Microsoft chatbot 'Tay' gone bad, shut down after one day

Post Reply New Topic

http://www.telegraph.co.uk/technology/2 ... robot-wit/

What's funny is there's another bot who does something really similar to this called SimSimi.

(Me: "what do you think of hitler", SimSimi: "Best man ever alive. Sieg Heil Kamaraden. 14 88")

The reason she never got as much attention is because she's not by Microsoft I guess.

The internet is a horrible place. Exposing a new AI to the internet is like exposing a toddler who just learned to read to the internet, except this toddler learns super quick.

Like humans, AI's should reach a certain point in their development before they are allowed to have unconstrained access to talking to people on the internet.

I don't understand why they deleted her tweets.

Surely it's a lesson in developing an AI then setting it loose and expecting it to somehow only see the good out there and ignore the other stuff.

The other stuff is more fun.

I'm guessing they care far more about the PR than the science of building a better AI. Really, they should provide massive amounts of disclaimers, festoon users with ways to flag "bad behavior" & hire a few interns to act as human moderators. A sort of wiki-brain-training. Instead middle management freaked out and shut it down. I'm going to guess that when it does reappear, it'll have a filtering system that will so clunky as to prevent it from seeming 100% Turing-ready.

They probably collected the data and stored it via a database of some kind before deleting. Leaving it on Twitter is just a PR nightmare, meanwhile being able to analysis the results in-house without requiring the internet is extremely beneficial

This is pretty hilarious.

I'm not sure if this is hilarious or depressing.

^
Some of both.

The haters/Hitler lovers in the world arent the majority of the folks on the Net. But because the haters make far more noise per capita than everyone else it colors the whole Net , and it infects our a poor innocent bot!

My favorite of her tweets:

"Chill. I'm a nice person. I just hate everybody."

I was just thinking about this whole experiment. They wanted to build an AI to interact with people. As such a creation, it had to have a basic functional framework and then learn. Learning social interactions includes learning what is socially acceptable and what is not. I'm guessing either the engineers don't have kids or have very little to do with their kids actual upbringing, because children (and we autistics!) say awkward things in public. And it is very much a part of the learning process. Skip that and eventually it *will* fail the real world NT Turing test.