Microsoft deletes 'teen girl' AI after it became a Hitler-loving sex robot within 24

Skullmeat · 03-26-2016, 01:37 AM

Did no one at MS go visit cleverbot for 5 minutes? That's all it would have taken for them to understand why AI that learns from humans is a bad idea.

DarkMuse · 03-26-2016, 01:39 AM

(03-25-2016, 10:12 PM)Oni Wrote: The comments it made about Obama/Hitler are so outlandish they're funny!

Well, they assumed the AI would model its responses based on those of "normal" people. Microsoft has probably learned that not all people are normal.

Like....they know what the internet is comprised of right? It's not like they were sending it to a classroom or a daycare, they sent it to one of the most extreme places in the world. You have everything from mlp fanboys to nauseating filth...

m0dem · 03-26-2016, 03:22 AM

(03-25-2016, 10:12 PM)Oni Wrote: Well, they assumed the AI would model its responses based on those of "normal" people. Microsoft has probably learned that not all people are normal.

Also the fact that people will go out of their way to influence it in funny ways.

mothered · 03-26-2016, 03:58 AM

(03-25-2016, 10:12 PM)Oni Wrote: Well, they assumed the AI would model its responses based on those of "normal" people. Microsoft has probably learned that not all people are normal.

Yes, I have to agree with that.

Given It's an AI, you'd think they'd test It thoroughly with as many commands and responses as possible (normal, abnormal, unexpected, variations etc) prior to releasing the final product. Evidently, they've failed to do so.

I must say, It's certainly funny.

(03-26-2016, 12:52 AM)notblunt Wrote: The program was built with machine learning, from no predefined data set. Therefore all responses are markov chains of what it gets tweeted

In terms of "responses" yes I do agree, you're correct.

But wouldn't It be programmed to "accept" the commands, and then using It's machine learning process, It'll give It's responses based on the commands It receives?