Brain Interface Lets Monkeys Control Prosthetic Limbs

himicos was one of many readers to point out one recent success of scientists working to develop working brain-machine interfaces, writing "A team at the university of Pittsburgh has finally advanced a 2002 technology enough for use in prosthetic limbs, the targeted application all along. Training computer models to the firing patterns of the neurons in the parts of the brain that control motion, they are able to project the intentions of a monkey to a robotic arm, which follows the will of the animal. The sad thing about the articles is that the beauty of the mathematics used to create and train the models is totally ignored." Reader phpmysqldev adds a link to coverage at the BBC, and writes "This of course brings significant hope to amputees and other other people with physical disabilities." (Note that this research has been going on for quite some time.)

Comments

Re:Explain the beauty?

[dezert_fox]

Shannon entropy has been a standard tool in data communications for a very long time--telcos use this math to make your phones work. It's effectively a way of quantifying the informational content of a signal, which can be used to determine exactly what kind of bandwidth you need in a bandwidth-limited environment. I'm uncertain what it's used for in the context of a brain-machine interface.
Any good data communications textbook would have some nice examples in it, and actually that wikipedia article posted is very readable and informative.

The Real Sad Thing

[DynaSoar]

"The sad thing about the articles is that the beauty of the mathematics used to create and train the models is totally ignored."

The sadder thing is that the discovery of response patterns of amputated limbs being mapped to other parts of the body is totally ignored.

A man had his arm removed. A psychiatrist attending happened to note that the man claimed to "feel" things in his missing hand when other parts of his body were touched. After careful mapping, three different response maps were found -- one each on his arm, chest and back. Each was so sensitive that individual fingers could be stimulated and he could correctly tell which.

This major discovery in neural plasticity makes it totally unnecessary to try to decode signals from electrode either drilled through the skull, or else placed on the surface and reading signals though the scalp, skull and dura mater, which reduces the signal by 3 orders of magnitude. Either way, these signals require some massive processing because a significant command/response signal (ie. an electrical response representing a single Hebbian cellular assembly that can be clearly decoded to an intent as stated in the article) comes from 0.3% to 3% of the neurons in the region being detected, the vast majority of the signal needing rejection as false positive or noise. Using the mapped response regions allows for signal analysis based on EMG patterns that are not expected at all in the area under the electrodes, making detection and analysis trivial.

TFA and most such research is not about giving amputees mobility. It is about decoding and using neural signals. If it were about the former, easier ways would have been used and the job already accomplished. It is about the latter because such things make more news, get more recognition, and therefore result in more grant application success.

The resulting technology will only be applied to prosthetics as a secondary result. Its primary use will be in such as hands-off controls for fighter pilots (see Clint Eastwood's "Firefox" for your obligatory Slashdot sci-fi/movie reference), tank crews and mobile missile launchers. Maybe this is the saddest part of all, but ignoring a more certain path to success as far as prosthetics is a sad piece.

Also sad, with a touch of irony, is the fact that the weaponry applications will be untenable because of the heuristic nature of neural processing -- getting it close but error prone will be fast, getting it right will be no faster or require less effort than hand operated controls. The slow speed and so the ability to use real-time perceptual feedback with prosthetics will make that far more successful. It remains to be seen whether after the war applications fail the research continues (ie. there is adequate funding offered) with respect to prosthetics. If someone like the US Veterans Administration picks it up when DARPA drops it, it might. I'm not hopeful.

The portion of the above that's assertion or opinion is based on the same professional experience as the portion that's not. That experience includes development of some of the "beautiful" maths decried as being ignored. Having been prosthetic-wrist deep in the research and from both directions, I find that a minor point to consider as "sad".