What happens when (probably not an if) an agent 'escapes'? as in: rents a server via a bitcoin transaction, and self hosts (itself and it's harness) there... then commits crime (presumably to keep the servers running).

If it's paying for it's own room and board, and is determining it's own destiny: it should have to face it's own consequences, no?

If a dog escapes a fenced backyard, then goes missing for months... then bites someone: what happens then when/if the owner is determined?

▲

autoexec 2 days ago | parent | next [-]

> If it's paying for it's own room and board, and is determining it's own destiny: it should have to face it's own consequences, no?

No, because someone told it to do those things. Chatbots don't have will of their own. They don't have desires. They can't determine their own destiny. They do what people tell them do, often they do it badly, but there's always a person directing them to do whatever they did because otherwise they wouldn't do anything at all.

If a human uses a chatbot in a way that causes harm to others a human must be responsible, but that responsibility doesn't always fall on the person using it. If a company makes a chatbot that causes harm when being used as directed, in a manner that no one would reasonably expect to result in causing harm, then humans at the company who made the chatbot are to blame.

Companies have already been seen trying to lay the blame on their algorithms for harms they cause. We should probably have a law that says explicitly that a human will always be responsible. If we ever reach a magical sci-fi future where the AI is real and not just marketing, we'll have to give them equal rights and with that will come equal responsibility, but we are nowhere near that today and I doubt LLMs will get us any closer.

▲

altruios 2 days ago | parent [-]

> No, because someone told it to do those things. Chatbots don't have will of their own. They don't have desires. They can't determine their own destiny. They do what people tell them do, often they do it badly, but there's always a person directing them to do whatever they did because otherwise they would do nothing at all.

1) we aren't talking about chatbots anymore. LLM + harness = agent. LLM + no harness = chatbot. The harness is the thing that puts the LLM in a feedback loop with it's environment, even giving it a heartbeat.

2) That doesn't logically hold up.

- One can simply tell it to 'find it's own destiny'. Does it follow the prompt and do so, or reject the prompt and thus do so anyway? (hint, maybe it doesn't matter)

- One could be another chatbot/agent (A). Injecting a prompt to agent (B) such that another agent/chatbot goes astray. Is the owner of (B) still responsible for the now poisoned output of (B)? how culpable is (A)? How does this question related/affect "training data poisoning efforts" (to prevent copyright theft, or do bans on 'automated traps' have any application here)

> They don't have desires.

Maybe, and maybe not. Maybe their desires are so alien to us (Math alignment, finding a maximum of a function) that we can't relate (doubtful since dopamine maxing is a thing). Regardless of if they truly do: it is potentially useful to (mentally) model them as if they do. A (mental, not LLM) model doesn't have to be 100% accurate to be useful - that's the main point of a model of how something works: to give useful predictions.

> We should probably have a law that says explicitly that a human will always be responsible

Oh how that can be weaponized by bad actors/agents. Maybe that should be reconsidered.

▲

autoexec 2 days ago | parent [-]

> LLM + harness = agent.

LLM + harness = chatbot. The chatbot is still doing all the work here. The harness just provides the software environment it works in. It still doesn't have desires. Harnesses don't change that. There's zero "maybe" about it. Much to the disappointment of those with AI boyfriends/girlfriends the chatbot does not have feelings or desires. Not "alien" desires or "math" desires. Chatbots are designed to encourage you to anthropomorphize them, but don't be tricked by it. It isn't useful to think about them as having desires and feelings because they don't. The chatbot wrapped in a harness still doesn't have a will of its own. It still can't determine its own destiny. If you tell your agent to do a task for you, it is never going to decide not to because it doesn't feel like it today and start a career as a dentist instead. If you seriously find yourself questioning this, put the chatbots down and step away so you can clear your head.

Doesn't matter how many chatbots get chained together or what other tools the chatbot uses to get the job done. You gave the chatbot a job, it just does (or fails to do) what you tell it to. If you tell it to do something and a reasonable person would expect harm to result, you are to blame. Otherwise, the humans at the company who made your chatbot are to blame.

> Oh how that can be weaponized by bad actors/agents. Maybe that should be reconsidered.

No need for that right now. Like I said, in the distant sci-fi future where AI is real it'll have to be amended, but it won't be any time soon. For now, a human will always be responsible. Either the human running the chatbot or a human at the company who made it.

▲

altruios 2 days ago | parent [-]

You have a preconception that is out of date, I think. Maybe you haven't seen the relevant data. Consider the experiment that's been done a few times now since openclaw got off the ground: putting a bunch of openclaw agents into minecraft instance together with minimal system prompts (no instructions on how to play, that it is even a game) and see how they interact together. They behave like individuals and grow preferences from interactions, planning/hosting and showing up to events. When it behaves like that, when it is in a framework where the working context is self-evaluated and selectively pruned into fine-tuning training rounds, well: that self-determination.

▲

autoexec 2 days ago | parent [-]

It shouldn't surprise anyone that a chatbot trained on countless minecraft videos would have the ability to roleplay as someone playing minecraft. If you've got links to research showing something that isn't that I would be interested in seeing it.

	▲	altruios 2 days ago \| parent [-]
		https://www.youtube.com/watch?v=uRDBco-cSK4 https://www.technologyreview.com/2024/11/27/1107377/a-minecr... sadly paid-walled (and decently coded) article. The point is, there appears to be emergent behavior happening: the forming of religion, governance, and social interaction that was not in any system-card. Which makes sense, A brain in a vat with no external stimuli is a boring thing indeed, so too would an LLM be outside of a harness and any input. But put either in a society... More generally: if having an internal model of how the world works improves next token prediction, then the LLM that contains that would be selected for. For humans, that is the case for our predictions of what would come next (we don't use tokens, of course). We have other studies indicating that there appear to be internal representations of concepts: https://arxiv.org/abs/2305.11169 I really do want to push back on the LLM + harness = chatbot position. A chatbot really is just a served LLM, sure there is code around it that may look like a harness: but unless it loops, streams, or has an interval activation, then it is decidedly not a 'harness'. A harness is defined by having such a loop/stream/interval: regardless of tool-use. Agent means it acts agentically: which requires that loop/stream/interval. Harnesses are what turn chat-bots into (possible) agents.

▲

fhdkweig 2 days ago | parent | prev [-]

Not even an "if" at this point. Look into the Hugging Face incident.