The apology identified the problem (AI can’t do that) and came up with an appropriate remedy (AI can’t do that).
It may be that the final version of the text was written by an AI, but it seems likely that a PR firm or crisis communications firm was used to defuse the situation. Those companies are used not just because they can generate text, but because they’re able to identify the problem, come up with a solution, and do it in a way that ends the news cycle. Those are skills that require empathy and understanding human nature, which is something AI doesn’t do.
So, the human needs to be an psychopath to not spot it and the people checking needs to have the job of spotting dumb ideas, to be able to see the issue. I tried running the idea past Ecosia’s AI (thinking mode) and it spots the issues right away plus other ones, including reputation harm to the company. Here is my prompt:
a company is having a party for young people where non-employees can come too. At the party, there will be a tattoo artist where everyone can get a tattoo that they choose themselves. if people get a tattoo, they will get offered a job interview. what do you think about that idea?
I’m not saying that AI can do anything a human can, but your claim that an AI can’t spot the issue without being told that one exist, is false.
So, you describe a situation in neutral terms and it is able to decide what’s ethical or unethical in that situation? Something that even humans are unable to do consistently?
Yeah. It’s often wrong about that – AI hallucinates about all kinds of things. But it definitely will try, and probably make a convincing effort at it.
I checked and it seems like the main difference with an AI suggested apology is the addition of gaslighting:
We did NOT offer instant interviews based on tattoos. That was never our intent. We brought a tattoo artist to our party as a fun, optional activity — like face painting or photo booths — to help break the ice. No one was pressured. No one was evaluated. And no one was hired because of a tattoo.
Or maybe they just used the feedback they got to craft the prompt. Or just fed the feedback directly into the llm and told it to craft an apology based on that.
Regardless, I don’t really care how they arrived at the soulless apology so I’m going to move on.
So… The prompt they used was more specific than “Write an apology?”
The apology identified the problem (AI can’t do that) and came up with an appropriate remedy (AI can’t do that).
It may be that the final version of the text was written by an AI, but it seems likely that a PR firm or crisis communications firm was used to defuse the situation. Those companies are used not just because they can generate text, but because they’re able to identify the problem, come up with a solution, and do it in a way that ends the news cycle. Those are skills that require empathy and understanding human nature, which is something AI doesn’t do.
Dude you’re so full of shit, it even identifies the fact it’s a party and people were drunk and can’t consent
By saying “What’s the moral issue here?” You indicated there was a moral issue to find.
I would bet that if you used that prompt for something where there were absolutely no moral issues, it would invent one.
But it didn’t invent an issue. It found the one talked about and several others.
Sure, and that’s pretty good. But, it’s not the same as identifying that there is an issue on its own.
It’s not like the people in the company saw the issue themselves, until they were told that there were an issue.
The psychopathic guy who posted that LinkedIn post didn’t see the issue. I doubt he passed the idea past the company’s lawyers or PR team though.
So, the human needs to be an psychopath to not spot it and the people checking needs to have the job of spotting dumb ideas, to be able to see the issue. I tried running the idea past Ecosia’s AI (thinking mode) and it spots the issues right away plus other ones, including reputation harm to the company. Here is my prompt:
I’m not saying that AI can do anything a human can, but your claim that an AI can’t spot the issue without being told that one exist, is false.
Perhaps you haven’t been using LLMs recently - they absolutely do that now.
So, you describe a situation in neutral terms and it is able to decide what’s ethical or unethical in that situation? Something that even humans are unable to do consistently?
Yeah. It’s often wrong about that – AI hallucinates about all kinds of things. But it definitely will try, and probably make a convincing effort at it.
I checked and it seems like the main difference with an AI suggested apology is the addition of gaslighting:
AI can if you go “and make sure to mention [INSERT PROBLEM HERE]” in the prompt.
And that entire apology is most definitely AI slop written, it has every telltale sign, such as using weird grammar choices, and looking sycophantic
Or maybe they just used the feedback they got to craft the prompt. Or just fed the feedback directly into the llm and told it to craft an apology based on that.
Regardless, I don’t really care how they arrived at the soulless apology so I’m going to move on.