Sort by controversial a.k.a. Shiri’s scissor (2018) | Hacker News Reader

Sort by controversial a.k.a. Shiri’s scissor (2018)

slatestarcodex.com

62 pointsappwiz15 comments

I like this story because I mentally replace “ML algorithm” with “exhaustive search”

We, as a society, are doing an exhaustive search of every possible person-opinion-framing combination and ranking them by controversial. I love this article for pointing it out.

It's a fun story, but we should remember machine learning is not magic. It's not going to produce a Basilisk, and we are not helpless marionettes manipulated by advertizement agencies and social media. And if somebody was gaming your political system via Facebook and Twitter, then maybe your system is not resilient enough and has to change.

OTOH, I really wonder what would happen if you tried to train an adversarial neural network to produce pictures that make somebody angry, or horny, or to provoke some other very primitive reaction. I guess it doesn't work or people would be doing it already.

Reminds me of BLIT: http://www.infinityplus.co.uk/stories/blit.htm

Yeah that is what I was referring to. Pretty chilling story.

MuZero is a basilisk from the perspective of Go players.

Go isn't the same as the human mind, though. That said, I think you could argue it's just a matter of time until we have enough power to do such things. But, you could also argue that such manipulation may simply not be possible, brains perhaps are not deterministic machines.

Could you explain this more? It's a basilisk how?

> OTOH, I really wonder what would happen if you tried to train an adversarial neural network to produce pictures that make somebody angry, or horny, or to provoke some other very primitive reaction.

That is, quite literally, the entertainment industry. I'm only half joking, I think the analogy works pretty well. You've you the same actor/critic dynamic that forces ever further refinement and innovation.

If curious see also

2019 https://news.ycombinator.com/item?id=21190508

Are there any examples of such scissor statements? This reads like fiction and would be easy to prove by showing us one of these

It is fiction, it's meant to make you think. The specific scissor statements would be specific to a given forum/company/etc. in any case.

It says "fiction" right at the top, but one given in the story was:

> A Republican Supreme Court nominee accused of committing sexual assault as a teenager.

It is fiction. You can tell it is fiction because no one cares enough about an advertising company to argue for hours over how code should be implemented. That's why the author never specifies the controversial statement. There's no such statement that could make 5 people do that.

There is some flaws and limitations to the AI story, but we already have self-improving thought germs: https://www.youtube.com/watch?v=rE3j_RHkqJc

Big if true.

Reply on news.ycombinator.com