Методологический семинар ИСИ СО РАН

The recent revolutionary successes of the large-language models in AI suggest that we are well on the way to understanding human intelligence and representing it computationally in artificial systems. And yet these systems remain far from capturing true human-like understanding. By taking a normative evolutionary approach, with the premise that an examination of the key earliest hunter-gatherer activities – e.g., food preparation, shelter-building, tool manufacture – will help uncover the fundamental cognitive architecture of the human brain, my analysis has identified a core set of abilities, including causal reasoning, metacognition, affect, metaphor, and a substitution mechanism not unlike the mirror or empathy mechanisms. I will sketch out how they were derived and how they all appear to work together to underpin human higher cognition. Importantly, these abilities remain lacking in artificial systems – and yet, it is argued, will be required to reach human-like general intelligence. Thus, the findings can be considered a roadmap and target destination for future-generation human-like artificial general intelligence.