We can edit, directly observe, and record/playback simulated brains. We can test ten million different models on top of a recorded simulation and see which one fits best. Eero doesn't think we could learn about how the brain works by simulating a trillion slightly different permutations on a brain, or (maybe more ethically) small subsystems of a brain, and observing how each behaves?
Sweet god, the economic implications! http://mason.gmu.edu/~rhanson/uploads.html
Horrifying, maybe, but I don't see how people can get away with suggesting this will never be valuable to anyone.
> I'm a little amazed people aren't seeing the uses of this.
That's not the issue - I imagine everyone here would like to see these goals reached. Nobody has to sell anyone on the rewards of neuro research. The question is merely whether this specific project can deliver them.
> We can edit, directly observe, and record/playback simulated brains.
The idea itself is a good one. However, the key issue becomes choosing an appropriate level of detail for the simulation. I believe a blanket choice of "let's just do the entire brain" is computationally infeasible right now, plus we don't have good enough models to actually program the thing - but most importantly even in a future where these problems are solved the device seems like a blunt and unwieldy instrument that won't give up its data easily.
> or (maybe more ethically) small subsystems of a brain
That's what's already happening all over the world right now, in thousands of independently scoped simulations and experiments.
>That [simulating small subsystems of a brain] is what's already happening all over the world right now, in thousands of independently scoped simulations and experiments.
Woah! Links? The searches I can come up with aren't turning up anything besides that simulation of a rat cortical column and the various attempts at nematode uploading.
Never is a long time ;)
> Woah! Links?
Any university with a neuroscience department does this. Go to your local university's website and browse what they're doing in that area: more likely than not you'll find something interesting. Basic research on neurons has become very common, and computer-based modeling is a fundamental part of it. The perception problem here is that, say, modeling the signaling behavior of locust neurons seems like a very inconsequential piece of the puzzle - but in reality it's what we need to do to figure this stuff out.
The fundamental problem in neuroscience research is not a lack of complexity, for now we need to move away from complexity in order to observe the behavior of basic building blocks. It may seem embarrassing how we're still at that stage, but it's where we stand.
http://www.math.pitt.edu/~bdoiron/
http://neurotheory.columbia.edu/~larry/
http://www.cns.nyu.edu/wanglab/
http://alleninstitute.org/our-research/modeling-analysis-the... (modeling group)
There are also some companies that are led by people with academic experience doing spiking neural network simulation and seem to be trying to commercialize it:
The main problem with the project though is not the missing scientific data - that just makes it unfeasible for now. What makes it unfeasible in any context is the detail level at which the simulation is supposed to be carried out. This quote from the article expresses it better than I could have done:
Eero Simoncelli, a neuroscientist at New York University. "Would you try to understand the universe by simulating every molecule? What would you have achieved? It’s going to be just as complicated as the real thing and you won’t understand it any better."