Sorry, you need to enable JavaScript to visit this website.

Who’d want to be friends with a robot?

Who’d want to be friends with a robot?

A machine held its own in a debate with humans this week. But this house believes that intelligence is pointless if you can’t crack a joke

A machine held its own in a debate with humans this week. But this house believes that intelligence is pointless if you can’t crack a joke

Tim Smith-Laing | June 22nd 2018

The internet has for some time hummed with anxious murmuring about the Singularity. The rate of technological progress is accelerating exponentially; the Singularity refers to the moment when computers have become so smart that they escape our control and eventually become super-intelligences capable of stamping out humans like so much vermin. Those tuned into news of the coming catastrophe keep a beady eye on IBM, whose scientists are doing all they can to ensure their own survival as obsequious quislings to our future mechanical overlords. On Tuesday, the company announced that it had brought us one step closer to “real AI” (an intelligence as smart as a human) with its snappily named Project Debater: a supercomputer dedicated to the art of competitive debating. After years of research, this week it finally competed against two real-life human debaters. The result? A thumping one-all draw – according to an audience that I suspect was almost entirely made up of people who thought that HAL, the genial yet murderous computer in “2001: A Space Odyssey”, was the real hero of the film. 

It was not quite John Henry versus the steam hammer. Even as IBM’s press office trumpeted the passing of another milestone on the road to true AI, one of the researchers offered the more-understated claim that Project Debater had managed to do something “sort of like what a human does when debating”. In fanfare terms, that is like hearing the Twentieth-Century Fox theme tune played on a kazoo. Most editors, even in the tech press, reached for their Brief-Recycled-Thinkpiece-on-the-End-of-Man button and left it at that. A good chunk of the rest of the internet just kept repeating the phrase “master debater”, as if it were actually a pun.

It is undeniably impressive, though. Set aside for the moment the following facts: that Project Debater is called Project Debater, that it manifests as a black monolith emitting the gentle, affectless tones of a child-killing psychopath, and that its “thinking face” is an animated set of gently bouncing blue balls. Ignore these, and you are left with a machine that can argue with a real human in real time. The hot topics at issue were the questions of whether “we should subsidise space exploration” and whether “we should increase the use of telemedicine”. It’s not clear what investment Project Debater was meant to have in either.

To put this in context, the greatest achievement of Google’s Duplex system (basically an AI secretary that can make phone calls for you) has been to book a haircut without the receptionist noticing that it was not a human. (One feels like the joke was actually on Duplex with that one. Given the number of times I have dealt with receptionists who failed to notice that I was a human, I am not sure they are the best control group.) And let us not even contemplate the notion of holding a conversation with Siri or Alexa. Project Debater, meanwhile, will independently formulate its own arguments for and against a proposition, listen to a human opponent and formulate counterarguments, and even – at least in the sense of the term understood by computer researchers – make jokes. Perhaps the most human thing about it, though, is that it cannot intelligently absorb more than four minutes of speech at a time. Me too, buddy.  

 

 

We are obviously still a fair way from the researchers’ goal of what is technically called Artificial General Intelligence: a machine that can successfully perform any task an average human could and even, perhaps, become self-aware. But what is really engaging in all this is the spectacle of watching IBM’s AI researchers gamely think through the kinds of problem-solving activities that, rolled together, would make something like a human brain. 

In their quest to do so, the company’s coneheads have over the last three decades created a chess player (Deep Blue), a quiz contestant (Watson) and now a debater (by which time they had apparently run out of names altogether). If you wanted to make machines self-aware, could you choose worse groups of humans to emulate than chess players, quiz nerds and people who think debating is a sport? Roll Deep Blue, Watson and Project Debater into one entity, and you would essentially be creating the most bully-able kid in school – one of those children who are not even popular with their own mothers. It would be like bringing Pinocchio to life and sending him out into the world with a note saying “Kick Me” pinned to his back. 

The irony of the route they have mapped for themselves is that if IBM does one day succeed in creating an intelligent machine, conscious of its own consciousness, capable of moving through the world and engaging its fellow sentient entities in conversation, it will most likely be a machine that no one will want to talk to. You may insert your own quip about God making man in his own image here if you wish, but what a way for artificial intelligence to enter the universe. In terms of the much-feared Singularity, we will just have to hope that Project Unloveable, or whatever the researchers christen it, reaches the conclusion that taking over the world would not be enough to make people want to be its friend, and starts learning how to be a real person.