The arrows of AI run in all directions

Thanks, Prashant. I only want to draw attention to the dimensions in the Russell-Norvig framework as arrows extending in all directions, as opposed to terminal points or boundaries.

For example, the AI-as-artificial-humans idea extends into a quagmire of thinking and behaviour that we don’t understand (e.g. environmental and metabolic influences, consciousness, irrational behaviours, emergent social phenomena). Similarly, rationality is a meandering path, whether you’re exploring various forms of reasoning (e.g. logical, probabilistic, causal, spatial, social) or epistemologies (e.g. common sense, induction, Bayesian, Popperian).

I think it’s wise that you’re focusing on the goal. The success of the prevailing conception of AI as rational agents illustrates the benefits of this type of pragmatism. I’ve argued for a narrow goal of automated scientific discovery. Goal-setting makes the problem of AI tractable by narrowing the scope and removing ambiguities.

And yes, you can certainly argue that dimensions of humanly-rationally, reasoning-behaviour can be used as a universal frame for AI. But given that those arrows run in all directions (and hardly straight lines at that), suggesting universality to such a limited framework is more confusing than clarifying, and leads to misconceptions.