9
Nuclear war and malicious use of AI threaten end of humanity by 2100, says Lancet assessment
(www.healthdata.org)
It doesn't look like the article as a whole is downloadable without an account, but from what is accessible:
The Commission identifies two threats as having extinction-potential:
- The malicious use of artificial intelligence (AI)
- Nuclear conflict
I think that it's reasonably likely that by 2100, we will have at least two other potential candidates (if we aren't there already):
- Engineered, self-replicating biological organisms that create pandemics or other situations problematic to human life (e.g. if we produced something like the Great Oxidation Event on a short timeframe). We are increasingly able to craft and modify life, and I would expect that we will explore all sorts of possibilities.
- Self-replicating nanomachines that do similar; something like the gray goo scenario.
In all cases other than nuclear conflict, I think that rapid replication that spirals out of control is the common factor that could render a risk existential.
I also do not think that the most-likely existential risk from AI comes from intentional, malicious use—the article is specifically identifying malicious use—but from an error coupled with a system that can expand massively and spin out of control. Something more like the paperclip maximizer scenario.