Of course a path to AI informed by neurological/biological structures is best. But does it carry the risk of Artificial Dementia? Not that that could be worse than LLM training degradation due to poor quality data lakes, which is already becoming visible.
You may find the TV show Pantheon interesting. The first season is on Netflix.
1 Like
Wonderful recommendation, thanks. I guess the creator of Pantheon is Ken Liu, the translator who brought Cixin Liu’s “The Three Body Problem” to the English world. I ask about dementia because I think it’s a fountain of knowledge that TBP could use. Sometimes figuring out how something is broken leads to insight into how it’s constructed.
Thanks for your question @scidata , it’s an interesting thought. While I’m not an expert, the most common cause of dementia, Alzheimer’s disease, is believed to relate to the build-up of malformed proteins and interplay with the immune system. These are low-level details of the brain that we are not simulating in Monty, so we wouldn’t expect these to be an issue.
Importantly, I’m not aware of any evidence that any forms of dementia are a result of over-learning (e.g. trying to acquire too much information), which would be more akin to catastrophic forgetting in deep-learning systems. If anything, evidence suggests that learning a great deal during one’s lifetime, such as having a high degree of education or learning multiple languages, is actually protective against dementia in humans.
Hope that helps!
5 Likes
Seems an interesting thought. My father had vascular dementia due to a heart problem and lower oxygen supply to the brain, perhaps causing reduced and misfiring synapses. Maybe a TBP research area for someone?
Also I have mild autism and a partner had schizophrenia which seems to have some similarities with synaesthesia. Perhaps more research areas for TBP?
I’m sorry to hear about your father. That said, I’m not sure how viable TBP would be for this kind of research. While it’s clearly heavily inspired by the cortex, it doesn’t fully emulate it. Plus, it’s like @nleadholm said, dementia and model collapse are actually pretty different things.
I’m sorry to hear about that @forestacorn . At the moment the focus of the Thousand Brains Project is on building artificial intelligence rather than neuroscience research that might have clinical implications. That said, if the Thousand Brains Project is as successful as we hope it will be, then it would likely serve as a useful mechanistic model of the brain, with potential downstream benefits for clinical neuroscience research. Unfortunately I think it would be a while before that’s the case, but it’s certainly worth having in the back of our minds.
3 Likes