Exploring how cognitive science can improve AI safety, governance and prioritization.
I’d be excited to intern for any research project.
Always happy to chat!
Daniel_Friedrich
Karma: 209
A new place to discuss cognitive science, ethics and human alignment
Exploratory survey on psychology of AI risk perception
When coming up with a similar project,* I thought the first step should be to conduct exploratory interviews with EAs that would reveal their hypotheses about the psychological factors that may go into one’s decision to take AI safety seriously. My guess would be that ideological orientation would explain the most variance.
*which I most likely won’t realize (98 %)
Edit: My project has been accepted for the CHERI summer research program, so I’ll keep you posted!
Daniel_Friedrich’s Quick takes
The core idea sounds very interesting: Increasing rationality likely has effects which can be generalized, therefore having a measure could help evaluate wider social outreach causes.
Defining intelligence could be an AI-complete problem, but I think the problem is complicated enough as a simple factor analysis (i. e. even without knowing what we’re talking about :). I think estimating impact once we know the increase in any measure of rationality is the easier part of the problem—for ex. knowing how much promoting long-termist thinking increases support for AI regulation, we’re only a few steps from getting the QALY. The harder part for people starting out in social outreach might be to estimate how many people they can get on board of thinking more long-termistically with their specific intervention.
So I think it might be very useful to put together a list of all attempts to calculate the impact of various social outreach strategies for anyone who’s considering a new one to be able to find some reference points because the hardest estimates here also seem to be the most important (e. g. the probability Robert Wright would decrease oversuspicion between powers). My intuition tells me differences in attitudes are something intuition could predict quite well, so the wisdom of the crowd could work well here.
The best source I found when I tried to search whether someone tried to put changing society into numbers recently is this article by The Sentience Institute.
Also, this post adds some evidence based intervention suggestions to your list.
What can an EA academic do to improve the incentives in the research side of academia? To help reward quality or even positive impact?
I’d love to see a deeper inquiry into which problems of EAs are most effectively reduced by which interventions. The suggestion there’s a lack of “skilled therapists used to working with intelligent, introspective clients” is a significant novel consideration for me as an aspiring psychologist and this kind of hybrid research could help me calibrate my intuitions.