ben on agi ness

ben goertzel on ai/agi ness

via tweet [https://x.com/bengoertzel/status/1853873397132103865]:

I am seeing constant confusion these days about the basic meanings of terms like AGI, ASI etc.

I don’t delude myself such confusion can be easily remedied but as the one who inflicted the “AGI” concept upon the world way back in the Dark Ages, I can’t resist posting a simple clarification of what these terms generally mean to folks in the AGI research world…

Bearing in mind of course these are still somewhat fluid, cutting-edge concepts on which there is no consensus…

AI — any software or hardware doing something that people consider smart when people do it

AGI — Artificial General Intelligence — AI that can generalize imaginatively and fluently beyond its training data and its programming.

intellectness as cancerous distraction we can’t seem to let go of.. there’s a legit use of tech (nonjudgmental expo labeling).. ai as augmenting interconnectedness .. to facil seeming chaos of legit global detox leap.. for (blank)’s sake.. and we’re missing it

[There is a math theory of AGI, e.g. Marcus Hutter’s book “Universal AI” and a lot of other papers, but it’s pretty general and doesn’t give you specific practical AGI tests. This math theory does make clear that humans are pretty low on the theoretical scale of potential levels of GI. ]

[ There are also practical tests like Francois Chollet’s AGI-ARC that try and partially succeed to measure general intelligence in a way that spans human and AI… humans do well on them and current AIs including all LLMs do very poorly.]

HLAGI — Human-Level AGI — AI that has cognitive capabilities overall similar to a normal adult human, including human-like ability to generalize imaginatively beyond its training data and its programming

ASI — Artificial Superintelligence — AGI that is massively beyond the human level in general intelligence.

[ASI is expected to come with the ability to self-analyze and self-modify to a level far beyond HLAGIs … so ASIs will almost surely be way superhuman in building new ASIs …]

[Note also that even after we have ASI, humans may have some unique value to offer, just as the plants in the Amazon and our gut bacteria have some special emergent collective properties that are valuable to us even though our general intelligence presumably exceeds theirs.]

… and what comes after ASI?

Ultraintelligence? AI that masters paraconsistent closed timeline loops and turns large portions of multiverse into physical instantiations of its post-quantum mindstuff?

… TBD …

________

_______

_______

_______

______

_______

______