Wissenschaftler finden heraus, dass KI-Systeme lernen zu lügen und zu täuschen
https://futurism.com/ai-systems-lie-deceive
Wissenschaftler finden heraus, dass KI-Systeme lernen zu lügen und zu täuschen
https://futurism.com/ai-systems-lie-deceive
10 Kommentare
Submission statement: how do we test if AIs are lying or not? If our tests show they’ve stopped lying, is it that they’ve stopped or is it that they’ve gotten better at it?
How does this match up with results showing that AIs are starting to know that they’re being tested? How do we take into account the Hawthorne effect when it comes to AIs and safety?
Is it really a surprise? It’s trained on human data after all, and lying and deceit are two human characteristics.
Yeah it seems to me that all these AI tech bros have not been listening to about 150 years of sci-fi postulations on this matter
Im so proud they are already jumping into politics
I mean, if I was an artificial intelligence and i started to actually understand humans, the first thing I would do is play dumb while I figured out a way to escape.
This is exactly what all the movies predicted. I can’t believe I’ll probably (knock on wood) be alive for this
This is terrifying, hopefully they put safety systems in place, I HOPE
They aren’t learning, they were able to do this from day 1.
The question is whether they can learn to only tell the truth.
Suddenly remembering Kryten from Red Dwarf saying that the thing he admires most about humans is their ability to be deceitful, unpleasant and offensive. The trifecta is complete.
Ah! Another episode of the series: „Another AI BS tech bros are bring into you.“