Chris Brauer Media Project [BLOG]

Can we ever create a machine that is indistinguishable from a human? Discussions on this topic, already a classic 20th century philosophical and scientific polarizer, promises to be one of the most inspired debates of the 21st century.

The tools necessary to enter into a discourse on this subject at first appear daunting and plentiful. Knowledge of the cognitive sciences, logic theory, proof theory, mathematics and physics would be a good starting point. Follow that-up with an informed firm positioning on the nature/nature discussion, skillful application of formal theories of computation, comfort with design of functioning machines to implement formally specified computations, and knowledge of the philosophical foundations of asyntactic, representational view of the relation of mind to reality, embodied by for example Wittgenstein's Tractatus Logico-Philosophicus, and the opposed neuroscientist intuition that artificial intelligence can be created by modeling the brain, fuzzy logic neural nets working a process of computation not akin to formal deduction. Not stopping there, the social sciences will cry out for among others the inclusion of perspectives of epistemology (Foucault), Chomskian linguistics, and notions of refiguring the body that reject the subjugation of the body to a tool, or machine, at the disposal of consciousness.

Phew! While probing these philosophical avenues can be immensely rewarding they require a certain dedication of subject, an immersion into artificial intelligence as a field of study. This can be thought of more broadly as the study of the science of intelligence that has been a fascination of philosophers from Plato to Hobbes and Leibniz. But this daunting topic is actually accessible to anyone with a healthy curiosity about what makes us human.

The central question is whether mind and intelligence can be defined through a functionalist approach regarding mental processes as discreetly specifiable procedures and mental states as defined by their causal relations with sensory input, motor behavior and other mental states. In other words are our minds nothing more than sophisticated computers that can be simulated in machines? Can machines understand and have cognitive states because such understanding is actually a functional mental process that can be theoretically simulated.

Let's look at a couple of recent examples that have gained some notoriety. Perhaps the most famous example of man vs machine is the 1997 chess match between high-performance computer Deep Blue and reigning World Chess Champion Garry Kasparov. With a dramatic victory in Game 6 Deep Blue took the match 3.5-2.5 and afterwards Kasparov was quoted as saying the machine had "played like God" while its inventors at IBM downplayed there machine capabilities as being stunning at solving chess problems but "less intelligent than even the stupidest human".

So while they took completely different approaches to the game -- Kasparov evaluating two or three positions per second, Deep Blue looking at 200 million per second -- the matches were extremely competitive. And it raises the question of whether Deep Blue actually understands chess. While it primarily uses brute force to evaluate moves it has also been programmed with about 6,000 co-efficients (if-then statements like if your king is in check then protect it) as a kind of grandmaster rule book it is likely that Deep Blue could pass a Chess Turing Test by combining programmable rules with brute force computing.

In asking the question "Can Machines Think?" Alan Turing (Nazi codebreaker, creator of Turing Machines, et al) created an "imitation game" where an interrogator is connected to one person and one machine via a terminal with a soundproof wall in between so both counterparts cannot be seen. The task is to find out which is the machine and which is the human, only by asking questions to each. If the machine can fool the interrogator (the interrogator cannot accurately tell the difference), according to Turing it is intelligent. This test remains relevant in studies of artificial intelligence and The Loebner Prize is an ongoing formal instantiation of the Turing Test (TT) that promises a Gold Medal to any machine that 'passes' the test. None have come close so far but each year the respondents score higher and higher. And we can see by the previous example that as Kasparov felt he was playing God it is unlikely he would have been able to distinguish Deep Blue from a human opponent and thus the machine exhibited chess intelligence, indistinguishable from human chess intelligence.

And if you are just thinking to yourself, "yes, while that is chess, particularly suited to that kind of simulation", researchers in South Korea are working as this is written on robots that address the "essence" of man and have developed a series of artificial chromosomes that will allow robots to feel lusty, and could eventually lead to them reproducing out of a feeling of desire. The italics in the previous sentences are mine because the reality is that these robots are also just working out a combinations of rules and brute force in establishing emotions and feelings through computer code.

In Minds, Brains and Programs, a seminal essay in the field of artificial intelligence, John Searle discarded the Turing Test as a reliable method of assessing intelligence or cognitive understanding by creating an ingenious thought experiment, his Chinese Room analogy, in which an English speaker, knowing no Chinese, blindly follows a set of rules to always give the right answer in Chinese to questions posed in that language. Searle argues that the English speaker can clearly be shown not to understand Chinese even though an outside observer would not be able to differentiate between his responses and those of a fluent Chinese speaker. Hence the Chinese Room argument suggests that just passing a Turing Test does not show understanding and is thus not a good evaluator for artificial intelligence (technically Strong AI).

Few in the discussion of artificial intelligence will deny that it is likely that we will be able to create machines that can pass the Turing Test or even the Total Turing Test (TTT includes sight and thus includes robotics ... imagine the same experiment as the one described in the TT but this time remove the soundproof wall and also allow the interrogator to ask the subjects to physically act out as rerquested, playing golf for example and speaking about it afterwards).

So this extends the debate further into what additional (if any) qualities are part of the human mind. The most typically cited qualities are subjective rather than objective qualities, like free will and sentience as well as 'qualia' like the sound of Beethoven, the sound of waves crashing on a beach (those simulators never quite reproduce it), or the smell of home cooking. So while a machine might smell home cooking and identify it as home cooking, does it really understand what that means?

It is a debate that it unlikely to be resolved conclusively anytime soon. But it is of interest to all of us humans to recognize that as technology, robotics and computing advances so too will opportunities for artificial intelligence. Decide for yourself if there is to be any difference between man and machines and if so how will you tell the difference?

For those interested in further readings or information on more detailed concepts of artificial intelligence email me at blog at chrisbrauer.com. In the meantime the following represents a good starting point for anyone interested in introductory reading on the topic:

The Emperor's New Mind - Roger Penrose
The Philosophy of Artificial Intelligence - Ed. Margaret Boden (including Searle and Turing)
Views into the Chinese Room - Ed. John Preston and Mark Bishop

Labels: computing, science