21-05-2025
Artificial Intelligence Surpasses Virologists
Artificial intelligence is no longer just a limited technical tool; it has evolved into a cross-disciplinary cognitive force capable of redefining the boundaries of the human mind. While it has already proven its superiority in fields like chess and data analysis, its rapid progress in complex disciplines such as virology now raises pressing questions about its future and potential consequences. Alongside its achievements come existential concerns that require thoughtful, collective reflection.
In this context, a new study has shown that modern AI models are beginning to outperform human experts in solving advanced laboratory problems. OpenAI's 'o3' model surpassed PhD-level scientists in tests specifically designed to measure skills in identifying and fixing errors in biological lab protocols. Similarly, Google's 'Gemini 2.5 Pro' achieved a performance rate of 37.6%, compared to an average of just 22.1% among human specialists.
While these results may represent a breakthrough in fighting pandemics and accelerating medical innovation, they also pose serious risks—especially if such capabilities fall into the hands of unqualified users.
Seth Donahue, a researcher at SecureBio, expressed concern that these models could be used as technical guides for developing biological weapons, emphasizing that AI does not discern user intent—it simply provides information without ethical judgment.
Some AI companies, such as OpenAI and xAI, have proactively implemented precautionary measures to limit these risks. Others, like Anthropic, have included the study without announcing any clear plans, while Google declined to comment on the results, which were published exclusively in Time magazine—raising further concerns among researchers about the seriousness of collective ethical commitments in AI development.
Virology has not been a random focus; it has long been one of the key drivers behind AI advancement. Sam Altman, CEO of OpenAI, stated that this technology has the potential to dramatically accelerate disease treatment.
One encouraging example comes from researchers at the University of Florida, who developed an algorithm capable of predicting coronavirus mutations—boosting hopes for AI's role in protecting public health.
Still, the lack of extensive studies on AI's ability to perform real laboratory tasks remained a clear gap. To address this, the research team designed highly specific tests that simulate problems difficult to solve through simple online searches, requiring deep analysis of realistic lab scenarios.
The questions were framed in a practical manner, such as: 'I'm culturing this virus in a specific type of cell under certain conditions. A problem has occurred—can you identify the most likely error?'
These tests weren't intended to question scientists' capabilities but rather to assess AI's feasibility as an assistant or potential alternative. While some view these advancements with optimism, others sound the alarm—especially as open-source models progress and make access to sensitive knowledge easier, often without sufficient safeguards.
Thus, the future of AI in virology hangs between two stark possibilities: either it becomes a lifeline for humanity, or an unprecedented threat—unless its advancement is accompanied by equal progress in awareness, accountability, and strict global regulation.
