[ExI] The problem of embodiment and uploading humans

[Ben Zaiboc]

On 12/09/2026 19:29, BillK wrote:
>
> I discussed with Kimi AI and Gemini AI the news that a fruit fly brain digital simulation has been trained to play Doom and other computer games. 
> During the discussion, they both mentioned the problem of embodiment. 


I don't know why it's being called a problem. It's a requirement, we know that. As far as I know, nobody who thinks seriously about uploading is going to be using the cartoon version (the "classic science-fiction fantasy of mind uploading") of a disembodied mind simulation, that's just silly. We know that the virtual brain will require a virtual body for it to work properly. It's kind of obvious, really. Just as obvious as the whole system needing a virtual world to exist in. In fact, almost all of the advantages of being an upload stem from being able to control your virtual body and virtual world.

It's no wonder that people who think that a disembodied mind is what uploading is all about aren't keen on it. They're missing nine-tenths of what it's all about!

I think the significant thing is that bodies are much simpler than brains. Virtual bodies especially, won't need to be individually tailored, as long as the interfaces are properly designed. To be an upload of your current self, your connectome must be replicated at a very high level of fidelity. The same is not true of the virtual body, and I expect an 'average body' will be arrived at, that would only need minimal tweaking to be a suitable embodiment for just about anyone's mind. Then it can be up to the individual upload to take it from there, and customise their body to be whatever they want.


What's much more interesting is embodiment as applied to AIs. They are going to need it as well. I think that's the main importance of the recent developments in robotics, rather than all the things people tend to talk about. The development of embodied AIs will be, I think, a major step forward. Not that bodies couldn't be virtual ones, of course, but nobody sees much advantage to that. Real-world robots with AI brains, though, is a different matter. They will inhabit the same world we do (the 'real' one), and we will find zillions of reasons to want that.

Once this happens, they will need subsystems for managing these bodies, which means feedback mechanisms to maintain homoeostasis, whatever that means in a robot. That will be the basis for emotional systems to develop, just like with biological creatures, as well as a more definite sense of self, etc.

That's when we'll start to get AIs worthy of the name, I think.

-- 
Ben