The Turing test is the best and worst artificial intelligence test we’ve got

There has been a lot of debate on whether the Turing test is a good test of artificial intelligence.

The good

Without having a formal definition of intelligence, Alan Turing created a test using the closest thing we could use to evaluate computers, humans. A computer talking to a human via a terminal is a decent proxy to understand if it is intelligent. From the start of the computer revolution until now, almost any humans could interact with a computer and know if it is intelligent or not. So the Turing test is still holding up while nothing better has come along.

The bad

Full end to end automated training of computers is key component of building and experimenting computer models quickly. Since a human must be involved in evaluating the computer, this intelligence test cannot be automated. If there could be a way to evaluate the model in with humans in the loop in a fast and automated way, then the Turing test could potentially be sufficient.

Many humans think intelligence is only about communication and so researchers spend time building computer models that can read and type like with GPT-3. If you like at living organisms, they are all considered intelligent to various levels, yet they do not communicate like us at all. We see their intelligence in the way they move, survive, evade pray, hunt for food,etc. Yet we do not focus on this kind of intelligence with computers. I would say that animal level intelligence is probably a prerequisite to human level intelligence. So the Turing test does harm in sending researchers to focus on the wrong areas. You can’t run before you can walk.

Leave a comment

Your email address will not be published.