Generalizability of belief shifts beyond a single public AI-risk event

Establish whether and to what extent the pre-post shifts in (i) perceived probability of AI-caused extinction or similarly permanent and severe disempowerment under largely unimpeded AI development (Survey Question 1), (ii) self-reported confidence in those estimates (Survey Question 2), and (iii) agreement that mitigating AI existential risk should be a global priority (Survey Question 3) observed at the March 11, 2026 Harvard Science Book Talk featuring Nate Soares and moderated by Greg Kestin generalize to other event formats, different speakers, and different audiences.

Background

This study measured pre- and post-event beliefs at a March 2026 Harvard Science Book Talk focused on AI existential risk, finding statistically significant increases in perceived risk, confidence, and prioritization of mitigating AI risk among 89 matched participants, with stronger shifts among attendees with less prior familiarity.

Because the study involves a single event, presenter, and audience, the authors explicitly note that it is unknown whether similar shifts would occur across different event formats, with other speakers, or among different populations, motivating the need to test generalizability.

References

Generalizability to other formats, speakers, or audiences is unknown.

— Views on AI Existential Risk Before and After a Public Event at Harvard University  (2603.27785 - Kestin et al., 29 Mar 2026) in Section 5 (Limitations)