AI has taken the grunt work out of many everyday tasks; from writing emails to planning dinner.  But are we starting to rely on AI to think for us and could it be making us dumber?  Today, Steven Shaw from the University of Pennsylvania on his research into cognitive surrender, AI and your brain. Featured:  Steven Shaw, post-doctoral researcher at Wharton Marketing School, University of Pennsylvania and incoming associate professor at Kings College London. Subscribe to ABC News Daily on the ABC listen app.Program:More from ABC News DailyAustralia, Computer Science, Science and Technology, Science, Technology, Biology, Neuroscience, Scientific ResearchTranscriptSam Hawley: Is AI making us dumber? Are we starting to rely on it to think for us? Today, Steven Shaw from the University of Pennsylvania on his research into cognitive surrender, AI and your brain. I'm Sam Hawley on Gadigal land in Sydney. This is ABC News Daily. Steven, you teach at a university, a place of learning. So you get to witness how AI is changing how we learn and think in real time. Just tell me about AI use among your students. What do you see?Steven Shaw: Sure, yeah. So almost all the students are using AI. They're using AI for everything, I would say, right? They have a nickname for each one of them, a chat, you know, for a ChatGPT. I talked to chat today. It's pervasive and it's been very interesting to see how students have integrated AI into their learning processes or how they use it for different types of assessment. I hadn't taught for a few years and I noticed a big difference in the way that students were communicating to me via email, a lot more formal and verbose, I think. And then obviously you see things like the em dash in there. And then with assessments, I would say the bottom end, the low end of student performance has gone away because basically students, if they are under time constraints or maybe have, are juggling a lot of different priorities, they can, you know, have AI do the work for them and come up with something pretty good.Sam Hawley: Wow, okay. And I'm guessing your students think AI is a pretty useful tool, just as a student would have felt when, you know, the calculator was invented or the internet. It just kind of makes things a bit easier.Steven Shaw: Yeah, it can make things very easy. And I'll also say many of my students were doing, separately from in my course, for example, amazing things with AI, like starting businesses and doing things that they would never have been able to do before AI. But I also know simultaneously that in the classroom, you know, at least some percentage of students are basically using it to do a substantial amount of the work, I would say.Sam Hawley: All right, so Steven, you've also been having a pretty deep look into how the use of AI changes how people think, how it can reshape our reasoning. Now you conducted an experiment with 1300 guinea pigs - people. So tell me, what did you do?Steven Shaw: Yeah, so we had participants come into the lab and we also did it online and answer a set of logic and reasoning problems. It's called the Cognitive Reflections Task. It's a sort of a well-known psychological task. I mean, it involves these, like I said, logic and reasoning problems, a little bit of math in them. And what's key about that task is that there is sort of an initial intuitive answer that pops to mind after you read the question, and that's the wrong answer. Right, and you have to deliberate a little bit and think a little bit slowly to come up with the right answer. But all you need to do is read the question and sort of recheck and you'll know that that first answer is incorrect. So that's a key characteristic of the task. And then our experimental group, we basically just said, okay, now you have access to Chat GPT. Here's a Chat GPT terminal here. You can use it if you want to, but you don't have to use it, right? And it's the same CRT task. And what we saw is a lot of participants used the AI during the task over 50% of the time. And what participants didn't know is that we manipulated experimentally whether or not Chat GPT would give them the correct or the incorrect answer if they consulted AI about that specific question. So that allowed us to see how much participants were relying on AI to get their answers. And what we saw is that even when AI was giving them incorrect information, the participants were adopting those answers and then taking them on as their own and submitting them.Sam Hawley: Okay, and you found that in 80% of the cases when participants chose to consult Chat GPT, they actually went with the wrong answer.Steven Shaw: Yeah, yeah, and that number's from experiment one, yeah. Basically, once participants chose to go to the chat, consult AI, they were quite likely to adopt the information from AI and take that answer regardless of whether it was correct or incorrect.Sam Hawley: And you call that cognitive surrender. So just explain that.Steven Shaw: Yeah, so we coined the term cognitive surrender and it's the adoption of AI outputs without verification. So following what AI gives you or tells you what to do without sort of thinking about it very much or giving it much scrutiny.Sam Hawley: Wow, so that's someone surrendering their thoughts to AI. They're letting AI think for them.Steven Shaw: Exactly, I mean, that's the question is, I mean, if you're adopting or just doing what Chat GPT tells you, where is the work being done? Where is the thinking being done? And that's part of sort of the theoretical contribution of our paper is to talk about how that locus of control, the agency is being transferred over to AI in the case of cognitive surrender.Sam Hawley: Well, let's consider now, Stephen, what is actually going on in someone's brain when they surrender to AI. Now you point to three types of thinking. So let's just run through those. First of all, the first is fast intuition. So what is that and who came up with that?Steven Shaw: Yeah, so this is the sort of classic dual process that we've been using for a long time. So there's a theory, there's several theories of reasoning, but most famously is by Daniel Kahneman, right? And you've got Thinking, fast and slow. And so, as you mentioned, there's sort of system one, which is fast and intuitive. That's that first answer or action, you know, that might come to mind when you're reasoning or thinking about something. And then there's system two, which is more analytical, slower, more deliberative. And so those two systems map to the two types of answers that participants could have come up with on their own in our task, right? Intuitive or deliberative, fast or slow, system one or system two. And one of the core arguments we make in our paper is that AI is so pervasive and can do all of these sort of domain-specific thinking type tasks, right? These are sort of thinking machines, artificial intelligence, that we believe that, you know, now this pervasiveness of AI warrants what we call system three, artificial cognition that can go alongside systems one and systems two. You know, systems one and two are in the brain, you know, evolved, whereas system three is designed and external, automated and data-driven, but performs the same sorts of artificial cognitions as systems one and two.Sam Hawley: Wow. Okay. And so system three, which you've come up with, is artificial thinking.Ad: Yeah.Sam Hawley: Outside of the brain.Steven Shaw: Exactly.Sam Hawley: All right. So does it then override the first two thinking phases altogether?Steven Shaw: It sort of depends how you use system three, I think. If you're still engaging system two in your decision-making, and then you're strategically offloading tasks to AI, right, so you've got some deliberative thought and you're saying, I'm going to intentionally use AI for this element of cognition, right? That's what we call cognitive offloading. Whereas if you aren't engaging deliberative thinking and then you're deferring, outsourcing your thinking to AI, or that's what we call cognitive surrender. So whether or not artificial thinking overrides, you know, systems one or two, depends on, you know, how you're engaging system three.Sam Hawley: Okay. But if you used all systems at once, so you adopt AI, but of course you still put thought, fast thought, and also deliberate thought, which, you know, complex decision-making into the whole mix, you could come up with a much better result, right?Steven Shaw: Yeah, absolutely. I mean, particularly, let's say for analysing large data sets, you know, very quickly, you can't really do that with your own critical thinking, but if you are careful about how you manage that data and prompt system three, let's say, you can extend your line of reasoning with augmented data analysis, let's say.Sam Hawley: But at this point, I mean, you've got students, how many students would be just using system three? That is artificial thinking.Steven Shaw: Well, that's a tough question to answer, how many? But I think what the key is, is that right now with the way that AI is designed, right? It's sycophantic and it's fluent. It'll give you answers that are confident to just about any kind of question that you ask it. With the technology that's available 24 seven like that, there are going to be students who, some non-zero percentage of students who will engage in cognitive surrender just because it's available and it's easy to do and they could spend their time doing something else.Sam Hawley: All right, well, Stephen, let's consider then what happens if we do switch to system three too often, I guess. AI is progressing so quickly, but I guess your message is that we need to understand how it can change the way that we think.Steven Shaw: Yeah, I mean, we don't necessarily need to be scared of cognitive surrender, you use the word reshape. I like that word reshape a lot, right? We're entering a new era of thinking and decision-making and system three is part of that, right? And so we have to have the right vocabulary to discuss issues and we have to understand the risks involved with cognitive surrender, but there are both risks and opportunities of engaging in cognitive surrender and so it's becoming more and more clear that our cognitive capabilities are changing and we may de-skill ourselves in certain areas or never skill in certain areas, but hopefully we can also gain some of these opportunities and adapt and augment our thinking as well.Sam Hawley: So we need to know when to accept and when to override AI generated results, but how do we actually do that? Just give me an example.Steven Shaw: Certainly, in a student or a learning context, if you want to learn something and you wanna become better at it, or if you want to gain a skill, if you're engaging in cognitive surrender during the learning process or while you're gaining that skill, right? You're not the one learning, the AI is learning and the AI is doing that skill. So you won't learn those things or develop that skill. And there's a growing body of research, I would say, particularly in the education space. There's some Wharton colleagues of mine, for example, that look at high school students. And in those studies, when participants have unaided, high school students have unaided access to AI, their performance goes up. They look quite good on paper. They get better grades or do better on these math problems than the students who don't have access to AI. But when you then take away AI, take away system three later and test those two groups, the individuals who relied and had access to AI perform actually much worse than the brain-only condition because they didn't actually learn anything. That performance, that math was being done by the AI, right?Sam Hawley: Okay, well, Stephen, I guess the real question is though, is there a risk that AI could make us dumber? And is it already doing that?Steven Shaw: I mean, I'm hesitant to say that AI is making us dumber. You know, we're going through a rapid technological transition. And as we talked about before, right, AI is reshaping this cognitive ecology of how we reason, how we think and how we make decisions. And so during transitions, I mean, transitions are challenging and some things are lost and some things are gained. And so is there a risk that AI will make us dumber? Yes, I think there is a risk, right? I mean, if we have people just blindly following what AI tells them, they're not even thinking for themselves, right? And that's kind of part of the definition of being dumb, I guess, right? Even if AI is telling them very smart things to do, they can't, they're not thinking independently and that's not really their own autonomy or agency. But I think, you know, we just have to be intentional and careful about how we use AI ourselves in our day to day. And hopefully policy makers will pay attention to behavioural science and the science that's coming out around use of AI. And we can implement effective policies in education, law, medicine that help us fight the rise of cognitive surrender in areas that we see as uniquely human.Sam Hawley: Steven Shaw is a postdoctoral researcher at the Wharton Marketing School at the University of Pennsylvania and an incoming assistant professor at King's College, London. This episode was produced by Louise Miolin and Ilaria Brophy. Audio production by Sam Dunn. Our supervising producer is Sydney Pead. I'm Sam Hawley. Thanks for listening.