GOLEM / QOLEM - simple models of subjective machine states
The GOLEM (Goal Oriented Linguistic Emulation of Mind) is a (comparatively) simple model of human cognition, or thought. It has four quadrants, consisting of two information channels, the input channel which manages sensor-side percepts and the output channel which manages motor-side concepts. Entities (data structures, basically) on the input side are divided according to phenomenality, into conscious (C) and unconscious (U) categories. Entities on the output side are divided according to governance, into voluntary (V) and involuntary (I) categories. Figure 1a depicts this simple system-based I/O model, which has four sub-channels.
Because the C/U and V/I binary divisions require few assumptions to be made, we can use them to further develop the phenomenal properties of this model. By definition, all information (U) must pass through the input channel AND the output channel.
U = input AND output = (C OR U) AND (V OR I) =..
since these are Booleans and follow both associativity and distributivity so we can use the FOIL (first outside inside last) binary pairing rule, so...
..= (C AND V) OR (C AND I) OR (U AND V) OR (U AND I)
This second form of the equation defines four quadrants of the qualiate GOLEM, or QOLEM- see Figure 1b
a b
Figure 1
The simplicity of the GOLEM/QOLEM model is continued by its neurodal form (a neurode is an idealised model of a real neuron), depicted in figure 2a. In this model, the same type of feedforward neurode is used in both input and output channels. In the input channel there is considerable convergence (fan-in) as we ascend the layers of the channel. In the output channel, however, there is divergence (fan-out) as we descend the layers of the channel.
These neurodal behaviours are possible because the axon (output branch) of each neurode can spread multiple copies of its output signal. That is, the idealised neurode is a multiple-inputs/multiple outputs modelling unit. For reasons that are outside the scope of this section (see William James' work), we adopt common coding, ie we represent BOTH outputs AND inputs in terms of perceptual units only. This modelling decision has two useful consequences-
(i) both inputs and outputs can be stored as percept categories, or 'types'. We observe that this is indeed what brains do- see the cover graphic on the HOME page.
(ii) the neurodal outputs can be recycled as inputs (global feedback)- see figure 2b.
