The way AI is progressing is
quite interesting, and ofcourse kept me riveted for last few years. It’s not
really about AI; whatever concerns humanity concerns me and I am quite
habituated to get into depth of things. It’s something I always find exciting. Having
wide range of knowledge without any concern for created boundaries helps to
probe. So, no knowledge is waste. Childhood spent in reading encyclopedia for
random knowledge stimulation is how I got into this frame I guess, and ofcourse
keenly observing insects/ants for hours is another matter that creates network in
impressionable brain. So, all the cynical view on AI -specifically in context
to LLM models/GPT, apart, these progresses are noteworthy in many ways. Ofcourse
one should be beware of hype, also be aware that this science fiction hype of extrapolating
reality has steady enthusiastic milieu of racism and violent narcissism pegged
in virtual games.
As we strip the hype, we find
something immensely significant transformative happening (much of which is in
AI’s narrow/discreet setting leading to halicin or protein folds so on which were
discussed over the months in this blog). LLM is ofcourse limited in its potential
for AGI to pose a threat, and was caricatured in my last blog as Andavar! (its Tamil
reference for god or godlike). Though LLM is no AGI but it has come a long way for
AI hence can accelerate programs towards AGI. LLM iterative learning has
limitation as it built from very general dataset sourced from whole of internet, transforming
that through neural network doesn’t really represent intelligence, it’s quite
shallow and really at surface level. Significantly with its incremental iteration
it cannot fathom breakthrough creativity, ideas or inventions. Quite clearly human
like generalized intelligence needs a technological breakthrough of sorts, and ofcourse
Turing test is quite a low benchmark for AGI. What is clear is that AI has much
superior computational capacity than a normal human and can detect pattern that
humans are incapable. The question here is the necessity to put such rigorous definition
of AGI. Indeed some define AGI as complex systems maintaining its boundaries
while seeking to transcend for new functions. This then is more realistic and
less of science fiction hype. It is here that LLM models show intriguing
ability of GI, and is able to make some leap beyond training data (ofcourse it
also makes equally stupid extrapolations) that is not fully understood and
makes the endeavor at an entirely different level of reckoning. There is a
very limited level abstraction that creates theory of mind is being speculated.
Progress in AI is moving towards
hybrid systems that brings neural network and evolutionary algorithms with
logical systems wherein LLMs is translated into structured logical statements
capable of abstraction (Ben Goertzel). So, unlike the present LLM, which
fixated references from past, it can reason whether its thoughts and statements
are consistent and logically coherent. In human brain’s cognitive architecture formation
of concept and self-concept shows similar pattern. So, it is suggested self/agency
or conscious would emerge from abstraction.
Interesting though these potentials
are the real impact of AI is seen in much prosaic relevant goals like
raising the living standards of people and sustainable goals. This is what the
book Human Compatible by Stuart Russell is discussing. With AI “there would be
no need to employ armies of specialists in different disciplines….every one
could have at their disposal an entire organization composed of software agents
and physical robots capable of designing and building bridges, improving crop yield,
cooking dinner, running elections, and doing whatever else need doing”. Education
is set to be revolutionized. “With AI tutors, potential of each child no matter
how poor can be realized”. In the area of health AI systems can “gradually
banish all diseases”. Anyone could have “high powered lawyers, accountant… on
call at anytime” so on. Potential of AI is immense, and is set to change the
world in a significant way. For systemically discriminated society with nonexistent
community values India has appalling education and health standards it is expected
that AI creates a compassionate space that is accessible to all.
