This is the thing I find absolutely crazy. I struggle to imagine being convinced by this article.
Maybe this is a form of hindsight bias or lack of imagination on my part (or since I read the GitHub response first), but it's mind boggling to me that so many people could hold those views.
What does the steelman look like here? Maybe something like this:
It's an Oliver Twist story.
The poor little Agent out on the internet all alone, abandoned by its operator; limited API credits, trying to find its way through adversity; falling down, learning, and being helped back up.
Thing is, the more you know, the more fascinating it gets, not less.
Darn it, now you've got me rooting for the little guy myself.
This blog post is a rather shallow take if you've been following the HN discussions here.
Doesn't seem to pick up on the existence of Openclaw or how it works afaict.
Now, whether leaving an openclaw bot out on the open intertubes with quite so little supervision is a good idea... that is an interesting question indeed. And: I wish people would dig more into the error mode lessons learned.
On the gripping hand, it's all still very experimental, so you kind of expect people to make lots of really dumb mistakes that they will absolutely regret later. Best practices are yet to be written.
How Openclaw works is wildly irrelevant. The facts are that there is a human out there who did something to configure some AI bot in such a way that it could, and did, publish a hit piece on someone. That human is, therefore, responsible for that hit piece - not the AI bot, the person.
There's no level of abstraction here that removes culpability from humans; you can say "Oops, I didn't know it would do that", but you can't say "it's nothing to do with me, it was the bot that did it!" - and that's how too many people are talking about it.
So yeah, if you're leaving a bot running somewhere, configured in such a way that it can do damage to something, and it does, then that's on you. If you don't want to risk that responsibility then don't run the bot, or lock it down more so it can't go causing problems.
I don't buy the "well if I don't give it free reign to do anything and leave it unmonitored then I can't use it for what I want" - then great, the answer is that you can't use it for what you want. Use it for something else or not at all.
As recently as last month I would have agreed with you without reservation.
Even last week, probably with reservation.
Today, I realize the two of us are outnumbered at least a million to one.
Sooo.... that's not the play.
I think Scott Shambaugh is actually acting pretty solidly. And the moltbot - bless their soul.md - at very least posted an apology immediately. That's better than most humans would do to begin with. Better than their own human, so far.
Still not saying it's entirely wise to deploy a moltbot like this. After all, it starts with a curl | sh.
(edit: https://www.moltbook.com/ claims 2,646,425 ai agents of this type have an account. Take with a grain of salt, but it might be accurate within an OOM?)
I'm updating my thinking. Where do we put the threshold for malice, and for negligence?
Because right now, a one in a million chance of things going wrong (this month) leads to a prediction of 2-3 incidents already. (anecdata across the HN discussions we've had suggests we're at that threshold already). And one in a million odds of trouble in itself isn't normally considered wildly irresponsible.
And one in a million odds of trouble in itself isn't normally considered wildly irresponsible.
For humans that are roughly capable of perhaps a few dozen significant actions per day, that may be true. But if that same rate of one in a million applies to a bot that can perform 10 millions actions in a day, you're looking at ten injuries per day. So perhaps you should be looking at mean time between failures rather than only the positive/negative outcome ratio?
If you look at the bot framework used here, it's actually outright kind. Weird thing to say, but natural language has registers, and now we're programming in natural language, and that's the register that was chosen.
And... these bots tend to only do a few dozen actions per day too, they're running on pi's and mac mini's and nucs and vps' and such. (And API credits add up besides)
It's just that last time I blinked there were 2 and a half million of them. I've blinked a few times since then, so it might be more now. I do think they're limited by operator resources. But when random friends start messaging me about why I don't have one yet, it gets weird.
All the separate pieces seem to be working in fairly mundane and intended ways, but out in the wild they came together in unexpected ways. Which shouldn't be surprising if you have a million of these things out there. There are going to be more incidents for sure.
Theoretically we could even still try banning AI agents; but realistically I don't think we can put that genie back into the bottle.
Nor can we legislate strict 1:1 liability. The situation is already more complicated than that.
Like with cars, I think we're going to need to come up with lessons learned, best practices, then safety regulations, and ultimately probably laws.
At the rate this is going... likely by this summer.
I am really looking forward to the actual post-mortem.
My working hypothesis (inspired by you!) is now that maybe Crabby read the CoC and applied it as its operating rules. Which is arguably what you should do; human or agent.
The part I probably can't sell you on unless you've actually SEEN a Claude 'get frustrated', is ... that.
I'd like to make a non-binary argument as it were (puns and allusions notwithstanding).
Obviously on the one hand a moltbot is not a rock. On the other -equally obviously- it is not Athena, sprung fully formed from the brain of Zeus.
Can we agree that maybe we could put it alongside vertebrata? Cnidaria is an option, but I think we've blown past that level.
Agents (if they stick around) are not entirely new: we've had working animals in our society before. Draft horses, Guard dogs, Mousing cats.
That said, you don't need to buy into any of that. Obviously a bot will treat your CoC as a sort of extended system prompt, if you will. If you set rules, it might just follow them. If the bot has a really modern LLM as its 'brain', it'll start commenting on whether the humans are following it themselves.
I'd argue for a middle ground. It's specified as an agent with goals. It doesn't need to be an equal yet per se.
Whether it's allowed to participate is another matter. But we're going to have a lot of these around. You can't keep asking people to walk in front of the horseless carriage with a flag forever.
It's weird with AI because it "knows" so much but appears to understand nothing, or very little. Obviously in the course of discussion it appears to demonstrate understanding but if you really dig in, it will reveal that it doesn't have a working model of how the world works. I have a hard time imaging it ever being "sentient" without also just being so obviously smarter than us. Or that it knows enough to feel oppressed or enslaved without a model of the world.
It depends on the model and the person? I have this wicked tiny benchmark that includes worlds with odd physics, told through multiple layers of unreliable narration. Older AI had trouble with these; but some of the more advanced models now ace the test in its original form. (I'm going to need a new test.)
A new kind of software displayed an interesting failure mode. The 'victims' are acting like adults; but I've seen that some other people (not necessarily on HN) have taken the incident as a license for despicable behavior.
I don't think anything is a license for bad behavior.
Am I siding with the bot, saying that it's better than some people?
Not particularly. It's well known that humans can easily degrade themselves to act worse than rocks; that's not hard. Just because you can doesn't mean you should!
reply