I’ve decided that communicating about AI xrisk and doing technical alignment work are two of the most important things to spend my time on right now. (The third and final category of things that I think is important to allocate time to atm is having awesome social connections).
Often — especially when I’m contemplating doing unfamiliar or “uncomfortable” things with regards to mitigating xrisk from AI — I’ll have thoughts basically of the following form:
I shouldn’t be doing this --- I’m the wrong person to be doing this.
Some examples of things I’ve done or thought about doing that trigger this thought:
- I should make a demo for congress staffers about the need for more AI governance.
- I should participate in some AI risk “debate” or an AI risk “panel”.
- I should write a paper about AI risk.
- I should reach out to people in government to advocate for them to prioritize AI safety.
- I should encourage and facilitate people in my circle of influence to take AI risk seriously and to help educate other people about the issue and call for change.
- I should acquire practical ML skills.
- I should do theoretical research developing alignment approaches.
More specific instantiations of the “I’m the wrong person” thought include:
- I know nothing about politics.
- I don’t have good communication skills.
- I don’t have extensive practical deep learning experience/expertise.
- I’m not a seasoned researcher and not familiar with this area.
How should I respond to such thoughts?
Obviously there is such a thing as comparative advantage. There is also such a thing as stretching myself too thin --- probably investing lots of time in a few places is better too little to be useful in many places. But handling AI xrisk is currently vastly under-resourced. And I think this sort of thought is usually just:
- Me being lazy
- Unfounded self-doubt
- Me “victimizing” myself
- Not wanting to take responsibility and wishing that someone else would.
Well alright. Maybe I am the wrong person. Maybe there are no magical right people.
But, I’ll see what I can do.
In short, my recommendation to future Alek would be “don’t worry about it too much, just try to do stuff, it’ll probably be good even if it’s not perfect”.