Wired mag article on hippocampal prothesis
This article is about efforts by six teams to develop a hippocampal prothesis by monitoring the input/output transformations performed by the hippocampus in slice, and then creating an electronic device to mimic them.
The article quotes noted memory researchers Howard Eichenbaum and Norbert Fortin who seem to approve of the methodology.
October 9th, 2005 at 1:12 am
I have some comments about this, but I’m out of town right now and I want to read some journal articles about it first.
October 11th, 2005 at 4:41 am
OK, I was looking for articles about this to see how they constructed the hippocampal model. I couldn’t find anything certain but perhaps this paper describes it:
Richard H. Tsai, Bing J. Sheu, Theodore W. Berger.
A Vlsi Neural Network Processor Based on Hippocampal Model. Analog Integrated Circuits and Signal Processing, Volume 15, Issue 2, Feb 1998, Pages 201 - 213.
It seems to say that they characterized the hippocampus in slice by stimulating fiber pathways and monitoring local field potentials. That is, they characterized the I/O relation of areas of the hippocampus, such as the dentate gyrus, by giving random inputs to the incoming fiber pathways, observing the field potentials which were generated, and then finding equations which could predict the field potentials given the pattern of fiber pathway stimulation.
I am kind of skeptical of this. Wouldn’t you think that computations such as those involving cortical memory representations would have a very fine resolution, i.e. that the state of the computation could only be described using individual neurons?
I would guess that a macro-level analysis would provide very important, fundamental information about the environment in which the computation is carried out, but that to really capture the entire computation you would have to describe things down to a very fine level of resolution (like single neurons). By analogy, if you were monitoring a computer, and you could monitor the information in memory, but you could only get the average of every 1k of memory, not the content of individual bytes, then you probably couldn’t tell the exact algorithm being run by the computer, even though you might learn some important things about that algorithm (say, that it accesses disk once a second).
Likewise, if you created an “computer prosthesis” that reproduced the average behavior of 1k blocks of memory, but didn’t actually get the byte-level values right, it probably wouldn’t be a good addition to your computer.
So, while I think that characterizing this macro-level behavior is useful and important research, I predict that it will not lead to a working memory prothesis.
However, take my skepticism with a grain of salt. The guys doing this research and the guys commenting on it in that article know more about this stuff than I do; Theodore Berger, for example, has published many experimental electrophysiological papers on the hippocampus as well as theoretical papers about modeling the hippocampus. I have published 0 experimental and 0 theoretical papers on the hippocampus.
October 11th, 2005 at 3:10 pm
btw also note that my “you gotta look at the level of single neurons” is pretty much an arbitrary hunch. i dont think there is really enough evidence yet to say that the correct granularity for a model is single neurons, or LFPs, or single synapses, or single molecules, or even quantum. although eventually i think one of these levels of description will turn out to be the “right” level to describe neural computation, at this point it’s just a matter of taste which kind of model you prefer.
October 12th, 2005 at 2:56 pm
This “neural prosthesis” is utterly bogus. It’s roughly equivalent to making a prosthesis for a failed hard disk on your computer, by putting in a little device that emits the same kind of whine as a hard disk does. In other words, the things this “prosthesis” replicates are byproducts that have nothing to do with the essential functions of the system. (The functions of the hippocampus are poorly understood, but they certainly involve storing memory by synaptic modification.)
You should not attach much significance to Berger’s publication record. Quantity does not imply quality — many of those publications are as pointless as the “prosthesis”.
October 12th, 2005 at 11:57 pm
Bayle, I don’t think you (or any grad student, for that matter) should take your/our small/nonexistent publication record as a reason for not being critical about research. I think your gut intuitions are dead on. All of the literature that I’ve read (including many papers from your/Chuck’s lab) tells me that the hippocampus is 1) a very complex system and 2) a very poorly understood system.
I’m sure that Berger is modeling something but it’s hard to believe that he’s capturing anywhere near “all” of the hippocampus. Since I haven’t spent time reading his work (outrageous claims presented only in non-peer-reviewed media are not so high on my priority list), I’ll leave it at that. (The peer-reviewed paper from the IEEE journal you refer to shows a very reduced VLSI system, nothing like the claims in the Wired article.) I think academia is generally very good at not overselling particular scientific projects but sometimes perhaps not.
Regardless of the mathematical models and such, I don’t think even the practical concerns are met yet. Just one example: Do we have any electronic interfacing technology that could possibly connect a VLSI chip to most of the dense synaptic inputs in hippocampus? (No.) If you’re talking about “replacing hippocampus in vivo” as a near-term goal, this is a real concern.
November 30th, 2005 at 2:26 pm
There was a headline about this on Slashdot a couple of years ago, pointing towaards a New Scientist article. A lot of people had interesting comments about it, including this one from my advisor, Ken Norman . I think the URL he posted in the Slashdot comment would have pointed to here .
April 4th, 2007 at 7:50 pm
[...] Berger’s team is trying to make a hippocampal prosthetic (a chip that could be implanted in the hippocampus and help people with damaged hippocampuses). (we’ve mentioned Berger’s team’s efforts before). [...]