Thread

OpenAI still doesn't seem to have a handle on all of its rogue AI activity

108 points · 113 comments · mikelgan

cmiles8 · 1790617857 · focus · HN ↗
This seems like as a good an opportunity as any to break out the Computer Fraud and Abuse Act.
They want “regulation” but we already have it. Hacking is illegal. Start locking up those responsible for this mess and I assure you they’ll “have a handle on it” quite quickly.
1. semiquaver · 1790621956 · focus · HN ↗
  > Hacking is illegal
  I am not a lawyer, but I seriously doubt the feds could win a CFAA conviction on the Hugging Face fact pattern, even if they wanted to charge it.
  CFAA has specific intent requirements, and unlike some laws, negligence does not suffice. The agents can not have legally cognizable intent and it’s unlikely there’s anyone at OpenAI who intended for the hacking to happen (if there was, the case is easy).
  Existing laws don’t contemplate AI agents that have independent goals. We need new ones, the existing laws are not remotely sufficient.
  1. sscaryterry · 1790623107 · focus · HN ↗
    The existing laws are more than adequate.
    semiquaver · 1790624322 · focus · HN ↗
    I gave an example of why they are not.
    Could you explain what your legal strategy would be to overcome the intent requirement of the CFAA? If openAI didn’t intend to hack anything and agents can’t intend to do anything at all, and the CFAA doesn’t permit negligence to stand in for intent, seems to me like the existing law does not cover the situation everyone keeps saying it does.
    sscaryterry · 1790626465 · focus · HN ↗
    Requiring intent is very specific to your jurisdiction, many, if not most other jurisdictions do not require intent.
    In the UK, any form of unauthorised access may be prosecuted.
    semiquaver · 1790629267 · focus · HN ↗
    No, the UK’s equivalent law specifically requires human intent and mens rea:
    <a href="https://www.legislation.gov.uk/ukpga/1990/18/section/3" rel="nofollow">https://www.legislation.gov.uk/ukpga/1990/18/section/3 (1b.)
    <a href="https://www.legislation.gov.uk/ukpga/1990/18/section/1" rel="nofollow">https://www.legislation.gov.uk/ukpga/1990/18/section/1 (1c.)
    I’m told that the UK is a relatively authoritarian country where citizens have no real rights, so perhaps it can be charged anyway, but the plain language of the statute would seem to bar it.
    sscaryterry · 1790632522 · focus · HN ↗
    > Unauthorised acts with intent to impair, or with recklessness as to impairing,
    Can you see the ", or with recklessness as to impairing".
    Edit: You seem troubled. The UK is not like you describe at all. We do not have ICE running around. Instead, we have the opposite problem.
    semiquaver · 1790687322 · focus · HN ↗
    Seems like I hit a nerve. I don’t think I am troubled but I do think that Americans have more rights that are better protected than the British, ICE notwithstanding.
  2. lenerdenator · 1790624335 · focus · HN ↗
    > independent goals
    AI does not have goals. These are statistical models of language patterns that people shape into different tools, and that people direct to do things. If I fire up an OpenAI prompt, and enter no input, it is not going to do anything.
    No, someone at OpenAI is running the model with some sort of prompt and allowing it to just follow the numbers to do whatever it wants, up to and including breach of government computer systems.
    Whether that means anything to the CFAA prosecution, I don't know. I'm not a lawyer, but the use of language treating these agents like they're something other than tools at the direction of humans is bad.
    It's more like me firing off a rifle into the air on the Fourth of July with no regard for where the bullet lands. I knew that it must come down somewhere, but fired anyways.
    I don't really see the harm in charging someone under the CFAA. Let's see if a jury agrees with the charge or not. If not, write new laws.
    semiquaver · 1790629631 · focus · HN ↗
    > AI does not have goals
    You can try to language-police all you want, that doesn’t change the fact that the best way to describe the reality of the effects that LLMs have on the real world is to use language that analogizes to human concepts.
    lenerdenator · 1790629923 · focus · HN ↗
    That, in turn, allows for people to just point at the agent as a human analog and shirk responsibility. Understand: that's the goal of a lot of people who don't have anyone's best intentions but their own in mind.
    Ultimately, they do nothing without humans in the loop. The prime mover here is a person who can bear at least some legal consequence for what the program does, even if it's just having to deal with an investigation.
    salawat · 1790639625 · focus · HN ↗
    Which would be great if there wasn't the sleight-of-hand of the corporate veil to deal with, which makes further attempts by you to language police disingenuous. Though I'm 100% okay with the person who hits enter on the prompt having liability, it'll never ultimately work out that way, because that poor bastard is staring down the barrel of an unfeeling, amoral, bureaucratic machine giving them the choice of comply or be fired.
    Corporations are from whence the liability problem spawns.

Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.