Existential risk narratives about AI do not distract from its immediate harms

Area of Science:

• Computer Science
• Social Science
• Risk Analysis

Background:

• Artificial intelligence (AI) presents multifaceted risks, with ongoing debate regarding their nature and urgency.
• Divergent narratives exist: one emphasizing long-term existential threats to humanity, another focusing on immediate societal concerns like algorithmic bias.
• A key debate is whether the 'existential risk' narrative distracts from present-day AI dangers.

Purpose of the Study:

• To empirically investigate the 'distraction hypothesis' concerning AI risk narratives.
• To determine if focusing on existential AI risks diverts attention from immediate societal harms.

Main Methods:

• Three preregistered online survey experiments were conducted with 10,800 participants.
• Participants were exposed to news headlines framing AI risks as either existential, immediate societal impacts, or beneficial.
• Quantitative analysis assessed changes in participant concern levels for different AI risk types.

Main Results:

• Respondents expressed significantly higher concern for immediate AI risks compared to existential risks.
• Exposure to existential risk narratives amplified concerns about catastrophic AI threats.
• Existential risk framing did not decrease, and may have even reinforced, concerns regarding immediate AI harms.

Conclusions:

• The study provides empirical evidence that existential AI risk narratives do not distract from, but rather coexist with, concerns about immediate AI harms.
• Findings suggest that public perception of AI risks is nuanced, with a strong emphasis on present-day societal impacts.
• The results inform ongoing scientific and policy discussions on managing AI's societal implications effectively.