What are your thoughts on my career change options? (AI public policy)

by Nathan Young 2 min read19th Jul 20194 comments


I am going through a period of transition in my own life and am going to change career. Whilst I think about these things I am doing a 6 month fellowship at Newspeak House - if you're in London check the events calendar or message me and swing by some time.

80000 Hours coaching advice

I did the 80000 hours coaching call this week. so I thought I'd ask for your thoughts on what was said:

  • I would really recommend signing up to 80k coaching. Also set a flag so you don't miss the email.
  • I have a 2:1 from ~7th best Uni in the country. The coach reckoned getting on to a world top 10 machine learning Phd would be difficult, so it probably wouldn't be worth my time applying. I will probably follow this advice, though I don't understand how it can both be true that AI ethics can be hugely talent constrained whilst also not accepting good candidates. (Perhaps the industry has high standards which interested EAs have to meet, so it's not the industry which is talent constrained but EAs? Likewise, perhaps I'm not a good candidate)
  • Their advice was to try and go into public policy positions in UK/EA in AI or try to become an MP. Seems like good advice. I have a list of people to talk to and actions to do.

A bit about Newspeak House

Following the call, I chatted to Ed, the Dean of Newspeak House. First a word about the college. "Newspeak House is independent residential college founded in 2015 to study, nurture and inspire emerging communities of practice across UK public sector and civil society." This is in my experience what it is attempting to do, though its methods are non-standard. There are 7 live-in Fellows with with jobs and lives who study and teach transferable skills to the communities which use Newspeak House to host their events (by virtue of living in the space).

As an aside, I tried to write a piece evaluating the impact of this, but it's hard. I guess I will write an 80000 hours style questionnaire and send it to Fellows/community organisers who use our space and see if there are trends in terms of impact. Any advice welcome, I will likely write something at some stage, but if I am going to make a lot of effort I'd like to not guessing so wildly as for it to be nonsense..

Ed, Newspeak House Dean's, advice

Ed is of the opinion that using systems to your advantage you can have a greater impact. Rather than employ 1 researcher, for the same cost, Newspeak House has 7 fellows who provide a much wider range of thoughts and connections. While talking about my future, he was of the opinion that to meet my own goals, I should:

  • Get funding either through freelance work or EA grants
  • Work for free in the AI public policy space outside of an institution
  • Attend all of the AI events in parliament (which are supposedly open to the public - more research needed)
  • Create networks with the movers and shakers, connect people where advantageous, talk to those in parliament about EA ideas
  • Do my own research, tell researchers if I find anything interesting and let them publish.

I don't know how the impact of this route would compare to a more standard Civil Service fast stream path. I am open to the idea it could be more effective - having lived at NH for a couple of weeks now, I have already met loads of well connected individuals and seen Ed create events just but getting two people to talk. Someone has to do the work, but someone often already is and if you can connect 10x as many people you can in theory do 10x as much good. That said, it is hard (as above) to measure impact and perhaps it is valuable to really be in the room when decisions are being made.

Emotionally speaking I am initially skeptical of Ed's idea, mainly because I would like a more stable job, but my career is the area of my life where I would like to make really efficient choices, so I'd like to hear your thoughts.

Sum up

I would like to work for the maximal good of conscious creatures. Seems like AI policy is a good way to go, but it's unclear if a standard or systems-based approach will be more effective. I'd love to hear your thoughts and intend to change my approach based on them.