Discussion of AI risk has gone mainstream in the past few months, and as usual whenever that happens, many of the people new to the field think they’re an expert and proceed to engage in the exact same arguments over and over. I think it would be convenient to have the following available:
An introduction to the core risk, explaining why it’s possible.
An index of common bad arguments about AI risk and in-depth responses to them.
An index of common good arguments about AI risk, and links to further reading about them.
All of these should require no background knowledge and be accessible to normal people with no philosophical or mathematical experience.
I was thinking of writing up such a guide, but I don’t want to duplicate effort. Does anything like this already exist?
have you seen the stampy project https://aisafety.info/ although it is currently a work in progress. there was also some examples of it here https://www.lesswrong.com/posts/EELddDmBknLyjwgbu/stampy-s-ai-safety-info-new-distillations-2
Raemon endorsed the Superintelligence FAQ for laymen. He recommended a different one for ML engineers but I don’t know where to find that comment. This was a couple months ago so he might have found something even better since then.
I recommend Yudkowsky’s The Power Of Intelligence. It has superb quotes like “Intelligence is as real as electricity”, and in my experience, one of the biggest hurdles is convincing someone that AI actually does dominate all other calculations about the fate of the earth. Once you pass that hurdle, the person will be less likely to see it as a flex on your end, and more likely to see it as something worth their time to look into.
I also tentatively recommend Tuning Your Cognitive Strategies, as it lets people get an actual up-close look at what intelligence is. Plus, it’s very accessible for allowing people to contribute; any findings that anyone discovers might end up being pretty huge discoveries in the history of human intelligence augmentation (which is endorsed for potentially being an ace-in-the-hole for solving alignment, and anyone can contribute).