[home]  our papers  ] Outline of our Theory ] About the Authors ] Links ]
This is the "Synaptic Darwinism" website of Paul Adams and Kingsley Cox, who work at the Kalypso Mind/Brain Center, and in the Department of Neurobiology at Stony Brook University.


"Synaptic Darwinism" is a general approach to certain problems of brain function. The starting point is the observation that both in neuroscience and in general biology one sees a similar process: the spontaneous development of complex structures which are well adapted to equally complex environments. Perhaps rather similar processes are at work in both cases. This does NOT mean that the brain processes information using DNA, though both DNA and brain hardware may employ similar logic. Several authors, notably Gerald Edelman, have proposed that the brain uses Darwinian strategies. But "Synaptic Darwinism" invokes a quite different, more specific, and more powerful, analogy. It proposes that the replicating entities in the brain are arrays of synapses. Instead of a particular polynucleotide sequence, one has a row of synapses. A connection is made up of many such functionally identical synapses, and as a result of the electrical activity of these 2 neurons, synapses can be added to the connection, or removed. These new synapses are replicas of the existing synapses, since they connect the same neurons. However, no replication process is perfect. An imperfect replica of a specific synapse is still a synapse, but it no longer exactly connects the same particular pair of neurons. The imperfect replica now connects one of the original pair to a neighbor of the the other member of the pair. These well-defined replication and mutation operations might ultimately be largely responsible for wiring up the entire brain, as a result of electrical activities caused by an animal's interaction with its environment.

There is recent electrophysiological evidence that events resembling synaptic "replication" and "mutation" do occur in the brain during learning - or, more precisely, in brain slices during procedures that mimick real learning. There is also recent evidence that mature neurons can form sprouts, which make novel synapses, during recovery from injury or even normal learning. If the same electrical events control both connection strengthening and novel synapse formation, the novel synapses are equivalent to "mutations". This view of synapse formation overcomes a serious limitation of conventional models of brain wiring. The bulk of the brain is composed not of computing elements (synapses and impulse-initiation sites) but of wires. If each neuron were connected to every other neuron, the wires would have to be of subatomic dimensions, and electrical spikes would have to travel faster than light. So the actual connections are just a tiny subset of the possible connections. Since the computing power of the brain lies largely in the detailed connections, either connections have to be pre-specified by the accumulated creeping wisdom of gene-based evolution, or they have to be constantly shuffled around and evaluated on-line. "Synaptic Darwinism" claims that most of the shuffling and evaluation is done rapidly at the level of synapses rather than slowly at the level of genes. This flexibility comes at a price. Experimental new connections degrade current performance. They may even proliferate until wiring becomes almost random. Thus synaptic mutation must be kept in check. One way to do this is to make replication more perfect, and this appears to have been a major theme in vertebrate evolution. Another is to only allow modification of connections across which electrical activity is appropriate. This would require rather elaborate circuitry, but this hypothetical circuitry corresponds remarkably closely to the most puzzling and characteristic features of neocortex. Quite astonishingly, a large neocortex undergoing Synaptic Darwinism must dream, sleep and, during the day, be conscious. Although the ideas and arguments underlying "Synaptic Darwinism" are plausible, in agreement with much recent data, and compatible with more conventional "neural network" approaches to brain function, they are still speculative.


Document made with KompoZer