You should look at the generalized recirculation algorithm, GeneRec, in O’Reilly and Munakata’s 2000 textbook. It essentially performs backprop, using only information local to each synapse. It gets information about error to each synapse by allowing the network to settle into a state constituting a “guess” as to the outcome, followed by a separate state that involves the correct output. Feedback and lateral connections ensure that the whole net is in a different state when correct output information is present. The learning rule then pushes weights toward the correct state, and away from the guess state. The learning rule at local synapses is very similar to the experimentally observed role of calcium influx at each synapse. Also, the algorithm seems to continue to work with very asymmetric connections, and would likely work without reciprocal connections between every individual unit.

This approach seems to solve the biological plausibility side of the error-driven learning puzzle. The remaining question, and one that must be answered by most forms of learning rule is this: where does the correct output information come from? I don’t think that anyone has fully thought through what information people actually use to learn from.

I looked a bit at what was happening in basal ganglia during cognitive control (in the study described in this post) and wasn’t able to make a whole lot of sense of it. The head of the caudate nucleus showed some (inconsistent) increases during cognitive control. I wasn’t able to find a clear interpretation, though I may need a dataset with a better encoding-period manipulation to better test O’Reilly’s theories.

One thing about O’Reilly’s model that has been unclear to me is how the dopamine/BG system received inputs in order to predict reward. Is there a feedback loop from prefrontal cortex, or do sensory stimuli directly impinge on VTA/basil ganglia?

I wonder if you’re familiar with Randy O’Reilly’s computational theory of cognitive control. It’s incomplete, and more work like yours needs to be done to extend our understanding. However, this theory is important because a) it works at a lower, more mechanistic level, and b) because it’s closely based on evidence from several different levels of research.

Just wondering what the rest of the community thinks of the type of work coming out of our lab. In the interest of full disclosure, I’m a postdoc in Dr. O’Reilly’s lab. I think the type of work his lab is doing is has an important role in figuring out how cognitive control works, along with empirical studies such as your own.

Hippocrates offered his philosophical opinion on this issue. I am not decided upon this.

This is the reason why I think you might be correct;

If the mind is not a direct product/implementation of neural activity then it must effect the brain all of the time - as the brain knows/remembers consciousness. (Obviously this action could not by definition be scientifically verified as it would lie within noise/physical probability of events). It is therefore simpler to assume the mind is a direct product/implementation of neural activity.

This is the reason why I think you might be incorrect;

Philosophically, what is the point of the mind if it does not do anything?

Thanks for your articles MC they are very interesting. I am working on neural network software myself and have found some very useful references.

Richard

[[Are we sure we know people’s dreams?]]

I think that Damasio’s arguments about the specific brain structures that support this (e.g. the exact source of “somatic markers”) is controversial in some circles, but I buy the dependency of decision making on emotion that he posits.

Igor.

I strongly disagree that anything but clinical application is a party trick. Basic science is extremely important for understanding the world around us, which is valuable in itself.

Additional value can typically be derived from basic research because of its usefulness to subsequent clinical research. After all, we must understand how the normal brain works before we can recognize the brain deficits present in abnormal brains.

MEG is too expensive for neurofeedback, but an EEG biofeedback, or an HEG hemoencephalography system can be just a few thousand dollars. There are many opportunities for MEG to settle some controversies with this community of clinicians and researchers who are, on a daily basis, reducing and even eliminating symptoms of autism, ADHD, migraine, seizures, etc. etc.

The main problem I see with TMS now is that it’s not going to just disrupt the region being TMSed. Most researchers using the method seem to ignore the fact that whole networks are going to be stimulated, since any given region in the cortex is connected to others. There is some hope, though. Concurrent neuroimaging may help with the problem, since we could see what’s being stimulated and make our inferences based on that. Alternatively, we could figure out with neuroimaging how regions are connected, then infer that TMSing region A likely also affects connected region D, which may be part of the story in disrupting behavior X.

Ultimately, all methods are ‘party tricks’ if not used correctly. Or, of course, if used to impress others at a party…