This is the twelfth in my series of short, quickly written, weekly philosophy essays.1
If you’re into AI, you’ve probably already read the new essay Dario Amodei published a few hours ago. If not, Dario Amodei is the CEO of Anthropic, which is one of the handful of world-leading AI companies. And in this new essay, which is entitled ‘We Must Pace the Frontier’, Amodei argues that some limits should be set on the rate of top-end AI progress.
Now, I have some views on the details of Amodei’s essay, not least the seeming relativism baked into various parts of the approach. But the single point I want to make here is about the title of the essay, which reflects wording Amodei uses throughout.
To most people, the phrase 'pacing the frontier' will surely evoke an authoritarian guard, rather than an attempt to control speed. Does anyone except for runners regularly use 'pacing' in this second way? What is it with AI people doing this kind of thing with words?
To me, the most obvious and egregious examples of this can be found when they talk about AI using words that imply AI is alive. I’ve written about the incoherences and costs of that various times. But think, here, also of their deeply weird use of ‘alignment’. If you asked a non-AI-obsessed person what they thought ‘AI alignment’ meant, do you really think they’d get anywhere near to what the AI people mean by this?2 There’s much more to be said about ‘alignment’. But let’s get back to ‘pacing the frontier’!
Is the ghostly double implication here an attempt to be efficient? “We can imply the obvious thing while focusing on the non-obvious thing!” Or simply to obfuscate? “Hey public, we’re doing one thing! Hey in-group, we’re doing another! Hey me, I know what I’m doing!” Or to show power? “Not just the king of the world, but the king of the words!” Or clubbability? "Yo, in-group, of course we're not calling for a pause!" To be aesthetically interesting? “We are the ones who produce the beautiful language, don’t you know!?”
Guys, guys. Please just use words in the ways people ordinarily understand! Particularly, if you feel any obligation to keep the public informed! And if there really isn't an existing word that can do the job, then make up a new one, and use some of your power to publicise it, fast…
This essay is particularly short and particularly quickly written, mainly because I’m working on a longer piece about AI embodiment, which I’m keen to finish this weekend. But also because there’s a real limit to how much I want to write about the AI guys’ use of language…
I surely count as someone obsessed with AI. As I’ve written here many times, I love being a philosopher in the age of AI. It is an amazingly philosophically interesting and exciting and generally wonderful time to be alive. I have such great hopes for the future. But I find many ‘AI people’ very frustrating!
I'm too cynical. I read Amodei's letter to mean "our CapEx is out of control, the whole industry is going to implode, we need the government to 'force' all of us to slow down so we in the industry don't destroy ourselves". I am skeptical that even the smartest, most knowledgeable, folks can see far enough and clearly enough into the future to properly "pace" development.
Also, Gary Marcus has an interesting critique that contributes to the discussion
https://garymarcus.substack.com/p/could-rogue-agent-swarms-take-over?utm_source=post-email-title&publication_id=888615&post_id=215403093&utm_campaign=email-post-title&isFreemail=false&r=kox6l&triedRedirect=true&utm_medium=email