TedEd: The Turing Test: Can a computer pass for a human?

A TedEd by Alex Gendler

One of the questions that has plagued computers. What is consciousness? Can a computer really think? These are all very complex questions. British computer scientist Alan Turing decided to try a different question. Can a computer talk like a human? He created a new way of measuring AI’s, and it is known as the Turing Test. In it, an evaluator has text conversations with unknown human participants. He then evaluates the responses. If a computer can take the place of one of the participants without the evaluator noticing, it passes. He predicted that by 2000, AI’s with at least 100 megabytes of memory would easily pass his test. However, only very few have. And many of those that have have done more trickery than true intelligence. The first computer that passed the test acted like a psychologist, letting the human talk and sending their own questions back at them. Another acted like a paranoid schizophrenic, dominating the conversation to his own script. This was a loophole in the test, which has now been largely resolved through formal competitions and judges knowing beforehand that some of the conversations would be with machines. Still, many chat bots use this method, like a robot that talked well about Bill Clinton and one that took the persona of a Ukrainian boy, thus causing judges to assume odd grammar was a language barrier. Some smarter programs have analyzed huge databases of conversations, or learned from previous ones, to sound incredibly human. However, they lack a consistent personality and can’t deal with unknown topics. Human conversations are incredibly complex, and computers, despite their increasing memory and processing power, can’t deal with simple pauses like “Umm..” or sentences that require previous knowledge. So it turns out that to solve Turin’s challenge, we may have to think about consciousness after all.