Dr. Charan Ranganath explains how memory actually works in the brain, why human vision beats computers at what matters, and what Neuralink can and can't do yet.
You're an expert on memory. If you could get, say, neuralink, and it gave you fully accurate memory to everything you've experienced, what effect would that have on your memory? I have a postdoctoral researcher I'm working with, and he's coming from the world of computer vision, actually. And one of the things that he's been studying is the limitations of computer vision models at certain kinds of challenges. For instance, if you give like an image and it can't tell which ones are the same and which ones are different, and it turns out the ones that the computer vision models struggle on and people do well, it's because people, people never actually see an entire image. They just move their eyes and grab little bits and pieces, and then the brain stitches that together into a feeling of what's in front of you, a meaningfully integrated image. And it turns out that the problems that these computer vision models have are on images that require people to make a lot of series of eye movements and then integrate that information. So what this tells us is people, on the one hand, are getting less information than a computer vision model. A computer vision model gets every pixel of an image all at once, right? A human is just grabbing little bits and pieces, but then the brain is meaningfully integrate that sequence, integrating the information that comes in from that sequence of eye movements into a meaningful representation of what's in front of you. And that's the magic of vision, that's the magic of memory. It's the magic of language, is that ability to make the most out of a small amount of information. And so to me, I think the challenge with something like, you know, if you were to come up with some kind of prosthetic, is you have to move away from this idea of recording as being the function of memory and something that's much more strategic, opportunistic, getting the information that's most new and useful and prioritizing that in building together a memory. And so I think there were neuralink is really brilliant in terms of the hardware, in terms of the AI. I don't think they're really that far ahead at all, but the hardware has been absolute genius. And I don't think they're anywhere near close enough to understanding anything to do with human memory.
