Hacker News Reader

Comment by Animats | Hacker News Reader

The demand here for federal preemption of state law has nothing to do with copyright. Copyright is entirely federal level today. It has to do with preventing the use of AI to enable various forms of oppression.[1] Plus the usual child porno stuff.

What AI companies are really worried about is a right of appeal from decisions made by a computer. The EU has that. "Individuals should not be subject to a decision that is based solely on automated processing (such as algorithms) and that is legally binding or which significantly affects them."[2] This moves the cost of LLM errors from the customer to the company offering the service.

[1] https://calmatters.org/economy/technology/2024/09/california...

[2] https://commission.europa.eu/law/law-topic/data-protection/r...

View on HN

> This moves the cost of LLM errors from the customer to the company offering the service.

So does that mean AI companies are going to have insurance/litigators like doctors and models will be heavily lawyered to add more extensive guardrails. I'm assuming this means not just OpenAI but any service that uses LLM APIs or open models?

For ex: If a finance business pays to use an AI bot that automates interacting with desktop UIs and that bot accidentally deletes an important column in an Excel spreadsheet, then the AI company is liable?

Conversely, would you suggest that if an AI driver has a programming error and kills 20 people, that the person who reserved the car should be required to enter into a “User Agreement” that makes them take responsibility?

If it's a "self driving car" that the person "owns" - Yes. If it's a "Taxi service" that the person is using - No.

If it's a car they own, they (should) have the ability to override the AI system and avoid the accident (ignoring nuances) - therefore owning responsibility.

If it's a Taxi they would be in a position where they can't interfere with the operation of the system - therefore the taxi company owns the responsibility.

Rightly or wrongly, this model of intervention capability is what that I'd use to answer these types of questions.

Oh? And how exactly do you suggest one override the electronics system of a drive-by-wire vehicle with integrated AI control, specifically?

No, the exact opposite. This says that if the AI that a bank is paying for locks your bank account in error because your name sounds <ethnicity with a lot of locked bank accounts>, it's the banks problem to fix, not yours to just live with (entirely. You still likely have a problem).

How is that the opposite of what I said?

> not just OpenAI but any service that uses LLM APIs or open models

Reply on news.ycombinator.com