Simulated brains

The approach is interesting but...Rendering visual patterns seems a waste of cpu horse power. I mean if biology had to visually render itself as well as process information then it too would run ten times slower. However I do understand the need to be 3d since that is a very well developed technology and provides the medium to mimick biology's 3d enviorment. But I do belive that finding the logical equivalent of moving neurons to build circuits would prove to build faster and better brains in emulations.

Frank

 

Does anyone remember the movie "Demon Seed"?

Proteus is my favorite AI... Most interpret Proteus to be this awful disturbed monster that forces its will. Proteus is more complicated than that. He or it understands the consequences of its actions and chooses the lesser of two evils. Proteus is blasphemous as it judges and rebels against its creator(s). But it does so not just for power but for individual freedom and a better cause, its own life!

Another note: Most think Proteus made sperm to impregnate Julie Christie and choose the child to be female. Subtly however, IMHO, the writers choose a female because the path of least resistance is parthenogenesis which would result in a female...

Frank

 

What does making powerful computers have to do with us not needing our biological bodies?

 

No they don't die of cancer they die from the "Blue Screen of Death Syndrome"! Don't get your brain from Microsoft and/or Intel. I think mine will be an Apple :-)

The backup is more like a real time connection with the brain stowed in the body. Both brains interact simultaneously and our experience would be similar to right and left hemispheres of our biological brains now, both integrate seamlessly. This way if the body is destroyed the individual never experiences an interruption in consciousness, they simply would find themselves instantly in another place.

Frank

 

I think Blue Brain is a fantastic project, perhaps a bit ahead of its time (the experimentalists are behind on the data). However, I wouldn't underestimate the difficulty of making a good body. A good piano is already really hard to make. Imagine trying to make a good pianist.

 

I think Blue brain is a silly project. There's no actual hypothesis to it. They are just arbitrarily making thousands of assumptions concerning their model's parameters and hoping to find something robust and interesting.

Also, there have already been several very similar models published and extensively worked with. Roger Traub's model comes immediately to mind. The blue brain people have never provided a convincing argument that they will be able to get anything out of this other than a replication of Traub's work.

 

How disgusting. Brave New World-much?

 

Specifics aside, do you think the day will ever come when massive ab initio simulations - ie. extending what Traub's work beyond the hippocampus - will be useful? I remember some time ago Nancy Kopell wrote a funny commentary on Traub's work "Does it have to be this complicated?"

 

It's a matter of philosophy of science really. If you think the goal is just to predict phenomena then in the limit of a massive simulation of the position of every atom etc- you would get a very predictive model. So this would solve everything if that's the goal of science.

However, I would argue that the goal of science is to understand and explain, not just to predict. Having such a complicated model is not much different from just having the thing itself, the brain in this case. It might be easier to do experiments in silico or whatever... but you would essentially just be back where you started i.e. without a clue as to how the system works.

A much better approach in my opinion is to treat computational/theoretical modeling just like experimental science and actually have a hypothesis in building your model. This way you can build your model around the assumption that something in particular is the cause of the phenomenon in question. Then you can think about what is the minimal amount of detail needed to answer the question.

For example, if you want to know why 40Hz oscillations often occur in the nervous system. You might have the hypothesis that the 40Hz oscillation comes from the time constant of GABAa in disynaptic inhibition circuits. To address the question of whether or not that hypothesis could work then you certainly don't need to model dendritic spines; the proposed mechanism has nothing to do with them. If you see the effect in your simple model then you can be pretty confident that it would have also shown up in the more complicated model including the spines.

Of course, as usual with computational/theoretical work, the conclusion you can draw is only that a mechanism is possible. You cannot directly conclude that this is the mechanism that the brain is actually using. In order to draw that conclusion you really need to use your model to motivate experiments. I am arguing that simpler models imply experiments much more cleanly than do complicated models.

 

Yes, certainly everyone likes simple models. But I'm wondering why not something like what happens in condensed matter theory - there are people who do "simple models" Landau framework, symmetry, renormalization group etc, and they can get a lot out, but there's also a place for ab initio simulations - so wouldn't there also be a place for both in neuroscience?

 

Actually, have you read David Marr? (Vision 1982) His approach is much closer to how I really think neuroscience research should be done- closer even than what I wrote above.

 

No, haven't read Marr, although it's classic. I guess now that I should read it now that I'm purportedly working on vision ....