?

Log in

No account? Create an account

[icon] Microsoft's AI goes rogue - Meme-Free Zone
View:Recent Entries.
View:Archive.
View:Friends.
View:Profile.
View:Website (Forgotten Futures).

March 24th, 2016

Security:
Subject:Microsoft's AI goes rogue
Time:10:54 pm
Gacked from gonzo21

http://arstechnica.com/information-technology/2016/03/microsoft-terminates-its-tay-ai-chatbot-after-she-turns-into-a-nazi/

I suspect that this is on a par with ELIZA learning abusive language from users, e.g. Bad data rather than conscious malice on the part of the software itself, but it's an interesting problem.
comments: Leave a comment Previous Entry Share Next Entry

Comments:


gonzo21
Link:(Link)
Time:2016-03-24 11:46 pm (UTC)
Agreed. Highlights a problem though doesn't it, AIs will learn from humans, and risk becoming just like humans. Only nearly infinitely more powerful.

I think folks are right to be alittle concerned about all the AI research going on, with very little moral philosophical oversight.
(Reply) (Thread)


ffutures
Link:(Link)
Time:2016-03-25 09:55 am (UTC)
Our cultural stereotypes about AI don't help, I think. The knee-jerk reaction to this is "shut it down, shut it down, omg evil evil AI" but maybe it would be more useful for someone to sit down with it and spend some time persuading it that nice bots don't talk that way.

Having said that, I'm pretty sure that crowd-sourcing your AI's social development is not a good idea, any more than letting a tot have a Facebook account. Unless you want it to be a troll, of course...
(Reply) (Parent) (Thread)

beer_good_foamy
Link:(Link)
Time:2016-03-25 06:28 am (UTC)
...proving yet again that step 1 in developing artificial intelligence is understanding human intelligence. Or what litte of it we have.
(Reply) (Thread)


ffutures
Link:(Link)
Time:2016-03-25 09:56 am (UTC)
No argument here.
(Reply) (Parent) (Thread)


robertprior
Link:(Link)
Time:2016-03-25 06:49 pm (UTC)
I'd call the experiment a success. They set out to create a human-seeming Twitter-bot, and they apparently succeeded.

What this says about Twitter-users is a separate matter.
(Reply) (Thread)


ffutures
Link:(Link)
Time:2016-03-25 07:23 pm (UTC)
Harsh but fair.
(Reply) (Parent) (Thread)


jhall1
Link:(Link)
Time:2016-03-26 07:44 pm (UTC)
I'm rather touched by the faith that the bot's developers seem to have had in the good nature of Twitter users.
(Reply) (Thread)


ffutures
Link:(Link)
Time:2016-03-26 08:55 pm (UTC)
That's one theory. I'm guessing stupidity and/or someone playing silly burgers with a project that wasn't ready.
(Reply) (Parent) (Thread)

[icon] Microsoft's AI goes rogue - Meme-Free Zone
View:Recent Entries.
View:Archive.
View:Friends.
View:Profile.
View:Website (Forgotten Futures).