Emotional difficulties of AI safety research
listing out only the difficulties arising from the subject matter itself, without reference to the AI safety community
- Predicting the future is hard, predicting a future with futuristic technology is even harder
- AI safety has many prerequisites
- One cannot tinker with AGI safety because no AGI has been built yet
- Thinking about death is painful
- Abstract utilitarianish thinking can infect everyday life activities (see Anna Salamon's post about this for more details)
- AI safety contains some memetic hazards (e.g. distant superintelligence stuff, malign universal prior, most famously Roko's basilisk) -- I think this one affects a very small percentage (maybe 1% or less than that) of people who become interested in AI safety
