Equipoise can't be simulated
Returning to an old gripe. AI types intentionally misunderstand this point, and even the anti-AI folks don't fully get it.

Aside from bigger questions of soul and consciousness, it's just plain PHYSICALLY IMPOSSIBLE for software to imitate even the SIMPLEST nervous system. This applies to everything from jellyfish to humans.

Early efforts at cybernetics, especially in Russia, did understand this point, and would have made more progress if they hadn't been shoved aside by the digitalists.

Negative feedback and balance are INCURABLY ANALOG. All nervous systems are based on negative feedback and differential input. Equipoise. Feedback acts INSTANTLY, and the total function of the circuit depends on the feedback in a way that can't possibly be factored out mathematically or softwarically. Software always acts in steps of time, so the effect of the error signal can never be continuous and simultaneous.

Here's a neat example of a HUGE simulated neuron, built in 1965 by government scientists who understood equipoise.



This is a tiltmeter measuring the tilt of a large stretch of land in Colorado, in an attempt to monitor and predict earthquakes. The sides are 200 feet apart.



Detail of each side.


Intrinsic equipoise, built with water instead of levers or tubes or transistors. Balanced and differential all the way to the pen.

Note that the output is an analog decision, typical of real brain and mind. The researchers could instantly see and 'feel' the changes.

Since the development of GPS and lasers, we no longer do differential measurement. We take the current height of each datum, then process the difference mathematically. This gives more nominally precise answers, but it's subject to all sorts of interference, and depends on all sorts of systems that aren't under the control of the researchers. A hacked GPS system or a badly coded program can give wrong results with no way of knowing that the results are wrong.

Later thought: A few of the modern cybernetics types still understand the point and try to work in analog mode, but they don't get the grants. The grants and corporate support go to digital. Apply Polistra's Law of Intent, and presume that the OBSERVED RESULTS are what Deepstate WANTS. They're never clueless. They WANT to have a failed simulation instead of a functional simulation, because they DON'T want real intelligence anywhere. Meat objects are not allowed to use real intelligence, and electronic objects are not built to use real intelligence.

Labels: AI point-missing, Equipoise

  ¶ 11:39 PM