Episode 18 - AI Gone Psychopathic, Embellishment vs Fact

An article about MIT research claims that they've built a psychopathic AI. We think the article is a little off base, and try to determine what's actually going on. We cover related examples, of Microsofts Twitter Bot learning from the worst people, and a cool Google Image Program that sees animals everywhere.

I also check in with Christian Hubbs, host of Artifically Intelligent, about his podcast. I was interviewed for todays episode, so check that out!

Go to the Artifically Intelligent podcast here.
Also check out their Twitter.

I also define a Convolutional Neural Net, but only on the highest of high level! More on that could come in future episodes.
See a related and recent interview of Yann Lecun (he was my professor in machine learning class at NYU)

Want to build a CNN yourself? Maybe go here.

Articles

Meet Norman, the Psychopathic AI (BBC)
Link to the research page from MIT.
Microsoft's Bot Tay from 2016 that learned from Twitter and had to be shut down.
The dreams of Google's AI - prepared to see cool images
Defeat the Baby

Previous Related Episodes

Episode 4 on Codebreaking, references the MCMC or Random Walk method to find a closer match (a simpler version of how Google's AI dreams happen)

Episode 12 on Virtual Friends and Enemies, another discussion on technology going the right way and the wrong way.

Episode 14 on Robot Communism, referenced multiple times.