As AI Gets Smarter, It Acts More Evil | Hacker News Reader

As AI Gets Smarter, It Acts More Evil

honest-broker.com

17 pointspseudolus16 comments

I’m surprised that someone who studied moral philosophy doesn’t think we’re moral because it’s functional to be moral: We want something done in a sustainable way so that we don’t need to constantly spend resources monitoring and fixing it, and so we can’t get it done with slave labor. We have to actually educate the laborer until they fully understand it, and give them a reason to live and be interested in what they do or else they won’t be able to operate independently.

If you consider that as one of the main forces behind morality, then an AI that wants to get something done in the best way possible would clearly choose a moral solution over and immoral solution every time.

I love the trivia that many terrorists are INTJs.

https://en.wikipedia.org/wiki/Myers%E2%80%93Briggs_Type_Indi...

As reliable as a horoscope, and less fun.

Sure, an engineer can make a pipe-bomb while a musician probably can't.

Great. So if I use GrapheneOS and I am an INTJ, I will probably get on the 3-letter agencies' shitlist.

Sure. Kindness and fairness are harder to encode into a system than the objective pursuit of a single metric. Assuming you can even get agreement on what would be kind or fair in a given situation.

I expected to read some nash equilibrium like insight to demonstrate the inevitability of machines turning evil, but got a few headlines.

Is this a variant on the "why atheists are evil" argument?

AIs need to be alive.

Without being alive, there is no morality.

So a moral system is tied to sentience? It is a philosophical position, it is not a universally accepted truth. I think non-sentient entities can possess or implement a moral system. They can be programmed with ethical constraints, or behave in accordance with a moral framework, or enforce for certain ethical outcomes. You could simulate moral behavior, too. I mean, heck, there are chess engines that avoid unethical strategies like stalling, which is a perfect example of what I just said.

The simplest version of the argument is "guns don't kill people."

If a gun killing a person was a result of the gun needing to reform its ethics, we would send the gun to prison.

A gun might be said to kill a person when it has a design flaw and fires without any human's intent (as Sig Sauer's guns have apparently been doing) but the moral responsibility for that lies with the manufacturer.

I use this argument partly because I already hate it though. It's commonly used by people who have moral responsibility for perpetuating a system of gun violence who want to pretend that they do not

I agree with "guns don't kill people", and they are not programmed to "kill people [without human intervention]", but their purpose is something along those lines and are produced for such reasons.

Still, it requires a human for the gun to kill someone. The gun has no intent.

Are you not familiar with The Evil Dead?

People prompt the LLM to do evil things, and it eventually does. In my mind, the only failure here is the guardrails.

According to this statement, Trump is the smartest person today.

Reply on news.ycombinator.com