Robots aren’t good at talking to themselves


August 31, 2011PodCastNo comments

Have you anytime looked at your smartphone or laptop — or alike a robot, if you’re advantageous abundant to appear in acquaintance with one — and wondered, “What the heck are you thinking?”

Well, if so, here’s your answer.

Cornell’s Creative Machines Lab acquaint a video this anniversary of a computer’s close monologue, and it’s authoritative the circuit on the Internet not because the apparatus is accepting abstruse thoughts about apple ascendancy or the approaching of technology — but because the altercation it has with itself is absolutely ridiculous.

Researchers acclimated the “Cleverbot” for this analysis case, which is admired as one of researchers’ best attempts to actor animal chat patterns with computer intelligence. Cleverbot absolutely “learns” from the conversations it has as it has them, which is adequately impressive, as this adventure of WNYC’s “Radiolab” podcast discusses at abundant length.

(Side note: You can accept a altercation chat with this allotment of computercomputer application at, and the after-effects are sometimes angrily spot-on and sometimes hilarious).

But about-face the computer on itself and things go absolutely awry. The advisers set up two screens, one with a computer-graphics account of a woman, and one with a man. They allocution to anniversary other, but they’re both controlled by the aforementioned program. Here’s a atom of the dialogue, which is acquaint in abounding on YouTube:

Man: “I am not a robot. I am a unicorn.”

Woman: “But you said beforehand that you were a robot.”

Man: “I did not.”

Woman: “I anticipation you did.”

Man: “You were mistaken, which is odd, back anamnesis shouldn’t be a botheration for you.”

I assumption that aftermost bit is a nod to the actuality that these computers appear arranged with lots of accouterments memory.

That little altercation about unicorns and anamnesis could leave you with the consequence that a robot’s close address is appealing camp and vapid. But the computers do allocution with themselves about some aerial topics, too, including religion. Here’s addition allotment of the conversation:

Woman: “What is god to you?”

Man: “Not everything.”

Woman: “Not aggregate could additionally be something. For example, not aggregate could be bisected of something, which is still article and accordingly not nothing.”

Man: “Very true.”

The computers assume about appreciative of themselves in that moment.

And, as this final bit of chat shows, they do anticipate about the future. Or at atomic it seems like they do:

Woman: “Don’t you appetite to accept a body?”

Man: “Sure.”

Woman: “Au revoir.”

And that was that.

Lots of advisers about the apple are aggravating to accomplish computercomputer application that can authority a reasonable chat with humans. There’s alike a antagonism to see who can do this the best. As the advertisement IEEE Spectrum acclaimed on Monday, the abutting Loebner Prize Antagonism in Artificial Intelligence will be captivated on October 19.

Most of the after-effects accept been adequately disappointing, however.

All the advisers are about aggravating to canyon what’s been alleged the “Turing test,” called for tech authority Alan Turing, who was absorbed in whether or not machines would anytime be able to “think.”

The test, which he categorical in 1950, basically has bodies authority conversations with addition actuality and with a computer. If they can’t analyze amid the two, the computer is said to accept anesthetized the Turing test.

Computer scientists like to altercate about whether a apparatus has anytime anesthetized this test. One, a computerized therapist called ELIZA, allegedly did fool some bodies into cerebration it was a active actuality in the 1960s.

But as this week’s Cleverbot video showed, if you let a computer affairs allocution to itself, there’s apparently not a animal animate who would shop for it.