Comment by rramadass | Hacker News Reader

Comment by rramadass | Hacker News Reader

One good framework to think about the problem of "Consciousness" in Humans vs. AI/AGI is the Samkhya School of Hindu Philosophy - https://en.wikipedia.org/wiki/Samkhya

See my past comments here for some background - https://news.ycombinator.com/item?id=40479388

This is how i work it out;

1) "Consciousness" defined as attribute-less pure awareness/witness and immutable "Self" (aka Purusha in Samkhya) will never be possible in AI/AGI.

2) All other human mental states which are mutable and considered as evolutes of Prakriti in Samkhya are possible in AI/AGI. A simple mapping would be;

2.1) Buddhi in Samkhya ~ Metacognition - https://en.wikipedia.org/wiki/Metacognition

2.2) Manas in Samkhya ~ Phenomenology - https://en.wikipedia.org/wiki/Phenomenology_(psychology)

2.3) Ahamkara in Samkhya ~ Emergent property from above due to "I Think therefore I am".

Now realize that (1) is purely experiential/internal and only (2) is observable from an external pov and you get the idea why AI/AGI may seem "Conscious" (in the common usage of the word) to external observers.

From an empirical view, 1 is an untestable internal state. It may as well not exist. Believe anything you need to, set policy based on things you can observe.

> 1 is an untestable internal state.

Not quite; it is testable through experience by following any number of practices, prescribed in detail in various texts. There is also empirical evidence in the form of changes in physiological and mental parameters in somebody who has attained that state.

Ah so assuming this isn't nonsense, if you can't define a set of such parameters that generalize to a non-human it still isn't useful to test non-humans.

The mentioned "changes in physiological and mental parameters" is a well known field of study.

See Effects of Meditation - https://en.wikipedia.org/wiki/Effects_of_meditation

Also see The effect of meditation on regulation of internal body states - https://www.frontiersin.org/journals/psychology/articles/10....

The problem is that the physiological parameters are biological in nature while the mental states can only be observed by indirect means i.e. changes in external observed behaviour mapped to changes in mental states. It is up to us to define these parameters in a precise manner so that we can maybe generalize them to non-humans. A good example would be, what parameters of Metacognition can be delimited precisely so that we can say "A Machine Thinks"? Pattern recognition, drawing inferences, prediction is certainly one but not enough. In a very very narrow and limited sense you could define it as "Thinking" but certainly not "Conscious".

As Dijkstra said; "The question of whether a computer can think is no more interesting than the question of whether a submarine can swim". The point is that concepts like "Thinking", "Consciousness" etc. have to be redefined when it comes to non-humans. What Samkhya gives is a very precise definition of Consciousness in Humans which we can use as a framework to start from.

Reply on news.ycombinator.com