Is AI the end of creativity and cognition?

Cognitive skills and creativity in the AI age

0:00ABC Listen. Podcasts, radio, news, music and more.

0:08The name's Bond. Alan Bond. Billionaire businessman, America's Cup winner and, oh yeah, con man. Bond cooked the books of Bond Corporation. He wanted everybody to know that he was in charge, they better look out. It was the biggest fraud in Australian history. I'm Mark Humphreys, and I'm here to tell the full Alan Bond story. Search ABC Rewind on ABC Listen or wherever you get your podcasts. An AI researcher warning the technology could, quote, kill us all within the next 10 years.

0:42An artificial intelligence agent has infiltrated an Australian government website. I'm not going to stifle growth of something that will be bigger than the Industrial Revolution. Every time that we make a new episode of All in the Mind on artificial intelligence, the clips that we use in the introduction, the things you just heard, they get a bit scarier. In the past weeks and months, we've heard about AI agents breaking free of internal testing and hacking companies, even governments, including Australian government websites linked to Medicare.

1:15Then there is the existential concerns, whether AI could cause humanity's extinction, as the workers at these companies that are developing AI keep warning. This has got some people reaching for the AI off switch, or at least slow down switch. And cognitive neuroscientist Ruben Rodeau and English professor Sophie G have been thinking a lot about the moment we are finding ourselves in. Sophie from the perspective of the humanities, Ruben from the science angle. They're speaking at a panel at the Curious Festival at the Sydney Opera House this week,

1:49but we've nabbed them first. So today, we're talking about clinging on to your cognitive skills and creativity in the AI age. From ABC Radio National, I am Sana Kadar. This is All in the Mind.

2:11Okay, I'll get you to start by actually introducing yourselves. So your name, what you do, where, Ruben, I'll get you to go first. So my name is Ruben Rodeau. I'm a cognitive neuroscientist at the University of Sydney. Perfect. And yourself? My name's Sophie G. I'm an English professor at Princeton University in the US. And at the University of Sydney, I lead a small team called the Elios Collaboration, lifting up our capacities to be more citizenly in the world. Okay. Very lofty. I like that. Yeah. Love a big idea. Yes. Okay. So talking of AI, a big idea, there has been so many fast-moving developments in the last

2:46couple of weeks between hugging face, between greater calls for regulation, between very fresh news that the Australian government websites were hacked. And then also, you know, having top researchers say they fear we're going to be dead within the decade because the AI will kill us all if we don't slow down. I'm curious to know, you know, how you guys are feeling about these developments at this point in time, because I'm not feeling so great. Sophie, I'll start with you. I'm certainly not feeling great. I think like everybody, it's been an overwhelming week of bad news for AI, with AI, from AI.

3:18And the question is how to respond to it, basically. I think that, you know, when AI first emerged as a technology, it seemed overwhelming, I think in part because no one could quite grapple with what it was. Now it seems actively dangerous. The fact that it's so under-regulated has become really evident. And I think all of us in our different fields and ways of working have to say, how are we going to stop this tide? How are we going to intervene in a purposeful way so that AI, in effect, is working for us,

3:53not the other way around? A tide is like the right metaphor. It feels like this gigantic ocean wave that we're not anywhere near the crest of, but it's already really overpowering, coming for us. Yes. Ruben, what about you? Yeah, I have to agree with Sophie. I think it's been a tough week in the news, and it's hard not to feel some concern about some of the developments in AI. And it seems as though we're in a bit of an arms race, or the companies that are behind these big models are in an arms race between each other that seems like it's heading in a direction that we're not too sure about.

4:25And I do think that we need to sort of see a slowdown in order to allow the regulation and the guardrails to catch up so that we can ultimately get this technology to do the things that we've wanted to do, and not the things that I don't think any of us really wanted to do in the long run.

How experts use AI in their respective fields

4:40Yeah. I'm curious how you guys are using AI in your work, in your respective fields, and what's meant for your work. Ruben, I know it's important in some of the work you're doing in your lab. Tell me about that. As neuroscientists, we have the unique opportunity to use it in this qualitatively, completely different way. And that's not as a tool, but as a subject of study. And this is because AI is fundamentally inspired by our real brains. And so what we can do as neuroscientists, especially computational neuroscientists, is we can train

5:16up these artificial neural networks, which are the components of artificial systems that we're familiar with, and then we can study them in order to try to learn things about our own brains. So what we do in my lab is we take, because I do a lot of research on sensory processing, we can model, using artificial intelligence systems, processes, sensory processes within the human brain that we know a little bit about from neurophysiology and other sorts of research. Which kind of processes?

5:46Visual processing. So you might model the visual system by creating a neural network that roughly corresponds to the different areas in the brain, in the real brain. And then you might train it by asking to do a task that we know that humans can do. So for instance, like image recognition, you might show it a whole bunch of images that you know what those images are, but all the model gets are the images, and it has to try to learn what they are through this process of sort of training. What will happen is it begins to form these solutions naturally in order to perform the

6:22task, these sort of little neural mechanisms that allow it to perform whatever tasks it is that you've gotten, that you're training it to do in this example, image recognition. The advantage of doing this, what we call in silico, but in an artificial neural network, is that you have unprecedented access to the neural network. Even the most invasive techniques that you can use on non-human animals don't compare to the level of control and interrogation that you can have in these artificial neural networks.

6:55And so once you know that there's some sort of solution that's emerged within that network to perform this task, then you can really deeply understand it in a way that's just not really possible in a biological brain. It's really fascinating hearing you talk, Ruben, because hearing someone speak so specifically and with so much expertise about what's obviously a very useful application of AI caused me to think about, well, what is the problem with AI? That sounds good. Yeah, no, that's a very hopeful kind of application of AI.

7:28Yes, very helpful. And as you were speaking, what I think I realised, and correct me if I'm wrong about this, it's very useful because you are trying to understand more completely and measure a physiological operation, what I would call physiological, I guess it's also neurological, within the brain that has kind of manifestations in the rest of the human body and is responding to sensory information coming in from outside. Have I got that right? So the pieces that you're moving around, you know, to put it quite simply from my point of view

8:01as a humanist, kind of have answers and you can verify whether those answers are correct or not. What I think is the core business of the humanities, which makes it both a really tricky field to be working in, but also a massive asset at this particular moment in human history, is that we in the humanities are most interested in the uncertain, in answers that kind of can't be wrapped up, that need to stay ambiguous. And I know that scientists also do this, so I actually don't in any way want to simplify what

8:31Ruben is saying. So, you know, put that in context. We get it. Yeah. But, you know, what I like to say, you know, of the humanities is that when something, the more ambiguous it is, the more complexity it has, the more incomplete the information is, and the more uncertain the outcome, the more attractive a problem becomes to someone working in any of the fields of the humanities. So, to answer the question, how do I use AI? I mean, how I personally use AI is in a very, very limited, simple way. When I'm making my classic books

9:04podcast, which is about the history of literature, and I need to know something quickly, I can ask Google, basically, you know, what year was, well, this morning we were talking about a Shakespeare play, what year was it first performed and what else was performed at around about the same time. And AI is able to, you know, provide a great deal more information on that than was previously possible. Do you feel, or do you see a place for AI and the humanities beyond that kind of fact checking or admin at some point? Yeah. So, I don't do this, but my colleagues,

9:35both at Princeton and Sydney, who are doing the most interesting work in the humanities with AI, they're using it speculatively. They're using it to process huge amounts of data in order to generate information that previously has appeared to be beyond the realm of human reach. So, let me give you an example. So, it's particularly used by people who are working with ancient documents, documents or objects, human-made objects that are extremely fragmentary. So, fragments of language tablets from the Assyrian world, for example, or from the Egyptian world,

10:12from any ancient or archaic world, basically, where we've only got fragments and traces of what was obviously a very complex and complete body of knowledge. AI is very good at using those fragments and traces and constructing a kind of speculative replacement for what is missing. So, what AI is great at in the humanities is what humans are sort of least good at, which is very quickly processing vast amounts of data and showing broad trends, broad possibilities, broad directions that you could

10:44take a piece of research in. And then humans come in, of course, because, you know, as we know, even from asking AI, you know, what you should feed your dog for lunch, AI makes quite a lot of mistakes. So, humans have to come in to do the differentiation work, the discrimination work. But, you know, the possibilities of what it would mean to be able to reconstruct records that we had imagined were completely missing is, that's an exciting possibility. That's fascinating. So, basically, filling in the gaps of sentences of ancient texts kind of thing? Yeah, or more than just sentences, filling in whole texts. Wow. Yeah. I want to talk about cognitive surrender, cognitive

Cognitive offloading and the risks of atrophy

11:22offloading, as well as creativity. Let's start with the cognitive offloading. So, there's a lot of talk and a lot of concern about how, by outsourcing our thinking to AI, we're going to atrophy our own ability to think. How credible is that concern? I mean, there's been research and reports to show that this is happening. So, Ruben will know a lot more than I do about this. I only know the sort of scary big headlines, but the headlines are scary. I mean, the main study that gets cited in the circles I move in was done at MIT. It studied students using various forms of what they call mediation between

11:57their own cognitive processes and the research questions that they were trying to answer. So, mediation means either some kind of artificial assistance, whether it's like, you know, like Google or a large language model kind of writing your answer for you. And what the study showed was that young people, and I think everyone, is cognitive sort of capacities instantly and catastrophically dropped off. And actually, any mediation causes our brains to work less, less well,

12:33and much more quickly than we think it will happen. And just very, very briefly to say sort of, why does that matter? So, one of the things that students, I think, feel about, you know, ChatGPT and its brethren is that it can help them kind of bang out an essay or bang something out quickly when they basically kind of know the material. And so, you know, why not use the thing that will get them the document quickly? And the answer that this and many, many other studies have shown is that bad work, the process of writing badly, thinking badly, stumbling your way, often

13:06with many errors through a thought process, that is the business of training your mind to think critically. And if you bypass that, then you've got a really major problem. And Ruben will be able to speak more about what the problems are. But in my field, they are many and they're catastrophic. And to kind of, you know, spoiler alert, it has actually led me to become a real sceptic of any kind of AR use in, you know, training, particularly young people to in analytical and critical thinking.

13:36Ruben, why don't you jump in there? Okay. So, I'd like to begin just by defining what cognitive offloading is for people. So, cognitive offloading is performing a physical task in order to reduce the load associated with a cognitive task, right? And so, a classic example would be, say, for instance, I needed to go to the shops to buy some ingredients for dinner. I needed 10 items. I could choose to try to remember all those 10 items on the way to the shops, or I could write them down on a list, carry that to the shops, and then look it up and use that to guide the... Okay. And that counts as cognitive offloading.

14:10That's cognitive offloading. So, that's, you know, in this case, the list is the physical performance, the act of holding them on my mind, in my memory, is the cognitive task. Cognitive offloading is actually a very adaptive behavior. We use it constantly throughout the day, whether it's using GPS to get to navigate to your location, putting your schedule in your calendar so that, you know, when your meetings are, or even looking something up in Google that maybe you already knew, but you can't really be bothered trying to remember, right? You just need it quickly. So,

14:42we use cognitive offloading all the time. There's a lot of research on cognitive offloading. There's a long history of it. Most of it's been done in the memory domain, but the most well-known costs and benefits are that you actually get a benefit when you have access to the tool, but then you have a cost when you lose access to the tool. That's sort of what's showing up in the research with cognitive offloading, isn't it? And students. Yeah, but it's important to be specific about that. So, when I say there's a cost and a benefit, I mean, for example, in the shopping example I gave you, say, for instance, you asked both

15:16Sophie and I to go to the shops and buy the exact same 10 items, but I'm allowed to use a list, right? And so, we go to the shops and we come back and you have a look at what we've bought and I've gotten all 10 items because I used a list and Sophie had to try to remember all the 10 items and maybe she forgets one of them, you know, we're fallible. She's a middle-edged English professor after all. So, maybe she gets 90% correct, okay? So, I've had a benefit there. However, if on the way to the shops, I have lost that list, then when I got to the shops, I just had to figure it out as best I could when we get back. Maybe I've only remembered

15:47seven of the items now to Sophie's nine. So, she's actually done better having not used the list in the first place. So, this is a really reliable effect of or phenomenon associated with cognitive offloading in terms of benefits and costs. But that doesn't mean that when I got back, if you ran a general memory test on both of us, that I would necessarily perform worse having used the task. These effects generally apply to specific knowledge, not general cognitive ability. So,

16:19I don't think the picture's super clear in terms of the effect of cognitive offloading on general cognition. There really is sort of a point at which cognitive offloading is really beneficial, but you can use it both too much and potentially too little. And so, if you think of sort of cognitive offloading on a dimension of not using it at all, all the way up to using it every single time that you do that cognitive task. Not using it also abstinence. Say, for instance,

16:52I was going to the shops. If I never used a list and I tried to remember every single item every single time, you know, that's not very adaptive. I'm going to forget things over time. Your partner would be very annoyed. Exactly. Equally, if you use it every single time, so even when I only had one item to buy, if I wrote it down on a list, even though it's only one item, that also wouldn't be very adaptive. So, there's this point somewhere in between this sort of using it all the time and overusing it that is actually quite adaptive. And it's when you supplement your existing cognition that it

17:25becomes very adaptive. So, I think the key is to supplement and don't substitute cognition with these tools. The other thing that occurs to me also is, Ruben, the examples you were giving of cognitive offloading, they involve memory, but the newer forms of AI that we're talking about, the LLMs, et cetera, they're replacing generation. Is that, like, that feels like a key difference. Yeah, it really is. So, I think that's a key caveat to a lot of this research. It has been almost exclusively performed on memory. There have been studies that look, recent studies,

18:00like the one that Sophie mentioned, looking at sort of cognitive offloading in the case of generation creativity. But the vast amount of our insight into this process has come from memory. And so, it's unclear at this stage whether or not what we have learned there is transferable to generative processes like composition and inference and generation. So, yeah, I think we need more time to really understand that and more research. I mean, these models are very new. People are researching it actively now, but, yeah, we still don't have a clear picture. And so, I think

18:33more research is needed in that space. Always. Absolutely.

18:39You are listening to All in the Mind from ABC Radio National. I'm Sanaa Kadar. Today, in the wake of new concerning headlines about AI seemingly every day, we are talking creativity, cognition, and staying human in the age of AI. I'm talking to neuroscientist Ruben Rodeau and humanities professor Sophie Gee, both from the University of Sydney, as well as Princeton University, for Sophie.

Embracing discomfort versus seeking convenience

19:07A word we haven't used yet in this conversation, which is sort of interesting, is discomfort. And something that we, you know, pervasively seek to avoid as humans, I think it's fair to say, and I'm sure other than human species as well, is discomfort. You know, we spot it coming and we seek to avoid it. And then we have this whole other body of human activity in which we purposefully, or semi-purposefully, expose ourselves to various forms of discomfort, whether that's emotional

19:38discomfort, which can obviously range from being in, you know, physical presence with a person who themselves are having very difficult, challenging emotions, being in presence with someone who, you know, is having physically very difficult things happen to them, you know, a person in hospital, the experience of reading a painful book, telling a painful story, looking at art, you know, the list goes on. My teenage son right now is somewhere in the Australian bush by himself in a tent for a week,

20:10you know, that's going to be... That sounds painful. Sounds really uncomfortable. And so, you know, one of the things I think that I find very alarming about the emergence of AI is that part of what makes it so seductive, part of what makes it so alluring is that it does rub away a great deal of the kind of friction of our capacity to tolerate discomfort, even in the case of AI that is supposedly sort of replicating discomfort of certain kinds,

20:40like a chatbot, for example, where you could talk through difficult feelings. But it's so sycophantic, it'll never be too discomforting. It's incredibly sycophantic. It's also not embodied and it's not, it doesn't have feeling. And, you know, one of the things that's both subtle, but utterly fascinating about human made objects, like writing, art, music, even just, you know, something more mundane, like a, you know, a basket or something like that, is that it has been made by a feeling human body. And the process of

21:15making somehow, and this is the miracle, imparts to that object, the sum of the emotion and the embodied experience that the person who made it has had. And those, those experiences then get passed to the person who is looking at the object or reading the object. It's, you know, it's a truly extraordinary thing about, about the way that humans make things. If we want to stay human and want is kind of, okay, the imperative is to retain our capacity to tolerate discomfort

21:49and actually in a way to sort of learn that through discomfort of many different kinds, including boredom, we can learn to use our human capacities in a deeper, richer, more meaningful, and ultimately, I mean, I hate to use this word, but sort of more virtuous way. And, you know. So you have a moralistic argument. Well, moralistic or about the idea that, you know, what are we supposed to be doing as humans,

22:20if not thinking and feeling and being in community with others in a generous and loving way? I mean, to me, that is our entire purpose. And, you know, there's a lot of talk about the decline of the humanities and, you know, how people aren't doing arts degrees anymore and that arts degrees are useless and so on and so on. This isn't the moment to sort of get on my soapbox about why that's not true. But, but what I will say is that if only we could wish away our need to be in sustained contact

22:50with difficult experiences. I mean, the history of human society is the history of unexpected, unwanted encounters with intense turbulence of different kinds, ranging from plague and war, catastrophic, you know, loss, you know, back in the, in the kind of dark ages, through now to some of the sort of climate catastrophes and geopolitical disasters occasioned by various acts of hostility and aggression. I mean, you know, there's, there's a lot of them, but humans are continually

23:22encountering discomfort and it's unexpected and unwanted. And really, if you look at human history and you really go broad, our survival, thriving and capacity to stay in meaningful community with one another depends on upskilling in collectively handling discomfort and disaster through our bodies and feelings being attuned to what's really happening. That feels a little bit like really the time we're in now. Intense encounter with turbulent times is right this moment. Yeah.

23:55As the AI. Like this week. Yeah. Tidal wave comes for us. Yeah. If I could just touch on the sort of difference between offloading memory and creativity just for a moment, it would be to say that I think there's a fundamental difference there possibly, not just sort of scientifically, but in the way that we value those two things. So I think that we are probably pretty comfortable with not being able to navigate very well now that we're all using GPS. That doesn't seem to be a skill that we sort of value that much. The same with sort of remembering your schedule for the day. I think you're probably pretty

24:26happy with, you know, not being able to necessarily remember every single meeting you have, and you're happy to basically allow that to be done by the calendar. But I think for creativity, we see that as fundamentally different. It is just another cognitive task, but we see that as something that is more intrinsic to us as human beings. And I think that really, it's not just a scientific question, but it's a value-based question of whether or not we are okay with allowing something to do that for us and to what extent. That's a good point. Because as you're saying that, I'm kind of like

24:59reflecting on the value of creativity versus remembering my schedule. And the first thing that's coming to my mind is kind of very obvious, but like my schedule, remembering that doesn't bring me joy, but creativity brings me joy and purpose and, you know, all of that. That's right. So, for instance, you know, I, one of the things that these models are incredibly, exquisitely good at is programming. And this is because it's a verifiable problem. And so, you know, they can probably program better than pretty much anyone else on earth at this point. But I've always really enjoyed programming. And I learned to program in the, you know,

25:32back when you needed to program. And even now that I can have a model, you know, program me something much faster and probably better than what I could do manually, I still choose to do that because I enjoy that process that, you know, there's a creativity to programming. There's a, there's a problem solving. There's something that's really satisfying about that process to me. So, I still choose to do that. And, you know, I wouldn't do that with, for instance, you know, remembering my schedule for the day. Yeah. Yeah. Right. That's my counter or my, my mimic to that example is I choose to continue to

26:07write my own scripts. Mostly, sometimes our producers write them too, and they do a great job. But I choose to continue writing them, not because I enjoy the writing process. I find it kind of angsty, actually, and difficult and discomforting, but I enjoy the end product. And so, to get AI to write my script for me, I would eliminate that kind of joy. So, I'll keep doing it despite it's, the pain it brings me. Okay. To wrap up, as we have established,

Parting advice on staying human and sane

26:30we are in a very angsty moment of existential dread. Do you have any parting words of advice for how people can kind of contextualize the moment we're in, how they can think about it and just try to stay a bit sane? The history of humanity, of human life, is a history of moments like the one that we're in right now. And therefore, we can look at human history and see, you know, what has been, what has worked. And the first thing to say is that nothing ever works completely.

27:03So, we're not going to get a kind of total fix. There is always going to be pain. There's always going to be loss. And that is a grim truth. And one of the many reasons that we need to stay in deep contact with our, actually, our human capacity to absorb pain and loss. You know, we are entities designed to do that. We're also beings designed to be in meaningful community and find great joy and happiness. And I think that by, you know, turning to anything that we,

27:34each of us can, to have a deeper sense of connection to our own bodies, to our own feelings, to the people around us, and actually to the physical places that we're in on a daily basis. I think those things are, it sounds a little cheesy, but those are the things that really sustain us. Yeah. Ruben? I would say, don't doubt the human brain. It is an incredibly powerful thing. It remains more powerful than AI. It can do amazing things that AI still cannot do. I think for people who are

28:08considering using AI or using AI, I think you should really try to focus on using it in a way that supplements your cognition, in a way that will make you, your outputs better than what you could achieve alone, and certainly better than what it could achieve. What you want to do is use it to improve your work, and then you'll always be able to stay ahead of it. And you will be able to, certainly now, and I think for the foreseeable future, be able to produce better work than it, if you're using it in combination with your own cognition. So use it to supplement. Don't use it

28:42to substitute your own mind. And beyond that, I'd say, you know, get out, go for a swim, go for a hike, you know, take advantage of the natural wonders that we have in Australia. Very, very good advice.

28:57That is Ruben Radeau, cognitive neuroscientist at the University of Sydney. And you also heard from Sophie G, an English professor at Princeton University and the University of Sydney as well. Sophie and Ruben are appearing on a panel discussion titled, Where's the AI Off Switch? at the Sydney Opera House as part of the Curious Festival on the 4th of October. And that is co-presented by the University of Sydney. We've got links to the event and the festival in our show notes. Thanks to producer Rose Kerr, senior producer James Bullen,

29:31and sound engineer Tegan Nichols. I'm Sana Kadar. Thanks for listening. Catch you next time.

29:45You've been listening to an ABC podcast. Discover more great ABC podcasts, live radio and exclusives on the ABC Listen app.