Medical Biological components in computing

[SynapticSelf]

Recently, someone pointed me toward this news report, which I have cross referenced (only to find that everyone got their information from the same limited source).

http://www.napa.ufl.edu/2004news/braindish.htm"

As I understand it, rat neurons formed an interactive network with an electrode in complete isolation. (The brain mass was kept alive in a solution.) The electrode was connected to a flight simulator, and over some time, the "brain" learned to keep the plane flying straing n' steady.

This raises (many) questions. Because the neural matter had no knowledge of an objective, in contrast to most Brain-Computer Interface demonstrations, where the subject is actively trying to accomplish some given task, how was this goal accomplished? Put another way: What was it about the straight, level path of the simulated jet that made it a favorable condition? Why wouldn't the rat brain keep conditions constant but less pleasing to the scientists involved, like a continual barrel role?

Was the electrode possibly wired in such a way that signals were only relayed when the jet was off balance, and the brain naturally seeks out the "path of least resistance", so to speak - resistance referring to interference, in this case.

Perhaps the mechanisms that govern the equilibrium in mammals could offer some insight. What do we know about this system? I will continue to read, and will update if I come across anything relevant.

 

[Q_Goest]

Hi SS. There's a home page here that also has a link to the published paper.
http://neural.bme.ufl.edu/page12/page1/page1.html

 

[SynapticSelf]

Thank you much, Q Goest!

Let me see if I understand this:

First, the brain tissue was "recycled" (chemically) and "ground/rubbed" so that there wasn't any bias in the original connection scheme. Then, by taking advantage of LTP, they determined that high frequency pulses to the system resulted in a decrease in action potenial feedback, and that low frequency pulses strengthened the feedback.

They then used two of the sixty channels to as both input and output, in effect creating two closed (although very entangled, I'm sure) circuits - one for pitch and one for roll. During each phase, nine separate high frequency bursts, proportional to the degree of error, were sent to the neural network. How was it proportional? Are we talking signal strength (amplitude) or duration?

Basically, the only change in biological feedback was the strength of the return, as expressed by the number of Action Potentials, which, of course, would be proportional to the incoming signal in any given evolution. At first the biological response was "faint", but strengthened over time. As I understand it, it would continue to strengthen, even past the point at which it was favorable. The only way to sustain the "perfect settings" of the biological system would be to invoke a seperate, electronic system to regulate the frequencies fed to the neurons. Basically, I see no uses for this in the near future, except as a means to study the pattern recognition that we see in natural cortex. (Which I'm not trivializing, believe me - I'm thoroughly excited about this!)

Lastly, I'm assuming that the mediating computer was already considering the solution to the pitch/roll error, at least roughly, because it would have to decide which direction to apply the neurons' feedback. We are dealing with absolute values here, aren't we?

What are your thoughts, people?

 

[Q_Goest]

Hi SS, What you said sounds about right. It sounds from reading the paper that they spread roughly 25,000 cells (neurons and glia) out over this microelectrode array (MEA) and allowed them to connect themselves. Once the cells connect, they start communicating spontaneously.

From various electrical interface points on the MEA, they can measure how one pulse statistically affects another one. It sounds to me as if the researchers simply produce some stimulation and measure the reaction. There are 2 stimulations sites and from what I gather, the researchers are looking at various reaction sites across the entire MEA. From that stimulation/reaction information, they determine how to control the reaction by changing the stimulation. With this information, they have enough knowledge to stimulate this array of cells and get the desired response, and it's that response which they now know enough about to have it provide input to a flight simulator.

So to respond to your original line of thought:

Put another way: What was it about the straight, level path of the simulated jet that made it a favorable condition? Why wouldn't the rat brain keep conditions constant but less pleasing to the scientists involved, like a continual barrel role?

It sounds to me like there is no favored condition, there's no intent carried out by the neurons, the rat cells have no meaningful understanding of anything at all. The same exact reactions they got from the neurons that made the aircraft fly straight and level could equally have been used to make the flight simulator do perfect barrel roles. The "calculations" done, if you want to call them that, are no different from finding a funny mechanism in which you can push here and it pops out there. The researchers are then applying that relationship between the pushing and the popping to do something else with. In this case, to fly an airplane simulator. It could equally be controlling a valve in a chemical processing plant or traffic signals in London. I have to conclude that these cells can't have any knowledge whatsoever of what they're doing which is very much unlike actual brain cells.

I wonder if a tiny portion of conscious person's brain could similarly be set up to do this without the person's knowledge of what's going on. Lots of crazy extrapolations one could make from this.