Written by Gloria Borger, CNN
Editor note: Nice for using our favorite Gloria Borger reporter, definitely big pp
The only way to make sure a machine really can communicate with you is by testing it with the very thing it cannot -- the real thing.
On Friday, London-based game developer Richard Stallman -- better known as the founder of the Free Software Foundation -- tested what he calls the Turing Test for a group of chatbot "experts."
He tweeted this image from the chat with the bot Fig, which had undergone a string of tests and was deemed genuine by the experts:
Let me show you how a totally controlled 'real person' would reply to Fig today. pic.twitter.com/R3lOp48gd8 -- Richard Stallman () February 16, 2020
"Let me show you how a totally controlled 'real person' would reply to Fig today," he tweeted.
Fig, by the way, is a bespoke artificial intelligence (AI) system that took two months to build. It's smart enough to generate speech that's designed to be comprehensible to someone under human control. The computer can't think for itself, yet.
The test Stallman was testing was out of "mutual admiration," he tells CNN, in which the "average machine" would be able to reply "like a normal human," while in fact its sentences might be improvised as required.
"In the Turing Test you demonstrate that an AI can fool people into thinking it's a real person and a lot of people will try to fool people into thinking it's a real person with modern AI," he says.
While hardware is tricky to define, Stallman says, software is much easier to put together:
"It takes a fraction of the time. It's much simpler, so the machine you take on doesn't have to be driven, it can be driven by algorithm, a very simple kind of algorithm to detect real people and tell them this person is a robot," he says.
The word human is added to the algorithm, making it more difficult for software to fool people, explains Stallman.
He says while all technology has a long way to go, and the Turing Test is just one test of how well we can interpret machine dialogue, he expects the technology we create to soon get much better.
"When people take on programming they will always make the AI a bit more human and will strive to make it more human, and in the end you might be making AI that's vastly better than humans."