Greatly enjoyed reading that, thank you! One thought, I wondered if you could elaborate what alternatives are there to prediction being the ‘essence’ of intelligence? Would you consider being reactive as an appropriate antonym and alternative? (As in backward looking rather than forward looking.) And if so, do you see current machine learning as predictive or reactive?
As a biologist, when I look at nature I see a world full of myopic intelligence without much prediction. Typically, animals respond to their environment with ‘plasticity’ — changing an instructional change in phenotype — rather than using the information to create some new insight. The instruction has its origin in some previous act of creation, usually from blind selection (Dawkins 1982, Cronin 1991). Prediction, to me, speaks of this creative use of existing information to create new knowledge, which is a higher cognitive function that we rarely see in nature. Even when a corvid or chimp is set a problem, they often do so using past experience rather than with ‘insight’ or ‘intuition’ to tailor their experience to the problem at hand. Prediction, to someone with my background, is something that is rarely found in biological systems because ‘prediction’ means something more than ‘reaction’. And yet, colloquially, ‘intelligent’ behaviour is often equated with the ability to solve problems — even though they can be solved unintelligently. (This is what Dennett (e.g. 2017) refers to as the strange inversion of reasoning (here and many times previously), which he credits in different ways to Darwin and Turing.)
I know the biological perspective is not what yourself or Yann LeCun were talking about. But I wondered if you had any thoughts? Thanks!