GOLEM is a finite state machine
Posture
Like all modern computers, as well as Alan Turing's canonical Universal Computing Machine, the GOLEM is based upon the finite state machine (FSM) model of computation [1]. GOLEM states are sets of numerical values, each of which stands for a somatic configuration unit, eg a joint angle. You are probably reading this while sitting in a chair. When you get up to make a hot drink or stretch your legs, you can feel your body change its overall configuration, ie its combination of joint angles and other proprioceptive factors. Some of these configurations (each one represented by unique combinations of angles) are more commonly used than others. These frequently visited configurations describe our individual bodily posture. We learn which ones are most comfortable, and use least energy. For this reason, they have been called equilibrium points [2].
Behaviour
Very generally, one's knowledge is a semantic property of brain state. Animals (including humans) use their knowledge to generate and govern behaviours which make best use of the situations they find themselves in. At first blush, there doesn't seem to be much in common between posture and behaviour, but this impression is a very wrong one. Instead of using a separate mechanism to generate dynamic actions, the GOLEM uses the bodies static posture system to generate them.
The GOLEM consists of two channels, an input channel and an output channel. The input channel contains a hierarchical data structure representing the current brain state of the animal, while the output channel contains a hierarchical data structure representing its current hierarchical brain state transition - basically, its proposed behaviour plans. A simple behaviour plan is almost trivial to prepare- all that must be done is compute the difference (= state transition) between the planned (next) configurational state and its current set of values.
The concept of a homeostat is used to explain how living systems self-regulate at all levels. This concept applies fractally [3], ie it is observed at both microscopic (at the cellular level) and macroscopic (at the system level, as in the diagram below) structural size scales. Systems which use 'standard' or 'canonical' homeostats as building blocks are called 'cybernetic'. They are represented by the red, broken line in the figure below.
The GOLEM uses an augmented form of homeostat (called a heterostat) to generate dynamic behaviour [4]. Note that the term 'heterostat' possesses the suffix -stat, meaning it describes multiple static (ie steady-state) mechanisms. The term 'neocybernetic' has been used to describe cybernetic systems which generate dynamic modes of action using heterostatic methods.
In the figure below, the 'standard' setpoints (regulated values) are not fixed, but are themselves variable. When they are varied slowly (ie tonically, to adjust system 'tone'), they are referred to as 'setpoint biases'. It is these bias values that are changed (involuntarily) when we adjust our posture to reflect changes due to aging (for example). When setpoints are varied quickly (ie phasically, to reflect the phase of a given multi-phase action sequence), they are referred to as 'setpoint offsets'. The term 'offsets' is a deliberate mnemonic, meant to connote the use of pointer (memory addressing) arithmetic, as seen in system programming languages such as 'C'. [5]
The neocybernetic architecture is notable in that it uses computations that are of maximal simplicity - everything is achieved using only scalar additions and subtractions These are directly comparable to the kinds of pointer arithmetic manipulations used in modern digital computers. It is therefore fair and accurate to describe the brain as a computer at both high and low levels of abstraction.
The main benefit (there are many other subsidiary advantages) of the GOLEM model, is that it is both bioplausible and neuroanatomically realistic. The state-transition paradigm fits GOLEM's two-channel architectonics as well as its neocybernetic functionality. Consequently, the two-channel template can be used to reason about primary and secondary (and perhaps, higher order, including those related to external language) notions of modelling selfhood.
1. It is the hypothesis of this research project that the GOLEM is an accurate, plausible model of one brain hemisphere.
2. Feldman, A.
3. I am using the term 'fractal' in an informal way, meaning at multiple size scales, not formally, with reference to the Haussdorf metric.
4. 'Dynamic Behaviour' is (obviously) a tautology, which is used deliberately to labor the point, ie as an educational device.
5. The term 'bias' is also a mnemonic, albeit somewhat more subtle, intended to connote the use of 'base', as in [base address + offset].