Neocybernetics
The basic idea of the homeostat and cybernetics is simple. It is exemplified by the biothermostat used by warm blooded animals. There is an optimum temperature at which their metabolism operates, and any temperature differences, either higher or lower, than the ideal 'set point' are used to automatically trigger corrective actions. Typically, shivering is used when the core body temperature (at the hypothalamus, usually) is too low, while sweating is used when the core temperature is too high. This basic idea can be applied to a wide variety of non-biological systems, such as home heating and cooling (a.k.a. 'air-conditioning') systems, for which the term 'thermostat' was first coined. A complex, sophisticated approach can be used in which the analog value of the difference between the actual temperature and the thermostatic setpoint can be used to control the exact amount of heating or cooling delivered, but an even simpler system is also commonly used, in which the setpoint is used as a an action trigger or threshold. If the actual temperature exceeds the threshold, then the cooler turns on. Alternatively, if the only ‘plant’ [1] is a furnace or heater, then the threshold switch is used to turn it off.
From this basic idea, some enhancements can be made. By varying the set point in discrete increments around the core value, we obtain a mechanism which can be used as a 'cruise control'. By allowing the set point to be varied smoothly, we obtain the function that forms the basis of 'fly by wire' systems.
However, if we allow the set point value to adopt any value at all, even values far removed from the original homeostatic level, we have created a way to use the vanilla feedback homeostat to implement feedforward dynamic governance. By using a homeostat to control an intrinsic variable such as temperature, we obtain a thermostat. The idea of an optimum temperature at which a system operates is not hard to grasp. But if we use a homeostat to control an externally defined (non-intrinsic) value, such as spatial position, there is no longer any meaning behind the 'set' part of the set point. It changes from an intrinsic set point that implements static governance to an extrinsic target point that governs dynamic position change ie motion.
The technique of augmenting static feedback control (setpoints) with dynamic feedforward command (target points, and guidance trajectories) is a completely novel method when it is used within a theoretical AI framework. In GOLEM theory, since these methods occupy a central explanatory (as distinct from computational) role, they have their own special name -Neocybernetics. [3]
1. not a decorative bush in a pot of dirt but the generic name for any machine, group of machines or industrial apparatus - go figure!
2. Kalman Filters are another manifestation of this concept, where feedback control and feedforward command are orthogonal aspects of the same overall idea, whether it is called guidance, direction or (as in GOLEM theory) 'goverance'. In calculus, the process of differentiation (obtaining the slope) is conceptually associated with orthogonalisation, both abstractly, and in a more concrete manner, if the functions are circular (eg sines and cosines), since f'(sine) = cosine and f'(cosine) = -sine.
3. Note that some neocybernetic ideas are used by Anatol Feldman to debunk efferent copy theory. He and I arrived at them quite independently, which adds well-earned credibility to both research efforts.