Non-traditional (Genetic Algorithm) Circuit Design; Unconnected Neurons Exerting Influence on Neighbors

A programmable chip (FPGA) loaded with the right program (code ‘evolved’ by an overseer measuring task performance to choose winners, with random sexual combination of code items in a generation) can reliably classify an analog signal with a far simpler design than any human-constructed approach.

Unfortunately, the code may only work for the particular physical substrate (programmable chip) it was evolved on, in that it exploits idiosyncrasies in that particular chip.

One particular winning program produced disconnected loops, that nonetheless were influenced by electromagnetism at distance from nearby circuits, and were essential to task performance.

You could still mass produce such devices by running the evolution on each somewhat-unique chip, seeding it with successful codes from other chips.

They use no clock signal; making changes only happen on transitions of a global clock signal is one of the ways abstract, designed programs can execute reproducibly in spite of small variations in the manufactured substrate (it’s also possible to design clockless circuits that converge reliably, but that’s not as often practiced).

The human brain almost certainly enjoys such efficiencies (compared to rigid, computer-programming-like design). Just a few weeks ago Caltech researchers found actual evidence of changes in neural activity from influences in disconnected neurons in proximity (in rats), under normal (not epileptic) conditions - ”Ephaptic coupling of cortical neurons. I don’t know if this mechanism will turn out to be essential to modeling our brains’ computation. This is definitely analogous to the useful disconnected circuit in the winning evolved FPGA design; if the learning reinforcement is supplied reliably by other already-working parts of the brain, neural ‘circuits’ could even arise by an analogous mechanism.