Some Reflections on AI: Part 2 — The God in the Machine

by Jonas Åsberg

Jonas Åberg argues that AI lacks true understanding, operating only with formal signs and fixed logic, while organic intelligence alone translates meaning across levels.

Read part one here.

Translated from the Swedish and originally published here.

2.1

Language, or more broadly signs, constitute the logical point of contact between organic and artificial intelligence. Both forms of intelligence, if we may call them that for simplicity's sake, use signs, i.e. discrete units that can stand for something (else), which can be combined in a regular manner into compound expressions and sequences. Organic intelligence can use signs but does not need to. Intellectual activity in a living being is possible without access to signs and language. (This depends on the fundamental autonomy of its life processes, i.e. that it interacts as a functional whole with the environment.) For electronically processed digital algorithms, or in other words computers, the opposite is true. Regardless of whether a computer can be characterized as AI or not, it always uses signs/sign systems in some form, and must do so. Signs thus constitute both a point of contact and a dividing line.

A process of understanding, as described in Part 1 of these reflections, may certainly use signs and language, as when we read a book or navigate traffic signs in a foreign city, but it does not need to, and the most important and fundamental processes of understanding do not. These include the processes that constitute a prerequisite for sign use, such as the ability to identify things, to see something as something. This is the main reason why language does not work as a criterion for intelligence. The signs that understanding uses are not formal, unlike the binary signs and sign sequences (codes) in a computer program, but representational or symbolic. A purely formal sign cannot be understood as such because what makes a sign understood as a sign is its conceptual content or what it stands for, by which is here meant something other than the sign itself and its external position in relation to other signs. It is only signs of this content-bearing type that contain two distinct and independent levels. It is thus only signs of this type that can be understood and that can be used for understanding.¹

Conversely, a sign-based process can use understanding (of the signs, i.e. of their content or what they represent) but does not need to. That a sign-based process can function without the signs having any independent content is a prerequisite for digital automata and thus for computer programming. By digitizing signs and combination rules (≈ language) and making them processable in this form (≈ thinking), one can liberate or separate them from their content and program them to perform tasks in this new form.