Cen Inc. Wants You to Help an AI Hack the Turing Test
Cen Inc. is taking a novel approach to making an AI pass the Turing test and, in the process, hopes to give AIs a little humanity.
It takes 21 years to teach a human to be a human. That’s kindergarten, pre-school, high school, college and graduate school, or perhaps you leave somewhere in the process and found a start-up or take a job. At some point along the way humans get smart enough to answer questions like a grown up and we call that human intelligence.
Alan Turing – who broke the Enigma codes in World War II – gave us the test for when a computer becomes intelligent –The Imitation Game. Setup a computer to pretend it is a human. Ask it questions for 10 minutes. If you can’t tell it’s a computer then it is, for all intents and purposes, a human. Since we think humans are intelligent the computer must be too. But what does it take to make a computer do this?
So far, it’s proven hard to make an AI pass the Turing test. The acid test is the annual Loebner Prize which offers a $100,000 prize to the first program to fool a bank of questioners into believing it is human. The prize has been running for 12 years but so far progress has been slow. Cen is taking a novel approach to the problem – add humans back into the loop.
The big problem for AIs is they tend to be fragile, making childish errors when they come across new information outside their training set. In the Loebner test these errors give the AI away, when they occur in real systems the AI fails, sometimes with disastrous results. Cen tries to detect these situations and switch the question over to a human fast enough to intervene and answer the question correctly. The answers get folded back into the AI knowledgebase and the AI learns. In the limit the system should be able to pass the Turing test, which is not surprising given the system can switch all questions to a human in which case you are talking to a human.
In this age of tokens, the solution would not be complete without a blockchain element and Cen uses tokenisation to log and track the answers and reward ‘Minders’ – the humans who intervene in real time. They get ERC-721 tokens for their knowledge and the chance to earn tokens in the future when their knowledge is used.
Cen is at alpha stage at present and all the pieces are not yet complete but you can start to help program it. If you just want to test the system, ask it questions and if you like the answer type a ‘+’ sign and if you don’t type a ‘-‘ sign. If you want to be a Minder sign up on Github and start inputting your knowledge. At the moment the interfaces are suitable for programmers but Cen is working on an interface that will make inputting knowledge possible for anyone.