Jayman; Yes (correct as usual): 1) Four AI categorical problems: – communication

Jayman; Yes (correct as usual):

1) Four AI categorical problems:
– communication(interface),
– world model(episodes),
– possibility (operations),
– morality (limits).

2) We had expected to develop in a different order: world model, possibility, morality, communication. However LLMs reversed the order by rapidly increasing the the ability of the AIs to learn with high error rates eventually reduced by falsification. This is the cause of rapid advancement in AI. Helping the AI learn by itself given the vast inventory and accessability of textual knowledge combined with human reinforcement for error handling.

3) There is no reason we cannot solve morality (in fact, we would have to teach it NOT to be moral if it’s learning by human reinforcement.

4) There is reason to believe we will go through another age of ‘conflict’ on the scale of literacy (age of monotheistic religions), the printing press (wars of religion), the industrial revolution (wars of empires), and the rise of innovation and adapttion in crime that preceedes each of these major evolutionar leaps.
This is going to be a nightmare for stock markets and disinformation and authoritarian propaganda on one end and seditious conspiracy on the other. But there is no reason we can’t shift the economy from emphasis on private consumption to directing displaced labor by AI to production of commons.
We might just repeat thepossibility of returning to a west at least, that like europe, is a vast open air museum.

Cheers

Reply addressees: @JayMan471