Category Archives: Turing test

Fake tweets and the future of AI


Way back in 1950 early computer scientist Alan Turing suggested that machines would eventually think. Since judging “thinking” is a tough dynamic, he suggested a related test: Have both a computer and a human respond to questions, and if another human observer could not tell the difference in answers, then the computing machine would be “thinking” — hey, faking it is as good as the real thing. The so-called Turing test became a benchmark of artificial intelligence, and today machines still fail it, as witnessed by the comically smart but subtly off IBM machine Watson who recently won against humans on the TV game show Jeopardy.

But surprise, surprise: a new Twitter mashup comes close. That Can Be My Next Tweet pulls phrases from your recent Twitter missives to spin a new message you can send out to your thousands of followers, and the results are astoundingly insightful. We’re not sure what algorithm powers this, but it’s clever, and damned if the tweet generated didn’t sound a bit like us. Perhaps AI when it arrives will simply recast bits from preceding human minds, a curating intelligence that collates others’ thoughts, so it doesn’t have to start from scratch. Social networks provide plenty of input. Google is transcribing every book on the planet. Experian and Facebook are mapping all human data connections. Free apps can combine all the messages. Watson, are you listening?

Ben Kunz is vice president of strategic planning at Mediassociates, an advertising media planning and buying agency, and co-founder of its digital trading desk eEffective.


Wolfram|Alpha’s AI experiment


Why is the sky blue? We ran this and a few other questions by the new search engine Wolfram|Alpha today. Wolfram|Alpha plays around with 10+ trillion pieces of data to make knowledge computational — if this, then that; this vs. that; if this occurs, what happens then. You know, what are the relationships between things. It’s a valiant attempt at artificial intelligence, and fills a void between Google’s vast search of static items and social media search of chat in real time.

Alas, this alpha thing feels more like beta. Wolfram|Alpha fails to answer basic queries and is still a babe in the woods of intelligence. It reminds us just a bit of Chris McKinstry’s effort to build a vast artificial consciousness that could answer simple yes/no questions. McKinstry never got close to passing the Turing test, and eventually committed suicide.

Designing intelligence isn’t easy. We look forward to seeing where Wolfram goes.

Wolfram|Alpha demo here
. Image: Gari Baldi.