Daily Archives: October 27, 2017

None of Us is as Dumb as All of Us! Unless, Of Course, You Include AI!

the doctor sees a lot of unsolicited pitches hit his inbox each and every day. Since SI does not cover press releases (since he just does not give a damn about your meaningless marketing sound-bites which do little to nothing to advanced education and technology) he ignores most of them. But this week he saw one of the most ridiculous headlines ever:

Can AI Harness the World’s 2 Billion Social Media Influencers?

Ignoring the fact that that the headline is factually incorrect (there are 2 Billion Users, NOT Influencers), this is one of the dumbest questions ever posed and anyone who understood anything about the state of AI today would not even want to ask it!

Generally, if you are going to train AI, you want to train AI on expertise. And where’s the last space you’d expect to find expertise? That’s right! Social Media.

But that’s just the tip of the why-you-should-not-do-this iceberg! If you include everyone, you not only include everyone of above average intelligence, but everyone of below average intelligence by very definition. So while you will have a few geniuses, you will also have morons, imbeciles, and possibly even idiots (as per the original Binet IQ scale). Do you really want them training your AI?

Moreover, what do people share on Social Media? Their most brilliant ideas? Well developed arguments? Philosophical contributions? Wisdom? Profound insights? Or pictures. Comments on politics. Viewpoints on pop culture. Their thoughts of the moment. Complaints. Rants. Digs. Manifestos. Insults. Self Praise. And so on. And most of it in blurbs, not sentences, and definitely not paragraphs. And in addition to the onslaught of bad grammar, the rate of spelling errors is atrocious.

Is this what you want to train an AI on? Really? You really want an AI that is going to make decisions like an angry dumb, self-obsessed, neurotic, troll with self-esteem issues making your decisions? And that’s likely a best-case scenario.

the doctor doesn’t know about you, but if he’s going to trust an AI, he wants that AI trained by experts for specific tasks, with performance analyzed and tweaked by other experts, since, as we know, there is no such thing as artificial intelligence, since no algorithm is intelligent, no matter how advanced, and what we really have are advanced automated reasoning algorithms. But if those algorithms were trained by the impaired, those algorithms are the last algorithms he would ever want to use. And those algorithms should NOT be on your list either.