What about the details of Edelman's books? Of course, they are now a bit antiquated, but I think people interested in making models of brain and mind can still learn a lot from Edelman.

My personal interests are centered in the area of learning and memory, and Edelman was on the right track in his approach to embedding memory in neural networks. Edelman gives a short historical review of the idea that memory storage in brains involves changes in synaptic function (Neural Darwinism, page 179). There has been a major revolution in the past 20 years in our understanding of molecular mechanisms (I point to some introductory references on memory mechanisms from here) for synapse modification, and this is still an active area of research. Edelman constructed his neural network models around some biologically motivated rules for how synaptic function might be modified by on-going synaptic activity. There is nothing magical about the connection strength modification rules utilized by Edelman in the network models he mentions (Neural Darwinism, page 183). There are many such possible sets of rules. The point is that Edleman is investigating what such rules can accomplish in a neural network model. He shows that they can accomplish a lot. The point is not that Edelman's approach can accomplish more than other types of neural network models that use non-biologically motivated connection strength modification rules such as back-propagation. The point is that we can make successful network models based on the real biology of synapses.

Who cares? For non-biologists (AI researchers) all I can say is, "Hey, maybe a billion years of biological evolution has found some tricks that you can use to make a better network." As for cognitive scientists who actually want to make models of minds that are rooted in models of brains, I say, "Pay attention here! Do not just take some easy to compute neural network method from the world of AI and apply it to your problem. Try to use a more biologically sound network method, you can then participate in the construction of models of biological minds rather than artificial intelligence minds." For the biologists who are involved in figuring out the molecular memory mechanisms of synapses, I say,"We can use computer models to organize the vast amount of experimental information that is accumulating and put it into its rightful context."

Beyond the level of what Edelman calls "local maps" (for example, his models of patches of cerebral cortex.....most people might call these local network models) is the issue of what he calls "global maps". I agree with Edelman that models of real-world cognitive issues need to involve many brain regions. Edelman stresses the importance of there being a diversity of synaptic mechanisms (Neural Darwinism, page 203), but I would also embed this in the context of a diversity of brain regions with different specialized functions. We need models of cognitive processes that link together many sub-models that utilize a diverse range of network connectivities and synaptic modification rules. I would argue that brain evolution has involved constructing many different kinds of local networks in small regions of the brain and exploring the search space of ways to connect these brain regions together so as to allow for useful functions like memory storage and motor control. This seems like a reasonable research program for us to emulate. Within our heads is a functioning mind machine, lets learn as much as we can about how it works and apply what we learn to our computer models.

go back to the first Edelman page

Go to John's Book Page.

Go to John's Home Page.