Ann: Investor Presentation - August 2018, page-50

In the brain, neurons work on a threshold basis. Think of a synapse (the space between two neurons through which they chemically communicate) as a space between a child and its mother. The child wants to get the mother's attention, and does so by tapping her leg. If the mother is occupied, a single tap probably won't get her attention. She has a threshold before she will be "activated". Also, if she has recently responded to that child quite a few times, that threshold might go up and the child may need to tap her more times before she will respond.

Like that analogy, neurons in the brain don't respond immediately to stimulation. They require a certain accumulation of stimulation before they fire. That threshold changes in response to various factors. This is why if you stare at an image for 60 seconds and then look at a white wall, you see the colour inverse of that image - the prolonged stimulation of the neurons responsible for the colours you are looking at causes them to adapt and be less responsive to future stimulation. Thus, when you look at the white wall they under-respond relative to the other neurons that weren't being stimulated before.

So, I think that in computing the idea is current hardware would require you to allocate multiple bits in memory to track signalling to a simulated neuron. The nature of the hardware with WBT I suppose might be able to do this with a single bit at a hardware level: for example (and now I'm guessing how it would actually work), maybe a small pulse of current into a memory cell would not switch the cell's state from 0 to 1, but it might mean that the next pulse of current, if it arrives within a certain time-span, would have a greater chance of causing the state to switch. Perhaps 6 or 7 pulses in a millisecond would not be enough to switch to a 1, but 8 would cause it to switch. This could potentially reduce the hardware resources needed for a simulated neural network by orders of magnitude.