DeepMind says reinforcement learning is ‘enough’ to reach general AI

When I read abstracts like this: "Cybersemiotics constructs a non-reductionist framework in order to integrate third person knowledge from the exact sciences and the life sciences with first person knowledge described as the qualities of feeling in humanities and second person intersubjective knowledge of the partly linguistic communicative interactions, on which the social and cultural aspects of reality are based. The modern view of the universe as made through evolution in irreversible time, forces us to view man as a product of evolution and therefore an observer from inside the universe. This changes the way we conceptualize the problem and the role of consciousness in nature and culture. The theory of evolution forces us to conceive the natural and social sciences as well as the humanities together in one theoretical framework of unrestricted or absolute naturalism, where consciousness as well as culture is part of nature. But the theories of the phenomenological life world and the hermeneutics of the meaning of communication seem to defy classical scientific explanations. "

My confidence isn't increased that this book has anything interesting to say. It sounds like the kind of post-modernist verbiage coming out of a lot of humanities departments. I mean seriously, why anyone in a STEM field actually write a sentence like "modern view of the universe as made through evolution in irreversible time, forces us to view man as a product of evolution and therefore an observer from inside the universe." Like what is "irreversible time" adding there? It's kind of redundant, and the awe-inspiring implication: man is a product of evolution, and therefore an observer in the universe? Earth shattering. And all phenomena, from fundamental life sciences to culture, are part of nature? No one's ever considered that before.

My guess is, if you asked a STEM major to write this abstract, they could sum it up with 70% less words. They'd also drop implicitly obvious claims that no one disputes.

I'm betting that this is book is just a less philosophically rigorous version of Searle's argument, given the implication that emergent phenomena "seem to defy classical scientific explanations".

Intelligence is simply a special side-product of evolution, there is nothing general about general intelligence. Many organisms can thrive without it.

There is also a non-negligible chance that all organisms would die out before reaching intelligence. We are fortunate to live in a world that produced us.

However, this "result" is trivial. It is obviously equivalent to the claim that intelligence arose naturally in the biological world without influence from God.

I'm not sure that's true anymore - pretty much any objective devised is being solved by ML solutions within months (with some exceptions such as Chollet's ARC, maybe Winogrande). But those same models will perform poorly on other unseen tasks, because ML takes shortcuts if it can. We used to have unsolved tasks for decades, such as Go. It's now comparably hard (if not harder) to create a good objective measure of intelligence than to reach human parity on said measure.

Personally, from reading the abstract, I disagree with the hypothesis. There's a trick where anything (even say, a database lookup) looks like optimization as long as you contrive the objective function just right, but that's kind of uninformative.