Meet a roboticist who’s training AI to write happening cookies and TED Talks

Alexander Reben wore a facade when he delivered a TEDx Talk in San Francisco progressing this month.

The mouth of a cartoonish face covering non-stop and sealed as nonetheless in speech, though Reben, a roboticist at Stochastic Labs in Berkeley, Calif., didn’t indeed pronounce when he was on stage.

Instead, Reben’s TED Talk was done adult of a infrequently inane, infrequently auspicious ramblings of an synthetic comprehension content generator.

Reben came adult with a suspicion after building a complement that writes happening cookie fortunes formed on inspirational quotes he found online.

“A lot of what a complement spits out creates clarity — if your widen your mind a bit,” he told Day 6 horde Brent Bambury.

For example: “Radical ideas competence be tough for everybody to register in their pants, generally. And there’s a dolphin,” Reben’s appurtenance pronounced on stage.

The origination irritated a seductiveness of TEDx organizers and they invited him to speak.

“I suspicion it would be interesting, instead of me articulate about my work, to have a complement indeed emanate a TED Talk from blemish regulating these technologies,” he told Bambury.

‘Ideas change so quickly’

The assembly wasn’t totally in a dark. Reben enclosed a disclaimer that a debate was created by a machine.

Still, he says, it took a assembly a small while to get into it.

“I had a crony in a assembly who described a greeting as befuddlement and fast branch into humour,” he explained. “Then people unequivocally got into it.”

The complement is identical to Reben’s happening cookie generator, though instead he plied a algorithm with a content of TED Talks from months and years past.

The AI separate out a array of mostly foolish phrases and Reben curated a best.

“The thing that’s tough is to make a thread of a story, since a ideas change so fast in this arrange of output,” he said.

Feed it a library

Reben’s complement produces clearly incongruous thoughts like ‘I am a bad situation’ since it’s means to commend structure and syntax, though not indispensably suitable word placement.

Take dual standard fortunes: one could say ‘your days will be filled with happiness,’ while another says ‘you’ll never be depressed.’

The AI competence flip some difference around and come adult with ‘you’ll never be happy.’ It’s still a fortune, though not a confident anticipation one expects after dinner.

The boundary of fortune or debate essay synthetic comprehension aren’t nonetheless clear, Reben says. While we could feed a appurtenance hundreds of poems, it would expected emanate something identical to what it has already read, instead of formulating a uninformed product.

‘I kind of like a some-more existential dismay ones,’ pronounced Reben about his AI-based happening cookie happening generator. (Alexander Reben)

Reben is interested in what would occur if we fed it an whole library.

“What if we went even incomparable and lerned it on literature, either it be communication or a play or a novel, and have it maybe invent a genre of essay that we’ve never seen before, haven’t even suspicion about as humans,” he asked.

The plea he’s confronting is that creativity isn’t indispensably structured or logical, and Reben believes AI competence strike a wall creatively. As it stands, humans are substantially improved served with an AI that recognizes stop signs in a self-driving car.

“Or there could be some breakthrough where it creates communication that we find awful, though competence be sensitive for other appurtenance training systems that decider poetry.”


To hear some-more from Alexander Reben, download a podcast or click a ‘Listen’ symbol during a tip of this page.

Article source: https://www.cbc.ca/radio/day6/episode-412-tourism-for-tokers-saudi-prince-under-scrutiny-how-to-kill-a-tv-character-april-ryan-and-more-1.4869003/meet-the-roboticist-who-s-teaching-ai-to-write-fortune-cookies-and-ted-talks-1.4869013?cmp=rss