Though invariably I like the people who say it, I don’t like it when people tell me they want to “work in AI policy,”^[1] then ask for career advice. “AI policy” doesn’t mean much on its own — the policy process is a pipeline, and what you should do depends on the part(s) of the pipeline you’re interested in.

The idea that I’m sketching out here — roughly, “there’s a whole ecosystem out there, from ideation to advocacy to implementation” — isn’t a new one and isn’t originally mine. Previous work has mapped the AI governance landscape much more comprehensively than I could: to name two, this EA forum post is a good intro, and this one has a super-detailed map.

Because I’ve found myself having this conversation with enough bright undergrads who “want to work in AI policy,” though, I thought this was still worth writing up.^[2]

What does “AI Policy” Mean?

You can think of policy, at a high level, as a process that roughly goes from abstract ideas to concrete implementation.

People have ideas, others translate those ideas into policy language, and still others then try to convince policymakers to implement them. There are advocates and lobbyists and organziers who try to influence legislators, to say nothing of the legislators and rulemakers themselves. Once laws or rules are implemented, someone has to enforce them.

The real deal is much less linear than that, of course — people can do multiple steps, you’ll sometimes cycle back around. There are also roles that touch the entire pipeline, like journalists, grantmakers, or field-builders.^[3]

Different roles deserve different paths

Most people who say they want to “work in AI policy” are actually drawn to one or two specific parts of this pipeline. Figure out which part(s), then test fit accordingly.

The career paths look different at each stage. Research roles need deep subject matter expertise and need to enjoy wonking out on details. Advocacy roles need political skills and relationship-building. Implementation roles need understanding of bureaucratic processes… and also relationship-building. You might not know whether you have or could develop those skills. That’s okay! it just means you should find a cheap way to find out if you do.

Talk to people who’re working in parts you’re interested in. Spend an hour doing what you’d do in that job and see if you hate it. Take 20 minutes to think through and fill out a memo on your career — I’m no Thomas Kalil, but I’ll leave comments if you share it with me!^[4]

Just be deliberate and really think through the ways you can help the most, instead of throwing darts at some shapeless blob you’ve labeled “AI policy.” Future you will thank you.

1. ^{^}

   I use “AI” here because that’s what I’m into — you can substitute any other kind of policy in for it, if you’d like.

2. ^{^}

   I’m also in the target audience, because writing helps me think.

3. ^{^}

   Again, this forum post is a good resource!

4. ^{^}

   Email a link to hi@henryjosephson.com. I can offer to do this publicly because almost nobody that I’ve offered this to has ever taken me up on it. People always promise to, and then never deliver. I’ve shamelessly stolen this from Michael Aird, who helps run RAND’s Technology and Security Policy Program, on the theory that I’d be less intimidating than he is.