Comment by polyfractal | Hacker News Reader

Comment by polyfractal | Hacker News Reader

Parent Full thread

I came storming into this thread ready to post exactly your statement. The title is way over-blowing the situation. There are so many caveats here its hard to do it justice.

First, they found single neurons which trigger memory, a far cry from containing the entire memory. Think of them as gateways to a house. A gate does not hold the house, but it does let you access the house.

Second, they didn't remove any neurons (from my brief skimming of the press release at MIT[1], I don't have time to read the paper now). There is almost a 100% chance that there is a large subset of neurons which will trigger the same memory. Which means this house has a lot of gates. Nothing is becoming "unreadable" here.

Third, this trick of using optogenetics to trigger a memory has been shown several times over already. Here it was reported by Boyden and Hausser in 2009 [2]. Presumably this piece is being reported because it was performed by Tonegawa's lab. Not discounting the research, there is likely a lot of advancement on the underlying science, the optogenetics (which are undoubtedly being borrowed from Boyden's lab across the street) and the understanding of mechanism...but it isn't new.

[1] http://www.mit.edu/newsoffice/2012/conjuring-memories-artifi...

[2] http://www.technologyreview.com/biomedicine/23767/

Exactly. The function of a neuron is to trigger other neurons. If you removed a neuron, but manually triggered exactly the neurons that the removed neuron would have triggered, it is to be expected that the same memory would be recalled.

Obviously, this argument breaks down at some point, since by induction we could remove 100% of the neurons and manually trigger the axons to the muscles that would have been triggered by triggering the single neuron in the intact brain. This would give the same observable behavior, but it is not clear whether the same experience would be had. But this is getting into the area of consciousness, which is something that nobody understands (yet).

If there are cycles in neuron graph (and I'm pretty sure there are), then you cannot be sure that removing one neuron and triggering all the neurons it was connected to in correct way is the same - you'd need to also simulate reaction of that neuron you removed to possible feedback it takes after the first stimulation, maybe cycling in such loop many times.

Theoretically, I'd agree. But "memory" is an extremely fuzzy definition. Perhaps the memory is not 100% identical, but missing a tiny detail about the color of the triangle on the guy's shirt in the background of the TV.

There is also a lot of redundancy in neural networks. Large swaths of neurons fire together in concerted patterns, and while removing one may tweak the pattern slightly, it is equally likely that the other neurons will just adjust their weights to keep the pattern constant.

They haven't just "found single neurons which trigger memory". They found the only neurons that were active while experiencing a specific situation and memorizing it and they managed to show that activating only those neurons is enough to trigger a reminscence of the very same experience. You cannot say some other neurons would trigger the same memory, because nothing changed about those neurons during "generating" this memory (under the assumptions science currently holds about the brain), so they cannot store it.

Boyden and Hausser gave a talk claiming they can do it in 2009. However, I cannot find a citation claiming they ever published this work. As it goes in the field, first to publish is more important than first to claim.

Reply on news.ycombinator.com