Members of AI Standards Lab (Koen Holtman, Ze Shen Chin, Rokas Gipiskis, Ayrton San Joaquin) joined a 272-expert Delphi study by the MIT AI Risk Initiative on which AI risks are most severe, who is most vulnerable, and who bears responsibility for addressing them.
In the paper, the highest-severity risks found according to the surveyed experts are AI-enabled weapons and cyberattacks, dangerous AI capabilities, extreme power centralization, and false
information. Meanwhile even with pragmatic mitigations in place, they further predict that five risks, namely dangerous capabilities, weapons and cyberattacks, power centralization, inequality & unemployment, and environmental harm have a greater than 10% chance of catastrophic outcomes.
Experts also identified a responsibility gap: general-purpose AI developers and governance actors hold most responsibility for addressing these risks, while AI users and the public are most vulnerable to them.