Quote from: Jonas Bee on January 13, 2025, 17:40:45Going with the assumption that AI has indeed exhausted real-world data (which is yet-another unsubstantiated claim by an loud business mogul rather than someone actually compentent in this field)
Right.
***
In the game of Go, earlier AI versions relied on human-generated data (played games). Afterwards, AI has relied on self-learning from AI-generated data (AI self-played games) to become much stronger (very clearly above human-level AI player).
In restricted domains, such as Go, this can work especially if the domain is deterministic, complete information (and, maybe not necessary, 0-sum).
In unrestricted, non-deterministic domains with incomplete information, self-learning might be harder but still possible. However, a key issue is the relation between AI, mankind, all other biological species and our habitat including the earthbound biosphere.
Firstly, AI might model reality or create fiction, such as a world without, or with different kinds of, living beings.
Secondly, besides operation of the AI itself while consuming energy etc., particular kinds of AI on its own or by (typically) human queries might, or might not, physically affect mankind, other living beings and our habitat.
Thirdly, for kinds of AI physically affecting mankind, other living beings and our habitat, ethics, laws etc. might be applied, violated, ignored or replaced by fictional, AI-created standards. E.g., AI might create the standards to enable its own independent life or to extinguish mankind.
Therefore, letting AI learn on its own from scratch is an insufficient aim. Instead, it must also be guaranteed that our ethics, laws etc. are maintained and mankind survives the age of AI. Elon Musk just speaking about self-learning AI without also considering ethics, laws etc. is irresponsible. Politicians (not only in the EU) ought to regulate, and enforce application of regulation of, AI before everything that can be done is done. At least, he reminds us that AI progress is fast.