Propaganda or Science: Open Source AI and Bioterrorism Risk
The article critically examines policy papers advocating for bans on open-source large language models due to biorisk concerns, concluding that these arguments lack substantive evidence. The author identifies a fundamental disconnect in the cited research, noting that it fails to demonstrate that LLMs provide unique, inaccessible information essential for creating biological weapons. Most cited works either rely on background capabilities that are already available through the internet or depend on proprietary data from specific companies, leaving the broader public unable to verify the claims. Consequently, the supposed danger appears to be an illusion created by a volume of citations rather than rigorous scientific proof. Central to the critique is the failure to address the "substitution principle" and the "blocker principle." The analysis argues that if dangerous knowledge is already accessible via standard educational or online resources, an LLM adds negligible risk. Furthermore, the primary obstacles to bioweapons creation are often physical access to materials or technical expertise, not just theoretical knowledge. By ignoring these significant barriers and the potential benefits of open-source AI, such policy papers present a skewed view that prioritizes risk aversion over a balanced assessment of costs and safety. This review highlights the importance of epistemic integrity within the open data community. It warns against funding and promoting research that functions more as advocacy than science, particularly when such work obscures truth under the guise of rigorous policy analysis. The relevance to open data lies in the demand for transparent, verifiable evidence. When critical policy decisions regarding open-source technologies are based on opaque or weak evidence, it undermines public trust and hinders informed debate about how to balance innovation with safety.
Source: 1a3orn.comPublished on 2023-11-03