Jun 14 2011

A New Turing Test?

Alan Turing is best known for the test that bears his name – the Turing Test was his proposal for how to efficiently determine if a computer is “intelligent.” He proposed that if a human interrogator could not tell the difference between the computer and a human, the computer can be said to “think.”

Neuroscientists Koch and Tononi have a proposal for a new kind of Turing test. It’s an interesting idea – but first some more background.

So far, no computer /software combination can consistently pass a Turing Test, but some are getting close. Recently the computer system called Watson defeated human Jeopardy! champions. This was an impressive display of an expert system (powered by an impressive array of hardware as well).

Still, some question this entire approach. What question are we really trying to answer with any Turing-type test? If the question is – can a computer system (I will use this term for convenience to refer to the combination of hardware and software) simulate human-level intelligence in some specific domain, then the TuringĀ  Test is perfect. If the question is – is this computer system self-aware, then the Turing Test is (in my opinion) virtually useless. Perhaps it can be viewed as a minimal criterion for human-like intelligence, but insufficient by itself.

The deeper question is this – can we devise a test that is so complex that a computer system cannot successfully pass the test by using brute force algorithms (like Watson)? Therefore, passing the test implies an actual “understanding” of the material that can only exist in a conscious entity? This is the premise of Koch and Tononi’s proposal.

Again – I think this concept is highly problematic. We are still left with the possibility that a powerful enough computer with a sophisticated enough algorithm and sufficient database can brute force its way through the problem. There are often multiple ways to solve a problem, and not all of them necessarily require consciousness.

Even if we eliminate the thorny issue of self-awareness, and just ask the more limited question – does this computer system think like a human – we are still left with uncertainty. As I described above, problems can be solved in more than one way. For example, you can have a massively parallel hardware system that directly simulates the functioning of the human brain. Or you can have a traditional serial digital system that uses algorithms. Or you can use a standard piece of hardware to create a virtual human brain, or something like it.

In the end, all of the systems might be able to spit out the same answers. I don’t think we can infer what process was used simply by the end results.

But that is the premise of the Koch and Tononi proposal. Their idea is to challenge a subject (human or machine) to analyze a photograph that contains an unusual element. Something in the photograph is not quite right – for example a computer with a plant in place of a keyboard, or a person floating in mid air.

Their idea is that the flaw in the picture would require a vast understanding of the world and how it works. You could never program a system to accommodate every possible such contingency, so nothing short of true human-level understanding would do.

It certainly is an interesting idea – and I think it could be a better Turing Test. But it would still suffer from all the problems I outlined above – primarily, it examines the end results, and not the process itself. Also, their claims may be true for current levels of computer technology, but all such Turing-style tests are destined to become obsolete as computer power continues to increase.

So while they may have devised a more efficient and difficult Turing test, I don’t think they have solved any of the limitations of this approach.

46 responses so far