Reading Thoughts with a Brain Scanner

Chris -   Also in the new this week, researchers at University College London have developed a way to read a person's thoughts and basically see what they're seeing in their mind's eye using a brain scanner.  Dr. Demis Hassabis is behind this study and he's with us now.  Hello, Demis.

Demis -   Hello.

Chris -   Welcome to the Naked Scientists.  Tell us, if you would first of all, what it is you actually have discovered.

Demis -   What we did in the study was to basically get volunteers to view three short video clips.  These are video clips of every day common events like someone posting a letter or someone throwing a piece of trash in a bin.  Then what we did was to ask them to memorise those video clips in as much detail as possible, and then a short while later they were placed in the scanner and they were asked to freely recall those three memories in any order they wanted to, and as many times as they wanted to.  After the scanning, we analysed their brain scans and we found we were able to tell which one of the three memories they were recalling and at which time, at an above chance level.

Chris -   What was the scanner seeing?

Demis -   We were focusing on this small region of the brain called the hippocampus that is known to be essential for this kind of memory.  We used quite sophisticated machine learning algorithms to try and spot patterns in people's brain scans, and that's what we're able to do here with just the activity patterns in the hippocampus, and we're able to tell from that which memory someone was recalling.

Chris -   How is the brain playing out that memory through that brain structure in a way that you're able to eavesdrop on?

Demis -   What we think is going on is that when they first see these videos in the training session, the hippocampus is responsible for laying down a memory trace, or a copy of that memory.  So that's what allows you to remember something in the future, you basically reactivate that memory trace.  So, what we've done here is try and investigate that memory trace directly and come up with a technique that allows us to look at that memory trace directly in vivo in a functioning human brain.

Chris -   And how does this inform our understanding of how that part of the brain actually works?  Presumably also, how we then extend that into what happens when it goes wrong with aging and dementia and things?

Demis - This study is part of a program of studies that are investigating the fundamental structure of memory.  What we'd like to know is things like - what aspects of an experience are preferentially recorded the brain?  Obviously, these are important questions because if we can understand how the brain does that, then maybe we can help form therapies for people who have disorders such as Alzheimer's or dementia, where we can try and enhance their memory for the things that they need to remember, over and above other stuff that is not so essential to them.

Chris -   And could you extrapolate the study to look into other modalities, other aspects of memory?  You just asked people to watch three short films, but could you make it much more detailed?  How far do you think you could take this?

Demis -   What we're planning to do next, and in the process of doing at the moment, is extending it further into looking at whether, for example, it's the content of a memory or the context, i.e. what happened or where it happened, that actually best defines the memory.  That's the start of actually breaking down memories into their components.  So we can actually eventually start looking at which features or which aspects of an experience the brain is coding for.

Chris -   But of course, it is a little bit artificial because your system had to learn from these people first in order to know what it was looking for and then record back when they did their free imagination and matched the two things together.  It's a bit further down the line before presumably you'll be able to put someone in the scanner and then workout what they're thinking about without having pre-learned?

Demis -   Yes, that's right.  We're long, long away from creating some kind of general purpose, mind reading machine or something.  What we did here is that these are predefined memories that we know that the volunteer is going to choose between.  Even then we're not 100% accurate.  So very much at the moment, it's still fundamental research rather than any kind of application such as that.

Chris -   So the HMRC, the Inland Revenue are going to have to wait a little while before they can tell whether people are being absolutely honest with their tax returns in the future.

Demis -   That's right.  That's right.

Chris -   Demis, thank you very much.

Demis -   Thank you.

Chris -   That's Dr. Demis Hassabis.  He is at University College London.  If you'd like to read a bit more about the paper he was discussing that he and his colleagues have published is in the journal Current Biology this week.