As activity passes through the brain, it changes the structure. From the point of view of the vast forest of neurons in your skull, the organisational problem is tremendous: the nervous system must physically change itself to optimally reflect the world in which it's embedded. The individual changes must each make the right contribution to the network to embody the new knowledge, and the changes must be positioned to make a difference to behaviour when the right moment arises, some time in the future. A simplifying error when thinking about memory has been to assume that it is underpinned by a single mechanism of change. The classical story of strengthening and weakening synapses has brought us a long way, and artificial neural networks employing those principles can perform impressive engineering feats. But memory is more than dialling synapses in a large connection diagram. As we saw, simple synaptic models quickly lose the capacity to represent old data as new data stream in. The way memories degrade— with older memories having more stability — reveals the secret of different timescales of change. The synaptic model would be convenient for neuroscientists and Al engineers, but it's almost certainly not natures approach. Instead, the changes underlying memory are distributed widely over titanic. numbers of neurons, synapses, molecules, and genes. By analogy, just consider how the desert remembers the wind: it does so in the slope of its sand dunes, in the shape of its rocks, and in the evolutionary pressures that carve the wings of its insects and the leaves of its plants.
David Eagleman, Livewired















