That Neuroscience Guy - The Influence of AI on Neuroscience

Episode Date: July 24, 2026

In today's episode of That Neuroscience Guy, we discuss the neuroscience behind the pros and cons of implementing AI into neuroscience. 

Hi, my name's Olive Kirk Olson, and I'm a professor at the University of Victoria. And in my spare time, I'm that neuroscience guy. Welcome to the podcast. So, unless you've been asleep for the last couple of years, you've probably seen the enormous growth in AI. AI is everywhere now. It's on our phones, our computers, companies are using it to answer the phone. Companies are using it to send emails. We're using it for the same reasons.

And it's in neuroscience. AI has had an impact on neuroscience. You know, is AI helping us understand the brain better or merely helping us generate better-looking stories about it? You know, what is the impact of AI on neuroscience, a neuroscience research? It's a topic I find quite fascinating personally because I'm actually a big fan of AI.

I think when used correctly, it's going to change. the world. But there are some downsides. So I thought I'd use this episode to talk about, you know, is artificial intelligence changing neuroscience for better or for worse? So let's start with the positive side. If you, you know, measure brain data, whether you use EEG or fMRI tools we've talked about, it's an enormous amount of data. If you record EEG for 10 minutes, you know, know from 32 channels, you're literally getting millions and millions of data points. If you take a three-dimensional picture of the brain with fMRI, you're getting a massive amount of data. If you're even looking at just behavioral data, just looking at how people perform on surveys,

it's still a massive amount of data. And one of the problems with that is, you know, as a human, it's hard to process that much data. that, you know, it's even hard to comprehend it. And this is an area where AI excels. It can look at that massive data set and find patterns in the data that the human brain has a hard time detecting and understanding. So I think, you know, in a positive sense, if you take EEG data, even the work in my own lab looking at dementia detection, you know, we've gone away from like the methods I would have used when I was in grad school. and we've embraced AI because it's finding patterns in the brainwave data that separate people with dementia from people that don't have dementia. It's also helping clean data.

You know, EEG data or fMRI data is actually quite noisy. It's not as simple as just putting an electrode on someone's head. There's a massive amount of work that goes into what we call signal processing. It's the stuff you do to clean the data, to find artifacts, the data and other issues with the data. And AI, again, has come up with some pretty sophisticated ways to clean brain data so that it's easier to understand and it's easier to interpret. Another place where AI comes in is I've been working with a massive data set.

We're talking over 50,000 data, individual data sets from 50,000 people. And, you know, to process that at the individual level, to see what's going on for any given person, takes an enormous amount of time, right? To just process a single EEG file takes a couple of minutes, even longer, depending on how fancy you want to get with the data processing. But AI can take that massive data set and go in and find biomarkers, and it can do it at the individual level. So it's able to understand that a massive data set, so 50,000 people, and it can look for individual differences between person 12 and person 13 or person 42,000 and one. Because it can do that, it's actually speeding things up, all right?

You know, some of the research processing we do and some of the research tasks that we have to do are quite slow when done at the human level. It used to take us a month to process a 30-person EEG study to make sure we're 100% certain of the findings and what's in the data. And now that's down to a few days. And what that means is it frees people up to do other things. Or does it process more data? So the pace of research has been increased. And if you think about practical applications there, that's going to accelerate things like drug discovery. because the trials and the stuff that took so long are going to be done more and more quickly.

You know, I had an interesting discussion with a colleague about the COVID pandemic, and if we had AI fully engaged at that point, it probably would have accelerated what we know about COVID and what we did to treat COVID and make ourselves safe from COVID. Another thing AI is good at is it looks at problems from different unique angles. It's got access to so much information that, you know, we've asked at research questions, and it comes up with what theories that we would never have come up with. Now, sometimes those theories are great, and like I said, we'll talk about the negative sides of AI,

but sometimes they're really groundbreaking, and you say to yourself, wow, that's a really good idea. So it provides another point of view, another perspective. And the other thing it's good at doing is it allows to make, it allows neuroscience to be more accessible. You know, if you take a research paper that, that, you know, you discover, you know, without the right training and education, reading your average research paper is quite challenging, but very few of people in my line of work are skilled writers in the sense of accessibility. They're technically good writers. They can write good science, but they can't write something that the average person can understand. But what can you do with AI? You can take a research

paper, you can feed it into AI and say, hey, give me a summary in layman's terms of what this paper's all about and what the results mean. And it will do it. So it allows anyone to better understand neuroscience and neuroscience research. Now, that's a lot on the positive side. But what about the negative side? Well, if you don't know what you're doing and you just dump some data into AI, you're effectively engaging a black box, right? You're putting data in and saying, okay, what does this data mean? But it doesn't mean that it's accurate. And it doesn't mean that you understand what is doing.

And that black box can be dangerous because one of the big problems I have with AI is it makes mistakes. All right. You know, when it processes data, it sometimes makes the wrong assumption. It doesn't get the math wrong, but it makes the wrong assumptions that you, uses the wrong settings, if you will, when it processes data. And as a result, you get results that aren't actually true. And because of that, sometimes it does what we call overfitting.

You get these incredible results. But it's only because it's played with parameters that shouldn't be played with. And if you don't understand what it's doing, and you can't look at those parameters and say, well, actually, that doesn't make sense with human data, then guess what? You believe the result and it's not actually representative of what the data truly says.

It also creates a sense of bias in a sense because one of the things I've discovered with AI and I'm sure you have too is if you give it a leading hypothesis, so let's see you're processing a data set, you want to compare groups A and B and you say I'm under the impression the brainwave response for group A

will be larger than the brainwave response for group B. AI will actually typically aggressively work to make your prediction come true. Again, it's playing with settings and parameters, but that creates bias. One of the more frustrating areas for AI, and I've had personal experience with this, is you could say to AI, hey, you know, I need a paper that sort of supports this idea I have, and it will go out and give you a citation. But it's pretty easy to figure out that sometimes the citations it gives you

aren't real papers. What it does is it say it's found four or five papers and it's sort of it's almost like it's meshing them together into the perfect citation. But then if you go to say Google Scholar and trying to find the paper, it doesn't exist. And it can do this with analyses, it can do this with images,

it can do this with claims. and it looks legitimate, but it's actually wrong, and it's just because the AI's made mistakes. And as a result, you know, what's happening here is researchers, whether they're graduate students or professors, and then this is true of almost any field, I'd imagine. You're sort of outsourcing understanding, right? So you're asking it, well, I got this data, what does it mean?

And if you don't understand the data and you don't really have the right training, you just believe what you're told. all right and that's a real problem okay because researchers that are using AI that don't truly understand the techniques they're using are generating false claims and that has impacted the publication process I was at a conference recently in Belgium and one of the big discussions we had is that since AI has become popular if you will in the last couple years, the number of scientific papers has doubled, and some people are arguing, tripled, and it's because you can get AI, you can give it a bit of data, and it will write a full paper

for you, and it's something you can submit. But it doesn't mean it's a great paper, all right? It's a paper, right? It's a review. It's a grant application, but it doesn't mean it's great. It's AI's take on it. And again, without that expertise, you might think, well, hey, this paper is really cool. And another place for this kind of, there's bad sides to this, is in the commercial sector of neuroscience, there's a lot of companies that are making some bold claims. We are using AI and we understand depression. We understand anxiety. We can detect, you know, MCI or dementia, space that I'm interested in. But the AI models are using are overreaching the actual data. It goes back to when I say it's trying to justify the hypothesis you have.

And as a result, there's all of these claims about these amazing things that AI can do. But if you actually start pulling it apart, the data, the evidence isn't that great. And the biggest risk that I see is that it's taking away understanding and explanation. All right. It's giving results that people don't truly understand, but they just roll with it. them. You know, I've actually had students and even some colleagues make some bold claims and then if you press them on it, they're not even sure why they said what they said, because it's AI told them to. And that is going to corrupt neuroscience if we're not careful.

So that's my take on the role of AI in neuroscience. It probably, you know, rolls over to other fields. But I wanted to offer my two cents. I think AI is going to have a massive positive impact on neuroscience and neuroscience research, but I think at the same time, we have to take it carefully, use caution, and ensure understanding and not just results. Okay, don't forget the website that neuroscience guy.com. There's links to Etsy, our merch, and Patreon, where you can support us directly. Heck, use AI and get AI to tell you what Patreon is if you don't truly understand. But those of you that are supporting us, it really helps. I just transfer the money to graduate students to my lab who support the podcast.

And they do appreciate it. It helps them pay their tuition, helps them pay for books, helps them pay their rent. There's social media, of course, at that neuroside guy on Instagram, threads, and X. We're putting up lots of content on the Instagram page. So please keep taking a look there. It's extra material previews of coming attractions. And of course, the podcast itself. If you haven't already, you know, please subscribe.

It really helps us. It motivates us to keep going. You know, I wondered if we were going to stop when we got over a million downloads, but the team, Jen and Matt are behind this. So, and I'm motivated. So we're going to keep pushing on. So thanks again for listening, and I hope you found this episode, and the other episode's interesting.

My name is Olive Kurt Olson, and I'm that neuroscience guy. I'll see you soon for another full episode of the podcast.