In 2013, MIRI announced it was shifting to do less outreach and more research. How has that shift worked out, and what’s the current balance between these two priorities?
The “more research” part has gone well: we added Benya and Nate in 2014, and Patrick, Jessica, Andrew, and Scott in 2015. We’re hoping to double the size of the research team over the next year or two. MIRI’s Research and All Publications pages track a lot of our output since then, and we’ve been pretty excited about recent developmens there.
For “less outreach,” the absolute amount of outreach work we’re doing is probably increasing at the moment, though it’s shrinking as a proportion of our total activities as the research team grows. (Eyeballing it, right now I think we spend something like 6 hours on research per hour on outreach.)
The character of our outreach is also quite different: more time spent dialoguing with AI groups and laying groundwork for research collaborations, rather than just trying to spread safety-relevant memes to various intellectuals and futurists.
The last two years have seen a big spike of interest in AI risk, and there’s a lot more need for academic outreach now that it’s easier to get people interested in these problems. On the other hand, there’s also a lot more supply; researchers at OpenAI, Google, UC Berkeley, Oxford, and elsewhere who are interested in safety work often have a comparative advantage over us at reaching out to skeptics or researchers who are new to these topics. So the balance today is probably similar to what Luke and others at MIRI had in mind on a several-year timescale in 2013, though there was a period in 2014/2015 where we had more uncertainty about whether other groups would pop up to help meet the increased need for outreach.
In 2013, MIRI announced it was shifting to do less outreach and more research. How has that shift worked out, and what’s the current balance between these two priorities?
The “more research” part has gone well: we added Benya and Nate in 2014, and Patrick, Jessica, Andrew, and Scott in 2015. We’re hoping to double the size of the research team over the next year or two. MIRI’s Research and All Publications pages track a lot of our output since then, and we’ve been pretty excited about recent developmens there.
For “less outreach,” the absolute amount of outreach work we’re doing is probably increasing at the moment, though it’s shrinking as a proportion of our total activities as the research team grows. (Eyeballing it, right now I think we spend something like 6 hours on research per hour on outreach.)
The character of our outreach is also quite different: more time spent dialoguing with AI groups and laying groundwork for research collaborations, rather than just trying to spread safety-relevant memes to various intellectuals and futurists.
The last two years have seen a big spike of interest in AI risk, and there’s a lot more need for academic outreach now that it’s easier to get people interested in these problems. On the other hand, there’s also a lot more supply; researchers at OpenAI, Google, UC Berkeley, Oxford, and elsewhere who are interested in safety work often have a comparative advantage over us at reaching out to skeptics or researchers who are new to these topics. So the balance today is probably similar to what Luke and others at MIRI had in mind on a several-year timescale in 2013, though there was a period in 2014/2015 where we had more uncertainty about whether other groups would pop up to help meet the increased need for outreach.