On “Machine Consciousness”

Consciousness in higher animals, with its 100-millisecond time constant, is a simplified version of more complex cognitive systems like wound healing and immune function, which emerged from information crosstalk between cognitive modules. This stripped-down nature suggests it should be possible to build a fast, single-workspace conscious machine mimicking the human global workspace. Tied to a backbrain AI for hyperrapid pattern responses, such a machine would exhibit elementary emotions. A designer might instead use high-speed electronics to create multiple-workspace systems less prone to inattentional blindness. The ultimate utility of such machines remains unclear, explored here through asymptotic limit theorems of information and control theories.

Study at a glance

Abstract

Consciousness in higher animals, by virtue of its 100 ms time constant, is a necessarily greatly simplified and stripped-down version of more complex multiple tunable workspace cognition/ regulation dyads like wound healing, immune function, gene expression, institutional function and the like. These more complex dynamic entities emerged through evolutionary exaptation of the inevitable information crosstalk between coresident cognitive modules. In consequence of the debrided nature of consciousness, it should not be difficult to construct a fast, single workspace “conscious machine” that mimics the human tunable neuronal global workspace system. Tied to a “backbrain” AI that has learned hyperrapid stereotypic pattern responses to some particular set of likely challenges, the result is an elementary “emotional” conscious machine. A clever designer, however, may want to use available high-speed electronics to mimic the more capable multiple-workspace/ workforce systems inherently less susceptible to inattentional blindness and related failings of overfocus and thrashing. Contrary to current social constructions, however, the ultimate utility of such machines remains obscure. Here, we explore these matters in formal detail, restricting argument to the asymptotic limit theorems of information and control theories.