Can artificial intelligence be intelligent?
perform better that a human, then we could believe it was intelligent. This test, although only thought as an approximation by Turing has a number of flaws in regard to defining Intelligence and consciousness. While passing a Turing Test maybe necessary for consciousness, it is not sufficient. For ex ample, Deep Blue’s defeat of Gary Kasparov at chess would pass a Turing Test, suggesting it is intelligent and therefore conscious when it is actually a very unconscious series of pattern finding algorithms. This is also another problem: any Turing Test is limited. A perfectly trained AI that could be perfectly created to fool the behaviour and language of a conscious person with a sophisticated enough algorithm and therefore passing the Turing Test without being neither conscious nor genuinely intelligent. Therefore, the Turing Test is not particularly useful but does show intelligence is not something that is easily tested practically but requires a theoretical approach. Theoretically, if you had a conscious intelligent Turing machine, how could you tell it was conscious and intelligent? This theoretical device is referred to as strong AI and outlines that, with sufficiently complex algorithms executed on a Turing Machine of any hardware, consciousness can be achieved. However, this idea has been deconstru cted by John Searle’s famous Chinese Room argument (Searle 1980). In this scenario a person is inside the Chinese Roomand passed Chinese messages through a slot. The personmust return an answer written in Chinese but does n’t speak any Chinese themselve s. They do however have a rule book that tells them how to calculate the correct answer for the given input. Seale argues that with a good enough book or algorithm, one can provide the correct answers without speaking or leaning any Chinese. In this way his answers will have the right syntax but no semantics. In the same way the symbols of Chinese may be unknown to the subject, the binary symbols 1s and 0s are unknown to a Turing machine. A Turing machine may well have an algorithm that can manipulate syntax, but it will never have an understanding of the semantics (Friedman 2002) . Without this capability it won’t be able to reason with language nor will it be able to form its own meta-language for creating thoughts. Consequentially, complex algorithms alone are unlikely to be sufficient for consciousness. At least in the human context, consciousness must have developed in human brains through evolution overmillions of years through randomevolutionarymechanisms such asmutation. In this way a strong AI algorithm, however complex, could potentially evolve by simulating and inserting randommutations into itself and running them. A programmer cannot attempt to recreate random evolutional mechanisms on a strong AI algorithm, just by programming randomness. The problem lies with computers being able to create random numbers. Computers lack true randomness, although what they can generate might appear to be like random numbers, they are actually ‘ pseudorandom ’ numbers that will eventually repeat for a given seed (Hoffman 2019) . You can’t generate randomness with a set of deterministic instructions as they are opposite concepts. Genetic algorithms in AI can still be used but their evolutionary mechanisms won’t be truly random and so their benefit in displaying con scious
74
Made with FlippingBook - Online catalogs