[ExI] AI

[Ben Zaiboc]

These two posts from Digest Vol 263, Issue 7:
>     6. Even social media with only AI users becomes dysfunctional (BillK)
>     7. Self-evolving AI (BillK)

Made me think.

I've always imagined that self-improving AI meant machine intelligences 
becoming more intelligent. Now it seems that it doesn't. It appears to 
mean large language models becoming more and more skilled at producing 
text that polarises human opinions and generates outrage and conflict, 
by pandering to existing prejudices and reinforcing circular reasoning.

This is not 'intelligence', by any sane definition.

Perhaps we need a different model for AI.

Or can we separate the news-worthy type of AI from the genuinely useful 
type, that should be, by now, helping us to accelerate scientific 
knowledge, technical ability and develop actually intelligent systems?

If they exist.

Looking at the average human being, it's no surprise that these 'AI's 
are getting better and better at manipulating people. But where are the 
AIs that can manipulate physics, chemistry and biology? Even LLMs should 
be able to be created that work in the languages of these disciplines 
(laws of physics, etc.).

Is this just less interesting to the average human, so we don't hear 
about them (another filter bubble)?

Even so, where is the evidence? SpaceX still has 'rapid unscheduled 
disassembly' events, Molecular Manufacturing (what used to be called 
Nanotechnology) is still a dream, Malaria still infects millions of 
people and kills hundreds of thousands annually, Fusion power is still 
as far away as ever (that may not be literally true, but we certainly 
don't have it yet), understanding of our immune systems still advances 
at a snail's pace, I haven't heard of any advances in stem cell 
manipulation for ages, there are still lethal cancers, the standard 
model of particle physics is still incomplete, we still pump 
hydrocarbons out of the ground instead of making them from CO2 in the 
air, nobody has a personal jet pack and the best answer we have is still 42.

It seems to me that 'AI' has become another term whose meaning has 
changed from the original intention. Do we even have any artificial 
intelligent systems at all? Any research projects left that haven't been 
hijacked by the LLM paradigm?

Still, at least we have Mr Beast, eh?

-- 
Ben