[lvturner]

Tried this on a few photos of my dog - one of her half asleep, lying on her back and clearly very relaxed.

It said she was angry.

To be fair, it mis-identified her, and she is also a mixed-breed Asian dog so I'm not entirely surprised.

But it brings up questions for the future - what happens when AI is wrong but used as proof of something?

[kortex]

> what happens when AI is wrong but used as proof of something?

False positives everywhere. There was an article on HN a few weeks back about cops picking up the wrong guy, flagged by facerec. Even though the system's printout had a huge warning, "This document is not a positive identification", the detective still had the density to say, "So I guess the computer got it wrong, too."

https://www.npr.org/2020/06/24/882683463/the-computer-got-it...

[rsa25519]

> But it brings up questions for the future - what happens when AI is wrong but used as proof of something?

You bring up a fair point, but I don't think AI is special in this regard. Imperfect sources of information synthesis (humans, naive algorithms, etc) are used all the time. Hopefully, AI should simply add one more data point from which to understand a given situation.

[lvturner]

People, presently seem much less willing to challenge what a computer says however.

I'm wondering if Little Britain's "computer says no"[1] sketch is somewhat foreshadowing here. It's certainly a situation I've ran in to in several banks (globally).

[1] https://www.youtube.com/watch?v=0n_Ty_72Qds

[dilippkumar]

> But it brings up questions for the future - what happens when AI is wrong but used as proof of something?

I worry that the pace at which things like this will get packaged as “absolutely reliable, scientifically proven” technologies to all sorts of consumers will be far greater than the pace at which our language catches up with abstract ideas with which nuances can be widely communicated.