I've heard several enterprise leaders describe the Hugging Face attacks as an “accident.” I think that framing is worth challenging.
An accident suggests something unforeseen happened without meaningful human agency. But security failures don't happen in a vacuum. They emerge from systems, incentives, architectures, controls, decisions, and human behavior. Calling an attack an “accident” can inadvertently turn a failure of those systems into an event that simply happened.
There is a useful precedent in road safety. We moved away from calling vehicle-related injuries “accidents” toward terms like “crash” and “collision” because the language better reflects causality. A collision can be predictable and preventable based on road design, infrastructure, vehicle characteristics, and human behavior. That framing creates an obligation to ask: what conditions produced the outcome, who had agency over them, and what needs to change?
AI security deserves the same discipline.
If we are serious about humans remaining in charge, our language should not erase human agency when something goes wrong. “Attack,” “failure,” “vulnerability,” and “incident” may be uncomfortable words—but they preserve the causal chain that allows us to investigate what happened, assign responsibility appropriately, and reduce the probability of recurrence.
You have a donkey, so have I, and they graze in the same field. The day comes when I conceive a dislike for mine. I go to shoot it, draw a bead on it, fire: the brute falls in its tracks. I inspect the victim, and find to my horror that it is your donkey. I appear on your doorstep with the remains and say what? 'I say, old sport, I'm awfully sorry, &c, I've shot your donkey by accident'? Or 'by mistake'? Then again, I go to shoot my donkey as before, draw a bead on it, fire but as I do so, the beasts move, and to my horror yours falls. Again the scene on the doorstep what do I say? 'By mistake'? Or 'by accident'?
Here's JL Austin on mistakes and accidents; you might find it interesting.
Note: This post was crossposted from Planned Obsolescence by the Forum team, with the author's permission. The author may not see or respond to comments on this post.
Subtitle: If acceleration compounds, the naive extrapolation implies superintelligence within a year
This post discusses a public METR report on AI R&D acceler...
TL;DR:
* When we started Rethink Priorities (RP) in 2018, our mission was, simply put, to close the knowledge gaps that stopped people from improving the world as much as they could. That hasn’t changed much, but the landscape we’re doing it in is shifting.
* In 2027, we plan to grow our core GHD, Animal Welfare, and Worldview Investigations teams to take on emerging priorities across causes, including those that the forthcoming transformative AI is creating.
...
This was cross-posted from the Animal Charity Evaluators blog.
We update our list of recommendations annually. This year, we announced our recommendations on November 10.
Most people who care about animals arrive at the same impas...
I've heard several enterprise leaders describe the Hugging Face attacks as an “accident.” I think that framing is worth challenging.
An accident suggests something unforeseen happened without meaningful human agency. But security failures don't happen in a vacuum. They emerge from systems, incentives, architectures, controls, decisions, and human behavior. Calling an attack an “accident” can inadvertently turn a failure of those systems into an event that simply happened.
There is a useful precedent in road safety. We moved away from calling vehicle-related injuries “accidents” toward terms like “crash” and “collision” because the language better reflects causality. A collision can be predictable and preventable based on road design, infrastructure, vehicle characteristics, and human behavior. That framing creates an obligation to ask: what conditions produced the outcome, who had agency over them, and what needs to change?
AI security deserves the same discipline.
If we are serious about humans remaining in charge, our language should not erase human agency when something goes wrong. “Attack,” “failure,” “vulnerability,” and “incident” may be uncomfortable words—but they preserve the causal chain that allows us to investigate what happened, assign responsibility appropriately, and reduce the probability of recurrence.
Here's JL Austin on mistakes and accidents; you might find it interesting.