Neuron Bit Energy

In fairness, the brain is a reaction machine, most of the information is stored as trained synaptic connection. Neural energy is expended within a framework that already contains predefined courses of action that are optimized for productivity and survival. Within that schema, a collection of neurons making a decision and generating muscle stimuli is using those bits far more efficiently (but with much less immediate flexibility) than a general purpose computer.

A better analogy would be a silicon chip that can grow its own gates and wires. A gate array can simulate changing its own wiring, but is far less efficient per bit processed than hard wired logic. A system that could actually move physical wires and gates around to adapt to tasks would be far more efficient per bit than any existing general purpose electronic digital processor - and we call those systems ASICs (Application Specific Integrated Circuits) - moving the wires is a design and fabrication process that takes years.

So - assume that neural flexibility provides a factor of 1000 advantage over general purpose bits, so that the Shannon efficiency is closer to 2e-3. Assume a nano-engineered system that can (after much engineering) escape Darwinian constraints can get close to 2e-2, in a context of continuously "rewire-able" connection systems. A 60 Kelvin system can probably do as much with 0.4 watts as a 300 Kelvin brain can do with 20 watts.

For comparison, see this recent Intel paper about integrated circuit logic energy:

NeuronBits (last edited 2016-07-16 18:44:41 by KeithLofstrom)