Intelligence, Artificial Intelligence and Vedanta Hinduism20 Jan 2020
Intelligence - What is it?
Meriam-Webster dictionary defines intelligence as
(1) the ability to learn or understand or to deal with new or trying situations, (2) the ability to apply knowledge to manipulate one’s environment or to think abstractly as measured by objective criteria (such as tests).
Hence, something to be considereed intelligent, it must have:
the ability to apply knowlege
the ability to pass the objective tests of
Test of Intelligence
Now, the question appears how can you objectively test the intelligence? When talking about machines, first thing that comes into mind is the Alan Turing’s Imitation Game3 paper which defines the test for intelligence of machine. The test is widely known as Turing Test. In summary, it says that if there is a machine, a human and a judge and if the machine is able to deceive the judge that it is a human and human is a machine, then it has passed the test of intelligence.
For purpose of this paper, however, I will be defining my own test based
on Hindu Philosophy known as Advaita Vedanta. In summary, Advaita
Vedanta 45 talks about the ways to seek
spiritual liberation through acquiring
vidya or knowledge. The text
talks at length about cognition, the means of acquiring knowledge and
provides proofto ascertain that the knowedlge has been acquired. Since,
as students of AI we are interested in the proof of
gain of knowledge
or intelligence, there is no other text that goes deep into this subject
of epistemology. Advaita Vedanta talks about six pramanas or proofs.
Pratyaksha (perception): acquiring knowledge through sesnses and worldly objects. In case of AI, if an agent can acquire knowledge without any supervision
Anumana (inference): applying reason to reach a conclusion from one or more observations or previous understanding. If an AI can use it’s knowledge to reach a connclusion, it is an intelligent AI.
Upamana (comparison, analogy): gaining knowledge through comparison and analogy. For example, in case of Raven’s Progressive Matrices, the KBAI agent can generate the candidate answer images and then use comparison to identify the correct answer.
Arthapatti (postulation): generally means implications, for e.g. if...then statements, similar to Production Systems discussed in section 2 above. It can also means the assumption of existence, fact or truth.
Anuplabdhi (non-perception or negative/cognitive proof): It says that knowledge of something’s non-existence or non-perception is a valid proof of intelligence. For example, if Watson can prove that there is no question or fact that supports the answer then it will be considered intelligent.
Shabda (word): Relying on past knowledge of experts like teachers, parents, etc. In case of AI, it will be equivalent to Supervised Learning
We can use these as parameters to test an AI Agent. If an agent is able to do perform well on each of these parameters then it is a intelligent enough to be an general AI otherwise low score on these will mean narrow AI or not an AI. These scores can be used to easily compare Agents vs Agents or Agents vs Humans in terms of intelligence.
Is Watson Intelligent? Is it more intelligent than Jennings and Rutter?
In 2011, the Watson computer system won the TV game of Jeopardy! against champions Brad Rutter and Ken Jennings. Let’s check if Watson was more intelligent based on the above Test of Intelligence.
Let’s check the following table:
|Pratyaksha (Perception)||No||Yes||Yes||Watson cannot on it’s own acquire knowledge. It has to be taught.|
|Anumana (inference)||Yes||Yes||Yes||Watson can use its knowledge base to quickly infer the essence and thus reply.|
|Upamana (comparison, analogy)||Yes||Yes||Yes||Watson is good at comparison and so are the humans|
|Arthapatti (postulation)||Yes||Yes||Yes||Watson can use his knoweldge base to postulate and draw conclusions|
|Anuplabdhi (non-perception)||No||Yes||Yes||Since Watson cannot acquire knowledge on it’s own, it can not guaranteedly proof the non-existence.|
|Shabda (Word)||Yes||Yes||Yes||Only way for Watson to learn is through Supervised learning|
|Total Intelligence Score||4||6||6||Hence, we can say Jennings and Rutter are more intelligent|
Thus, using the pramanas above, we have proved that Watson is not as intelligent as Humans. It is at best a narrow AI, which is good at tasks for which it has been made. But for other tasks it is not as intelligent as humans. The power of perception and non-perception and to reason for them is what makes humans different from machines.
References and Footnotes
Gottfredson, Linda Susanne (1994). Mainstream Science of Intelligence. url: (https://en.wikipedia.org/wiki/Mainstream_Science_on_Intelligence). ↩
Neisser, Ulric et al. (1995). Intelligence: Knowns and Unknowns. url: (https://en.wikipedia.org/wiki/Mainstream_Science_on_Intelligence). ↩
Turing, Alan (1950). “COMPUTING MACHINERY AND INTELLIGENCE”. In: url: (https://www.csee.umbc.edu/courses/471/papers/turing.pdf). ↩
Vidyabhusana, Satis Chandra (2006). A History of Indian Logic: Ancient, Me- diaeval and Modern Schools. url: (https://books.google.com/books?id=0lG85RD9YZoC&dq=taittiriya+pratyaksa&source=gbs_navlinks_s). ↩
Deutch, Eliot and Dalvi, Rohit (2004). The Essential Vedanta - New Source Book of Advaita Vedanta.url: (https://archive.org/details/EliotDeutschRohitDalviTheEssentialVed page/n19). ↩