Briefing Chat: What tickling a chimpanzee can tell us about the evolution of speech

**Nick Petruchow** (0:04)
Welcome to the Briefing Chat Podcast, the Friday show where we discuss a couple of articles that have been highlighted in the Nature Briefing. And joining me this week to chat science is Marin Hunsberger. Marin, how's it going?

**Maren Hunsberger** (0:18)
Hello, hello, I'm good. I'm a little overly warm thanks to London's heatwave, but we're here and we're chatting briefing.

**Nick Petruchow** (0:24)
Yes, it is very hot here in London and across most of Europe. Records are being shattered right now, but that's not the news we've got for you this time. I was reading a story in Nature this week about AI and it's about whether AI is, well, kind of ruining our skills.

**Maren Hunsberger** (0:41)
Oh, I don't like the sound of this one, but I bet you I can guess some of the results.

**Nick Petruchow** (0:48)
Well, the one thing to say is these are early results. And as is something we often like to say on the podcast, more data will be needed, more studies, etc.
But this is, as I said, an article in Nature that's taken a dive into the early results that have been coming in around this issue. So you probably heard it said that maybe our skills will atrophy because we're using AI, the idea being that AI will do things for us and we won't need to do them and therefore we'll lose the skills required to do them. And for physicians and software engineers, that does seem to be the case according to these data.

**Maren Hunsberger** (1:25)
Uh-oh, interesting populations to look at. In what scenarios are doctors using AI to replace skills that then they're losing?

**Nick Petruchow** (1:33)
So this particular study was based in Poland and it was looking at physicians who specialize in endoscopy. So this is like using flexible probes to have a look inside people. And in this case, these were experts doing colonoscopies. They've done at least 2,000 in their careers, and they were trying to flag precancerous intestinal legions called adenomas.
And so, in this study, there was an AI tool that was available some days and not others that would also flag these adenomas. And before the tool was introduced, the physicians could identify at least one adenoma during 28.4% of colonoscopies. But after the tool was introduced, if they didn't have access to it, that decreased to 22.4%.

**Maren Hunsberger** (2:19)
Wow. It really is use it or lose it. You got to practice or you get out of practice.

**Nick Petruchow** (2:25)
Yeah, it's interesting as well because this was during just a three-month period after the tool was introduced. So it seemed like quite a rapid and in this case, statistically significant drop in their ability to detect these adenomas. So the study offers say that continuous exposure to such tools could cause conditions to become less motivated, less focused and less responsible when making cognitive decisions without AI assistance.
Uh-oh. That wasn't the only study. So there was a pre-print, so bear in mind, this is not a peer-reviewed study yet that looked at computer science. So in this case, there was a randomized control trial with 52 individuals. So randomized control trial, some people do one thing, some people don't and you're able to infer the differences.
In this case, half of the group were encouraged to use AI tools and half were not. Both groups were able to use the Internet to find answers, and they were given some sort of rudimentary coding tasks, and then were assessed on how much they'd learned at the end with a quiz. And in the group that were encouraged to use the AI, they scored about 50% on this quiz. Whereas the group that didn't use the AI scored around 67% on this quiz, which implies that the people not using the AI are actually learning a little bit more, especially because the questions that the AI group did poorly on were those that required them to diagnose errors in the code, which suggests that they had failed to learn the concepts behind the code, even though they just produced it. So it really seems like the learning itself is the thing that's been affected here.

**Maren Hunsberger** (4:04)
Interesting. Yeah, I think I've seen a couple of different studies come out, you know, in neuroscience and psychology journals regarding how much of the brain is actually involved in completing tasks using AI or helped by AI. And it's really interesting to see where that research is leading in terms of, okay, how involved is our brain in some of these tasks if we use these tools? And is that affecting how much we actually are able to learn?