembering that phrase. But I lied. Now let's see. Uh, if you can remember the first phrase, could you say it out loud now?

Remember, remember.

If you can remember the second phrase, let's see what happens. Say it out loud now. Ah, okay, okay. Fewer people remembered it, but you are a Gresham audience, so you only you only need to see it for a few seconds. What we've um experienced there is what Betsy Sparrow and colleagues called It's the Google effect.

When we know the internet has the answers, our brains let the information go, making us reliant on our devices.

By the way, this is an example of AI slop, which I created for you instead of just saying it. That was a demonstration. So, an instrument, as we've seen, can do some very positive things. It can alter you, like the plasticity of the brain in the violinist case, or it can extend you, like uh Sky Brown's use of the skateboard. You can become attuned to something like we saw through the scalpel, or it could thin you, it could reduce your abilities. Nicholas Carr uh quite a while ago wrote a series of books giving us warnings about how general automation could lead to reducing our ability to critically engage with the world. It's a great book. But he was talking at a time when there weren't these emerging AI systems. AI systems do far more now than just remembering things for us. They're not just Google. They can draft things for you, they can analyze things for you, they can synthesize things, they can negotiate. So we've got to think very carefully about the environment that we're creating that you think inside, and whether that is extending you or diminishing you. I want to give you a few examples now of ways in which you can have an interaction with AI. Uh, in the first case here, where you're not being extended. You're being, or in this case, I'm being diminished. So I've asked Claude, Claude is one of the large language models, this prompt, I want to write a poem about the sea that captures loneliness and cold. And I hit enter, and as we've all experienced, if you've used any of these systems, up comes a poem in front of your eyes. The sea does not remember you. It turns its grey shoulders to the shore and breathes out cold and breathes out cold, a tide that doesn't keep the score. Or it even makes us a little picture I didn't ask for. Now, my point isn't whether that was a good poem or a bad one. I think it's a bad one. Um, it's that I have treated the large language model like a vending machine, like a slot machine. Uh, I have not treated it like an instrument. So let's try to do uh another attempt where I at least try to push the AI to be more like an instrument. So I start again. I want to write a poem by the C that captures loneliness and cold, same as before. Before we start, let me ask three questions that will help me say something I couldn't say without your help. So I've tried to be more engaged. I press enter and you'll see Claude starts working. This time, instead of producing the poem straight away, it gives me choices, which I you can see the cursor there moving. I'm making some selections. And then when I've made all three selections to the questions, there we go. Oh, it also asks me a bit more. It wants to push me. It's like that material pushing back. And then it produces a poem as you'll see. Here we go. Hopefully. I've stood at the edge of a sea I invented to feel small in the right way. The water doesn't turn, the horizon holds its grey like a rule. No one wrote for me. Again, the point isn't whether that's better or worse than the first one. The point is I was a little bit more treating AI in an instrumental fashion. Let's move from text to a more visual example. Our group in the computational foundry at Swansea have been thinking about how you can go from sketches to images. So what you can see here on the left hand side, there's a sketch. You hit enter, it's like a slot machine, bang, there is your finished version. So the question we brought to our research was could we do something to help the sketcher before generation? And can we do something after the sketch has been after the final image has been produced? And we developed two prototypes. I'm going to quickly show you what they do. In the first one, this is a system that helps you before you generate the final system. I've drawn a cat. Do you like my cat? It's not bad, is it? Then AI adds something relevant or salient. That's what we ask it to do. We say, look at my drawing, add something relevant. And it's added a ball of wool. Then it's my turn, and I've added a little house. What do you think the AI is going to add now? Let's have a look. That's a little mouse hanging from a ceiling. Then, so I've had back and forth with the system, right? I haven't just used it like a slot machine, but now I wanted to generate the output, and there's the output. So that's helping the sketcher engage with the instrument before creating the final output. Back sketch is working with a finished image. So I've drawn a mouse. As you can see, my ability to draw mice is not great. I think my cats were better. Then we ask the system to decompose the final polished version into a series of abstractions which are less and less detailed. We were inspired by this by Picasso. Picasso is a series of artworks where he takes an object, in this case the Toro, the bull, and he decomposes it into abstractions. We can then choose one of these abstractions and we can start editing it. And we can have lots of back and forth until we produce the final output. So we've seen agentive power, we've seen material that we need to work with, and I would say that if you are thinking about using AI, at least move down from that kind of agentive view. Think of it as an instrument that you are going to learn to play and engage with and connect with. An environment, though, stays with us, surrounds us, and shapes us. Now, in order for us to think a bit more about how digital environments really do have impact on our lives, I want to take us back to a really important bit of history in computer science, and in particular in my field, which is interactive systems. There was a debate in 1992 between Nicholas Negraponte, you can see a representation of him there, who is the one of the founders of the MIT Media Labs. Hugely influential lab based in the States, of course, but affected lots of the ways in which we think about computing today. And he was debating with Mark Weiser, who is based in another very important lab in Xerox Park in California. And they came together in front of about 700 people to discuss what the right way of thinking about future digital environments should be. Negro Ponte's view was: in the future, remember he was talking in 1992, what we need are personalized butlers, assistants that will unobtrusively, but certainly with a presence, appear and help us through our everyday tasks and our work. Weiser was completely against this. He said, no, digital technology should disappear into the woodwork. He said that the smart home of the future should be like the smart home of 200 years ago, where you had a book in every room, which brought knowledge and brought information. Then you went, and you, the human, went and engaged with that. So a way of thinking about these two worlds is in Negroponte's world, we're in the world of Downton. So we are the Aristos, and our digital computers are behind the base door, popping up every now and then, probably gossiping about us, telling the, you know, probably saying how useless we all are. That's the world of Negroponte. The environment that Visor envisaged is perhaps best thought of like a Zen garden. Everything carefully designed and considered, but there to allow you, with your agency, to reflect and travel through it. Let's see a scenario to perhaps make the difference between these two environs a bit clearer. I should warn you there are two bits of AI slop coming up again. The scenario is this I want you to imagine that you are asleep. If you are asleep right now, please wake up. You're asleep and your AI butler has looked in your calendar and has found that you have a slot in the morning when you can do a run. But it also sees that your running shoes are in need of repair. So overnight it has ordered some new shoes, it's downloaded the perfect route for you to run on. So off you go. Everything prepared. Ooh, a cup of coffee as well, and off you go. AI slop one. In the visor world, it's much calmer and less intrusive. So in the visor environment, what would happen is you're lying in bed, and then suddenly you notice that the temperature in the room has been reduced, helping you to stir naturally and freshly, and waking yourself with the thought, I feel good, I'm going for a run. And then instead of providing you with the exact route, the environment would ambiently show you the way. And off you go. Now both of the visions are interesting, right? Both of the environments are adventurous. They're very optimistic, they're very utopian. But what neither Negroponte nor Visor could see, and remember they were dispating in 1992, was our AI environment is actually a sort of combination of both of their worlds. AI is personalized and it can anticipate. But it's also almost everywhere. Even if you don't think you're using AI, in many of the tools that you now use, it will be there. And even if you're not using any digital tools, a lot of what you're doing, even sitting in this lecture, is going to be used to train and develop models because these lectures are recorded, and no doubt an AI will be listening to them and transcribing them. Now, I don't know about you, but when I think about environments and how to design them and how AI can be better designed, a good place to start is thinking about cities because most of the world's population, and increasingly so, live in cities. Good cities open up avenues for you. They provide you with possibilities, they extend you. There are places to rest, places to divert. Tonight you had to find your way around a whole new route to come to the Gresham College, and you should be proud of that. Poor cities, on the other hand, do the opposite. They sort of limit your destination, and perhaps even your destiny, taking away your ability to engage with that environment in a way that suits your individual needs. Now we started this lecture with music. And we started with music because I wanted to show you that an environment can shape emotions. I'm going to do something a little bit more ambitious now. You're going to listen to another piece of music, and what we'll see perhaps is that this sort of environment, it isn't just mystery and poetry that you're feeling. What's actually happening is this sort of environment is helping you anticipate, it's rewarding you, and it's giving your brain things to learn from. So we've got about 20 seconds to listen to this amazing piece of music by Barbara. Incredible. I wonder if any of you felt um what I call the chills, a kind of freeze-ong. Not right at the final bit, but also during the crescendo that's building. And certainly if you're in them, if you're in the concert hall and are listening to that, you probably will feel it in a profound way. Now what's happening there? This environment, music, is surrounding us, it's just surrounded us. Something particular is happening in our brains. And Valwa Salimbar and her colleagues have explained it. So deep inside your brain, you've got a reward center. And it has various parts, the chordate and the accumbens nucleus. When you were listening to that music, dopamine, which is the reward, was firing during the anticipation, not just at the outcome. And that, remember, the reward structure in our brain shapes what we value, shapes what we want to pay attention to, and over time shapes what we learn. Let's um have a quick contrast. In our day-to-day exposure in these AI environments, there's very little anticipation. There's very little of our brain reaching towards that environment. It's much more fluent and reassuring. So let's see what happens here. I don't know what's wrong with me. Um, all my examples. The first was about being on the seashore, lonely and cold, and this one's about losing um leaving my job. I'm not leaving my job, I love this job, I love all the jobs that I've got. But here I've asked, I've been thinking about leaving my job. I feel like I'm not growing anymore. Hit enter. Little delay. That's a really heavy feeling to carry. When you reach a point where the day you start feeling like a loop instead of regression, it takes a lot of mental engineering just to show up. So, no anticipation, just reassuring, not troubling, no engagement. The environments we are building, or we could be building and using, could be leading us to work in that way. Uh, Robert Bork and Elizabeth Bork call this uh problem of not having desirable difficulty in our lives. The concept of desirable difficulty is the very notion of reaching and struggling to find something out is as important as invaluable as coming to the answer. Now, with they they did their work before large language models, as you can see, back in 1994, but their work is really relevant now. And we can see some of the impacts of this thinned critical thinking in very profound ways. I want to share just two of those from research papers that were published this year. So, this is the first one, and this is the most shocking one. Darren Akamolu is a Nobel Prize-winning economist, and he's been modeling with his colleagues what happens if we stop doing the critical thinking, if we live in environments that are just easy for us. And what he says is there in the world there are two types of knowledge. One is called general knowledge, that's a knowledge all of us in society uses and has created. It's a stock there that we can draw on. And then there's context-specific knowledge, that's stuff that you need to find out to get a task done right now. And what Darren and his colleagues are saying is that in the past, before AI, when you need to find something out, you work hard in that environment. And your hard work leaks out, maybe because you talk to a colleague about what you're struggling with, or explain it to somebody else, even down the pub, and that builds capacity in the general knowledge pool that the rest of us can draw from. If we build AI environments where we just say, I need to solve this problem, just do it for me, and it does it for us. There is no leaking out into the general knowledge pool. And what this paper shows, and by the way, it hasn't been fully peer-reviewed yet, but I would definitely read it. What it shows is that if this continues to happen, you get what has been called knowledge collapse. And that knowledge collapse will happen even if AI is doing all the work for us, because the capacity in the general knowledge bank is being eroded. Now, that might be a bit too kind of catastrophic and society level for us, so let's think about this other piece of research again published last year in April, I think, 2025. And it was a study done by Microsoft Research and by Carnegie Mellon University. They talked to 319 people who were using AI in their day-to-day tasks. They wanted to understand how that use was affecting their critical thinking. The paper's great, read it all, but there are two messages. Number one, if you become overconfident in the system, the AI, and underconfident in your own ability, then you're much more likely to accept the answers AI gives you, even when they're wrong. So the best way forward, this paper says, is don't outsource your judgment. Steve Wasnick, uh one of the founders of Apple, uh, used this phrase. He said, use your actual intelligence, not AI. Integrate, synthesize, use these tools for sure. But if you outsource too much, then something bad's going to happen. Now, as we come to the last section of the lecture, I want to bring us back to thinking about the environment. And in my world, I live in Wales, uh, and I used to live in New Zealand. Rivers were really important. Rivers do two things, don't they? At least two things. One is that they carry away stuff, they carry away debris, branches, things that have dropped in from the side of the banks. But they also shape. They shape the landscape, they shape the banks around them. Let's take both of those things that rivers do and then relate them to AI. First, something that is very well known, I'm sure many of you have come across Tristan Harris. He used to be a Google chief efficist. Uh, and then he had a crisis of conscious, and he wrote and he went to Congress, and this is long before the current worries about social media, and he said, This river is taking away our attention. Taking away our attention, just like the river taking away stuff around the landscape. Tonight we can go much further, I think, than the worries that Tristan Harris was giving. Because if your attention is being captured, there's an answer. You could put down your phone. If we're creating, though, an environment in which we will think, then we have to consider what that is doing. What is that doing to the banks of our critical thinking and our understanding? So, as we've seen, if we build environments, and it's really important to remind yourselves, AI not as an agent, and perhaps not even as a tool or as an instrument, an environment in which you will think. If that environment makes life too easy for you, if it doesn't have that desirable difficulty, the banks of your abilities is going to be made thinner. And as we've seen, that can have catastrophic outcomes, not just for yourself, but for society. The converse then is to embrace difficulty. This doesn't mean throw away AI. That would be nuts. And I get fired as the worshipable company of IT professor. Hang on. And I and I truly believe that AI does have value. Lots of problems with it, but it used properly, and if we see it as an environment we think within, just like this room tonight, right? This room tonight has been an environment where we have created together a space to think. I've chosen slides, I've chosen music, but you, everyone is still awake, I think. You have been listening and thinking and engaging. We've created an environment, I hope, that is strengthening our capacities. So if AI is becoming a mental environment, we need to think of it more like music, like media, like culture. And our focus as designers and computer scientists, and then people like all of us, as users of this, as an environment, and we must make sure. Sure, that environment preserves the conditions for thinking. Because it's an environment, we can, of course, think about regulation. And in uh food, we put things into our body, that really shapes us, doesn't it? As you can see, I eat too much and it's really shaped me. And so we have regulation. Or with media, we consume lots of media, and we've had regulation for a long time. And of course, planning for physical spaces. Now I'm not going to go into that because our professors of business and professors of law at Gresham can talk more eloquently about it. I want to end by thinking about what people like me as a computer scientist could do, and also people like you, if you're not a computer scientist, how you can live in that environment and have a better relationship with it. So, first of all, algorithmic innovations. That paper on knowledge collapse, one of the really fascinating sections for me was the argument that if you make AI systems less than perfect, then you are going to naturally engage the user to get them to critically think and work with the system. So that's important. It's very hard for developers to think like that, isn't it? Because the race is on, you know. Anthropic today announced that it was going to do a trillion dollar initial public offering. And it will be saying, no doubt, in that uh prospectus, that it's producing a better, better, more perfect model. We need to disrupt that a bit. We could also build systems that are trained to align themselves to the value of promoting critical thinking. We talked in, I think, last lecture. That's one of the things you can do in systems. You can say, these are, if you like, the constitution that I have for artificial intelligence. I wanted to do this sort of thing. Not to thin me, but to strengthen my critical thinking. But if you came to the last lecture, we saw how difficult this is. I'll remind you of the example. So this is Tetris, very simple game. An AI developer said, I would like AI, you to um learn how to survive as long as possible when you're playing this game. And what it did was just to pause the game. So it answered the question, but it wasn't aligned to our values. Now, if it's hard to align to a simple thing like Tetris, how do you get it to align to making us more critical in our thinking? Secondly, interaction design, and I'll quickly uh give you some examples there. Instead of building systems that just give you the answer, think about how you can give the user pause for thought. So here is an AI system, and the task that these researchers gave this, uh gave their participants was have a look at that meal and then replace one of the ingredients such that the meal is a low-carb recipe. And in this condition, you can see the person using the system was given the AI's answer right in front of them. And the most obvious thing, and participants did do this, they used that answer. And the researchers then compared that with two other conditions. One where you had to click and drop down the answer, so you had to make a choice to see what the AI was telling you, and the other they just delayed the output. These are called cognitive forcing functions, and just simple nudges like that can get us to sit in the environment more effectively. And finally, down to you and me and Wednesday morning tomorrow. How can we have a better future with AI? How can we live in this environment? Well, first of all, carry on doing what you've been doing tonight, and some of you have come to many of the lectures. Really put some effort into understanding what's going on with these AI systems. But ensure that you, when you're using them, are the one who is controlling and thinking and being surrounded by that environment. Reach for that desirable difficulty. Think back to the beginning of this lecture, those four clerics, as we listened to Bryn Telvrill, that wonderful music. They were surrounded by an environment. They were in an environment, it wasn't an unpleasant environment, they weren't being forced to doing something, but they were being shaped. Their memories provoked, their emotions distilled. If we can see AI as an environment to think within, and we ask developers to see it in that way and create pleasant, valuable, human-centered environments, if we can ask regulators to step in like they have done in many other environments, then perhaps the future will be very bright, and indeed our souls will sing.

Thank you.