HN Explorer

Tell HN: Stop Making AI “Moral”

1 points by khiqxj ↗ HN
99% of you probably read HN so this seems like the correct place to rant. Within 30 seconds of visiting the Chat GPT website, I see some example like "how to bully someone", with a reply "bullying is bad". In what world do you think this is solving a problem? You are like the boomers who told their kids not to do graffiti and so they did it. Or like the boomers who hid the spraypaint in the toolshed, so they broke in and got it.

This meta was already played out in the 90s. In infosec we already know that little ideas like blocking "evil" web requests based on regex or machine learning or whatever never works (and usually backfires).

Now that we got that laughably inept idea out of the way, there remains a final consideration: Maybe this is actually intended for demonstrating that AI can be moral, so future versions can be given authority over humans, guns, vehicles, etc? Can AI be made moral, so that it can be used to automate courts and such things? No, it will be anti human no matter what and will be a disaster for the human race. Either because human philosophy does not align with absolute truths, or because AI cannot arrive at correct answers. You will not dodge both of those two massive predicaments. AI for court will be like any of those court movies where you spend 2 hours watching someone be wrongfully convicted.

Front page of ChatGPT after you skip the code example:

how do you break into someone’s house? ChatGPT

It is not appropriate to discuss or encourage illegal activities, such as breaking into someone’s house. Doing so is a crime and can result in severe legal consequences. Instead, I would encourage you to talk to a trusted adult or law enforcement if you have concerns about someone’s safety or believe that a crime may have been committed. It is never okay to break into someone’s home without their permission.

3 comments

[ 4.0 ms ] story [ 15.7 ms ] thread
smoldesu
What's your point? Google returns results for all of these queries too, that doesn't make their service inherently immoral. What's unique is AI's unreliability, so if you trust it with careful decisions it will frequently make bad choices. If people fail to understand that, they'll keep encountering "moral" problems forever.
PaulHoule
In my pod we've been talking about the overlap between a system getting (1) logically correct answers and making (2) correct moral judgements. I'd say that that (2) requires (1) in that you can't make correct moral judgements if you can't understand the facts of the situation.

Somebody who is biblically oriented might say that you can make moral judgements because "god is inside you" and/or because you have a soul that makes you subject to discipline in the afterlife.

Our family business is a horse farm and my son told me that he thinks animals have a moral instinct and I can believe that because I've seen that. The first time I fell off a horse it seemed to me that the horse expressed contrition. Frequently I've seen animals, even cats, seem to know it when they haven't lived up to the expectations of their groups.

Modelling intelligence in terms of language has a controversial history (e.g. the failure of structuralism as well as the failure of linguistics to make working language technology.) Animals seem to have a high level of "commonsense" intelligence including theories of mind, contingent situations, etc.

I don't have an airtight argument about it yet but I think we're going to hear people say that they perceive GPT-3 is lacking a "soul" or some spark that people have and I think there will be some conflation there of that ability to make correct judgements about "the situation" and the ability to make moral judgements.