AI Scary

[astroron]

The more technology savvy of you may find this interesting, and
perhaps like the author a bit scary.
I guess some countries governments will also find it interesting. ;)
https://arstechnica.com/information-technology/2019/02/researchers-scared-by-their-own-work-hold-back-deepfakes-for-text-ai/
Cheers:thumbsup:

[silv]

Thank you for posting!

Interestingly, a Fraunhofer institute has recently finalised a software to detect fake news in which they utilised a German language corpus derived mainly from an online "newspaper" run by Russians. That's a fake news newspaper, frequented by far right Germans but written by Russians with German as 2nd language.

Or - that newspaper is already written by a Russian-developed A.I. like GPT-2 (or Chinese-developed, for that matter, lets not forget China in the fake news and hate speech circus).
In this case, the new Fraunhofer A.I. has "learned" unwittingly from another AI. And is now used to lead a war against that "father" AI and its siblings.

A re-enactment of Greek mythology :D

[Wavytone]

I’m not surprised - using “likes” to select the material to train an AI engine is just plain wrong - there is an implicit assumption that because something is popular it must therefore be valid (ie true).

But sadly that which is factual is often unpopular.

Might as well try to train a brain surgeon by feeding it episodes of “Bold and the Beautiful” interspersed with ads for kitchen knives.

[LewisM]

It is my understanding this is PRECISELY how US medical students are trained.

[gaseous]

There's the basis for a reality TV show if ever I heard one.