In my efforts to specify several different developmental and architectural
paths for general intelligence I pointed out that a defining problem for
epistemology (computational epistemology perhaps) and computer science is
to determine if and when a program/organism has the cognitive capacity to
"do physics", to decode its very own environment, as we humans think we are
doing with our universe. It is a recurring feeling in human history that we
have nearly completed the physics project, including recent Theories of
Everything, while my intuition is rather the opposite, that it is probably
impossible to decode the physical world and we are just children building
sandcastles. The intractability of these problems is usually exploited by
dualists that maintain that some "magic sauce" from another dimension is
needed to keep the intellect reverse engineering the world, while strangely
my intuition is more in the non-dualist direction (please note that physics
is epistemologically and practically different from general problem-solving
intelligence, but probably AGI-complete in its own right).

With this out of the way, I realized we do not need to keep speculating
about non-dualism, we can attempt to create the non-dualist, Advaita
Artificial Intelligence. The idea is to build software that attempts to
decode the physics of its virtual or otherwise world, while being embodied
in it. A variation would be to try to "understand" computer languages which
of course are much more like physics than natural languages. As I have
claimed repeatedly, creating AGI will give rise to its twin and nearly
intractable problem of translating from an alien language with very few
clues. Perhaps we would have to depend on visual confirmations of a
successful AdvaitaAI, where the measure of success would be to observe
simulations inside the virtual world that correctly predict the outcome of
virtual experiments again and again - a dream within a dream! And we would
be well advised to include surprising physical laws that are only rarely in
effect, but the initial challenge would be to get any virtual physics
decoded at all by an embodied agent. It could turn out that very simple
systems are undecodable, for example an "organism" in the Game of Life may
not be able to simulate any other pattern "internally" because of the
absence of light information in the game, the pattern would have to be
inferred by edge interactions which may be impossible or destructive
(please note that such virtual experiments in common parlance would take
place outside the organism, but from a philosophical and computational
point of view the virtual experiment whether it takes place in the "head"
of the virtual organism or not is more of coupled system - I am not sure if
this is related to the weirdness of the "observer" in quantum physics or
not).

The Game of Life was of course mainly supposed to resemble physical
multicellular growth patterns, perhaps an obvious extension would be to
equip it with a faster light signal (every generation that simply
influences the neighboring cell could also weakly interact with something
10 or 100 cells away too). Strangely I can see the benefits of our own
light's physics emerging, for example probabilistic interaction with cells
could allow the signal to travel very far, and changing the speed of the
"light" depending on density gives information that otherwise would be very
difficult to acquire. Interesting!

AT



-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com

Reply via email to