abundance of silicon in the Earth ' s crust and the atomic simplicity of transistors when compared to chemical neurons presents a more parsimonious solution for evolving a thinking entity . Yet the fact is that we do not find these structures anywhere in nature , which lends weight to the idea that the ultimate ontology of thought might not be fully reducible to computation , which is what would be necessary for Zuboffs ' silicon-chip / neuron replacement ' or Bostrom ' s ' whole-brain emulation ' to be realistic possibilities , that unfortunately I do not have the space to fully critique here . Nevertheless , these ideas fall prey to what I consider to be category mistakes - that chemical neurons might be fully modelled by digital algorithms running on electronic microchips , or that we can nondestructively measure and replace both individual neurons and their entire connectome while the brain is on — all in reality may be impossible . Invoking non-Turing machines is still limited to science-fiction , not empirical science nor sensible , well-grounded philosophy .
To reiterate , the rule-following ability of the digital computer is rigidly fixed and can only be deterministically followed in a predefined way . Hence , no matter how complex the analogs of ' neural-networks ' or ' generative adversarial networks ' the connectionist might create with their computer languages , all must ultimately compile down to the logical ones-and-zeros of machine code used to electronically switch transistor gates on-and-off , to represent strings of bits composed into binary encoded numbers . While these numbers are subsequently recomposed to form higher-and-higher levels of abstract software structures , that nowadays perform so rapidly that they appear to us to react in real-time , because of this fundamentally symbolic underpinning , they nevertheless remainunthinking . Indeed , if the deterministic logic of the process we desire to run on the microchip is not perfectly described , we will get an error , or the programme will grind to a halt or attempt to compute its algorithm endlessly without ever outputting a result — the Entscheidungsproblem that Allan Turing originally created his universal computation machine to prove ( Turing , 1935 ).
This decision problem is not a problem thinking minds can ever have -we would simply say " I don ' t know " and know that we don ' t know or that we should stop going on . Hence one important function of thinking — logical intelligence - can be externalised into a machine , but that abstract logical structure in itself is not thinking , and so thinking is a mistaken label to apply to the machine . It took the higher-order metacognition of an actually thinking human to consciously selfreflect on its own thought in order to produce the now-automated intelligent process , yet that process does not possess the same selfreflective higher-order degrees of freedom as the original thinking mind .
This serves to reveal to us what is important about thinking , both as a process and how it should correctly be used as a verb — that thinking is not merely rational intelligence amenable to Boolean systematising , but is necessarily tied into what we consider as a consciously alive and subjective mind - one able to self-reflect , to know that it knows , understand that it understands , spontaneously conjure up jokes and non-sequiturs , chew over contradictions , cheat at chess , and at a deeper level , feel what it is like to be the doer of all those things .
Yet these things are what we would demand from our thinking machines — that they must have what is labelled ' Artificial General Intelligence ' or ' Strong Al ' and achieving this
SCHOOL OF THE ARTS 2021 / 2022
146