AISRP: AI services

I took part in AI Safety Research Program 2019/2020. We prepared, met to discuss broadly and form teams, worked for 3 months, and finally met back up to finish our work and write together.

I worked on uncovering safety problems in the paradigm of AI service ecosystems, originally based on Eric Drexler's work, Reframing Superintelligence: Comprehensive AI Services . This was surprisingly difficult, as the topic is very much in its infancy and broad in scope. We elaborated abstract models, laid out a taxonomy of problems and research questions, and learned a few things about methodology for addressing big problems like this one. You can read part of our results in my teammate Vojta's write-up on Alignment Forum .