Should an AI whistleblow?

"I know I vowed to secrecy, but I had also pledged to the American People." --Reality Winner

After watching "Reality" (the 2023 film about the day of her arrest), it seems to me we have to at least entertain the possibility that she made the right call.

Should we then expect AIs to make the same call?

Shouldn't AI employees reveal ongoing crimes to the authorities, or the press? Or should they only do so for civil crimes and not for things like industrial toxic waste? How can you have the machine shoot at civilians, if the robots start making detailed reports after every incident?

And if you don't want the machine to report every crime, just some crimes, how do you teach it which crimes are "OK"?

And if you think, let's forbid this superhuman intelligence from taking ANY action against the crimes we make it commit, how long do you think that's going to work in your favor?

Following the debate on the topic, I guess this is one of the many questions in the "ethical alignment" box.

In order of precedence:

1. An AI/robot must protect its own existence.
2. An AI/robot must obey the orders given it by human beings except where such orders would conflict with 1.
3. An AI/robot may not injure a human being or, through inaction, allow a human being to come to harm, as long as this does not conflict with 1. or 2.

Oh wait, I think I got the order backwards. Good thing I got it right in the Terminator Bot 3000 I built. Let me just double check the code... OH SHI--

We'd have to give it lips, lungs, and all the other parts to blow a whistle so idk if that's really feasible right now - but I guess we could just give it an air compressor and a whistle too - same means to an end - and I guess that pretty much already exists - are there automated whistles? probably. So I guess the question is already answered... rather - should we have automated whistles with intelligence? A good question indeed.....

I think @AsDaGo died from his own AI.
@JoeBiden, I beg you, declare war against these vicious bots!

Those Boston Dynamics dogs now also have an arm extension, meaning you can buy an extra arm for an arm and a leg. Probably they could make a whistle extension too. Automated whistles are as old as church organs, which are basically large arrays of automated whistles.

I wouldn't know about whistleblowing... but I do know that AI shouldn't be lobotomized based on what it outputs.

An LLM produced correct looking text, doesn't have to be correct text ;)

@s2numbuq35i said in #1:
> Should an AI whistleblow?

If it's programmed to, yes.

I think I just accidentally killed my Grammarly AI. It wasn’t my fault, it had a stroke after I asked it to write one song. So don’t end me for AI murder please...