31
Three Inverse Laws of AI - Susam Pal (susam.net)
submitted 4 months ago by to c/technology@lemmy.world
10 comments fedilink hide all child comments
 
you are viewing a single comment's thread
view the rest of the comments
[–] 2 points 4 months ago* (last edited 4 months ago) (1 child)

you have to have some awareness of what AI even is or you are at risk of being harmed.

Yes that's how it is, but that should not be the case, AI should legally be considered like asking expert advice, like asking a lawyer or a doctor, those are not considered risks, because they have legal responsibility for their advice. The same must be the case for AI, AI must have similar legal responsibility covered by the company offering the AI service.

If AI responses can't be trusted and are false information, it's not a service but a disservice. It can never be the case that normal users should have particular skills to use an AI service. That's legally a slippery slope we should absolutely refuse to allow.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 2 points 4 months ago

    AI should legally be considered like asking expert advice, like asking a lawyer or a doctor

    That is impossible. AI has no consciousness or ability to reason about the answers it is giving. Without a thinking domain expert between the user and the AI, this simply cannot be done.

    If AI responses can't be trusted and are false information

    AI response can't be trusted. Everyone should know that. There is no possible way to design a system where AI can be trusted. Hell, you can't even design a system where a human can be trusted to be infallible.

    You don't need a particular skill to use AI as a service — which isn't to say the results are going to be worth a crap if you aren't a domain expert. Anyone can ask AI to build a web service and get it approximately correct, but you need someone who knows how to build a web service to tell the AI what it needs to build it correctly.

    But one does need to understand that the output from AI should only be used when the marginal cost of failure is near-zero.

  • permalink
  • fedilink
  • source
  • parent