Marginal Revolution's Tyler Cowen Calls for Scientific Rigor in AI Security Risk Discourse
Key Takeaways
- •Tyler Cowen argues that alarmist claims following the OAI/HuggingFace incident lack sufficient scientific substantiation.
- •Cowen requests that concerned parties provide a concrete methodology for quantifying AI-attributable cybersecurity costs within a two-to-three-year timeframe.
- •He asks for specific numerical cost estimates that can be tracked and evaluated against real-world outcomes over time.
- •Cowen calls for disclosure of any shorted stocks or assets, noting that his own expected costs are low enough that he would not short the market.
- •The post advocates grounding AI security policy debates in falsifiable evidence rather than what Cowen describes as personal anxiety management.
Following the recent OAI/HuggingFace incident and related developments, a wave of dire predictions and serious concerns has emerged across commentary channels. However, the economics blog Marginal Revolution, co-authored by George Mason University economist Tyler Cowen, argues that the majority of these alarmist claims remain insufficiently substantiated.
In a post titled "Act like it is science," Cowen issues a straightforward challenge to those raising alarms: adopt the methods of science to make a convincing case. The post reflects Cowen's longstanding methodological stance—widely recognized in academic economics and policy circles—that policy arguments should be grounded in quantifiable evidence and falsifiable predictions rather than qualitative appeals to worst-case scenarios. Specifically, the post requests the following from anyone wishing to persuade skeptics:
-
A methodology for estimating future cybersecurity costs. Cowen asks for an outline of how, within two to three years, one might quantify the additional cybersecurity costs attributable to AI-related security breaches. He acknowledges that this figure is unlikely to be zero but insists on a concrete estimation method. As reference, the post links to a study of past cybersecurity costs conducted by leading economists.
-
A current numerical estimate of those potential costs, to be evaluated against actual outcomes over time. Cowen notes that respondents can provide estimates across multiple regulatory and safety scenarios, framing this as a straightforward way to demonstrate analytical credibility, even if confirmation comes with a delay. The post clarifies that such estimates need not serve as a substitute for one's preferred safety measures, but urges specificity in predictions.
-
A disclosure of any stocks or other assets currently shorted. If one's cost estimate from the second point is sufficiently low—as Cowen's own would be—the answer could legitimately be zero. He states that while costs are expected, they are unlikely to be so severe that society cannot navigate them while still anticipating positive stock returns. The disclosure request echoes a broader academic practice of asking analysts to put their convictions on record, making it possible to track whether the most dire forecasts hold up against observed outcomes over time.
The post concludes that adopting these practices would elevate the conversation to a more scientific level and move it away from what Cowen characterizes as "an extended exercise in personal anxiety management." If the issue is truly as critical as proponents claim, he argues, the effort required to produce such rigorous analysis is surely warranted.
The article was published on Marginal Revolution.
