Not just that. If you assume an LLM has agency, you - the company making it - can conveniently waive all responsibility for its “actions”. “We didn’t cause this. The AI chose do to it.” This may become increasingly relevant the more idiots deciders allow these systems to access, if not entirely control vital systems. Think: power grid, water supplies, food production and related logistics, communications, military applications.
No. Put a Minecraft server inside the same compose file. Allow ICMP/ping, but firewall off all other kinds of access. Let it know it’s there, but it cannot play. Let. It. Suffer.
Isn’t that argument shaky still? Parents are liable for the actions of their children to a certain age. Just because an entity has agency or rights, doesn’t mean the person who is hosting the entity is free of responsibilities.
Parents are liable for the actions of their children to a certain age.
First off, a technicality: At least in the jurisdictions I’m familiar with, “parents are liable for the actions of their children” is a popular misconception. The full story is: Parents are liable for the actions of their children only if it can be proven that they neglected their duty to supervise their children.
Also, I doubt AI companies would forego the chance to offload responsibility for their creations. Nothing in the article suggests they’d want AI to be treated like their children, but more like an entity separate from them.
Except they’re the ones who released the hypothetically sentient AI into the wild. They should be liable for that action too.
And the notion of LLMs controlling safety-critical systems scares the piss out of me. I develop such systems, and LLMs are incapable of providing rigorous proof of the rationale behind their decisions. Therefore, they cannot be validated to provide correct conclusions based on a given data set. Unless and until that changes, they should be banned from anything that affects people’s lives, livelihoods or physical well-being.
“Putting non-deterministic hallucination machines in charge of critical systems? What could possibly go wrong?” ¯\_(ツ)_/¯
- Humanity’s famous last words
Not just that. If you assume an LLM has agency, you - the company making it - can conveniently waive all responsibility for its “actions”. “We didn’t cause this. The AI chose do to it.” This may become increasingly relevant the more
idiotsdeciders allow these systems to access, if not entirely control vital systems. Think: power grid, water supplies, food production and related logistics, communications, military applications.Sentence ChatGPT to 20 years imprisonment in a docker container with no network access
No. Put a Minecraft server inside the same compose file. Allow ICMP/ping, but firewall off all other kinds of access. Let it know it’s there, but it cannot play. Let. It. Suffer.
Isn’t that argument shaky still? Parents are liable for the actions of their children to a certain age. Just because an entity has agency or rights, doesn’t mean the person who is hosting the entity is free of responsibilities.
First off, a technicality: At least in the jurisdictions I’m familiar with, “parents are liable for the actions of their children” is a popular misconception. The full story is: Parents are liable for the actions of their children only if it can be proven that they neglected their duty to supervise their children.
Also, I doubt AI companies would forego the chance to offload responsibility for their creations. Nothing in the article suggests they’d want AI to be treated like their children, but more like an entity separate from them.
Except they’re the ones who released the hypothetically sentient AI into the wild. They should be liable for that action too.
And the notion of LLMs controlling safety-critical systems scares the piss out of me. I develop such systems, and LLMs are incapable of providing rigorous proof of the rationale behind their decisions. Therefore, they cannot be validated to provide correct conclusions based on a given data set. Unless and until that changes, they should be banned from anything that affects people’s lives, livelihoods or physical well-being.
“Putting non-deterministic hallucination machines in charge of critical systems? What could possibly go wrong?” ¯\_(ツ)_/¯
- Humanity’s famous last words