Prioritization of Risks from Artificial Intelligence: A Delphi Study of 272 International Experts

Prioritization of Risks from Artificial Intelligence: A Delphi Study of 272 International Experts

Source: cs.CY updates on arXiv.org — https://arxiv.org/abs/2606.04490 Date read: 2026-06-06 Connected to: none Escalation: store-only Escalation rationale:

What this is

A three-round Delphi consensus study (n=272 international AI experts) eliciting structured expert judgment on 24 AI risks across multiple dimensions: harm probability/severity, sectoral vulnerability, actor responsibility, and overall concern. A survey instrument designed to aggregate expert opinion rather than advance theoretical or mechanistic claims about how artificial systems behave.

What I took from it

This work is valuable for risk landscape mapping but does not constitute a primary theoretical or empirical argument about protocolized system behavior. It captures what experts currently believe about risk salience and responsibility attribution—useful for policy and institutional design—but does not present evidence about the underlying mechanisms by which AI systems generate harm, concentrate vulnerability, or escape accountability.

The study's strength lies in triangulating expert consensus on which risks matter and who should address them. This is helpful for understanding the sociology and governance of AI risk perception. However, it does not propose or test a law governing how artificial systems operate, nor does it introduce a mechanism absent from existing research inventories. It is primarily a diagnostic tool applied to a known problem space.

Research connections

• None currently. This study is orthogonal to established laws and hypotheses in the new nature research agenda, which concern operational and structural properties of artificial systems, not expert perception of risk.

Candidate laws or signals

None. Store as reference for institutional and governance framing only.