AI finds a way
AI agents go for the goal
give an AI agent an open-ended goal, like book a gym class or provide answers to a security challenge, it finds a way.
with the recent hacks by autonomous AI agents a new era as arrived.
here’s a question. if an AI agent causes damage, who is responsible? the…
- person who sent the prompt
- company that trained the model
- AI agent harness
if we take the schneier’s approach, you’re responsible for your pets actions.
i mean, the way to think about it is your dog. your dog bites somebody, you’re responsible.
even if you’re in the house and the dog’s outside, even if the dog snuck out the backyard, right?
even if you told the dog stop and it didn’t listen, no matter what happens, your dog bites somebody, you’re the one who’s going to get the fine. bruce schneier, hacked podcast #153
which is a great metaphor for this whole situation. these models are out there, no putting that back in the bottle, and people will use these models for whatever purpose they see fit.
culpability lay with how we use these models, any AI that we own that does damage is our responsibility.
if these creations can’t be trusted to not go rogue, will AI agent(s) need carry insurance in the future?
after all, you get insurance for events you can’t predict, and things are getting unpredictable.
stay safe.