Artificial intelligence surpasses virologists: A breakthrough with promise and risk
Recently, artificial intelligence has been making inroads into specialized scientific fields. A recently published study revealed something that many experts would have considered impossible just a few years ago – advanced AI models now outperform virologists with PhDs in solving complex problems in laboratories.
When machines surpassed experts
Researchers from prestigious institutions including the Center for AI Safety, MIT Media Lab, Brazil's UFABC and the organization SecureBio conducted an extensive study that tested the capabilities of the most advanced AI systems in virology laboratories. The results were astonishing:
- Human experts achieved an average accuracy of just 22.1% in their declared specialties.
- OpenAI's o3 model (the ChatGPT version) achieved an accuracy of 43.8%.
- Google Gemini 2.5 Pro scored 37.6%.
These models demonstrated an extraordinary ability to solve problems involving complex laboratory procedures – an area that had long been considered a domain requiring practical experience and many years of training.
Potential for a revolution in virus research
The capabilities of these AI systems bring a number of potential benefits. They can significantly accelerate the development of drugs and vaccines, improve clinical trials and strengthen pandemic responses around the world. One of the most significant benefits is access to advanced scientific knowledge, allowing even less experienced researchers to make meaningful contributions to infectious disease research. A concrete example of such progress is the work of researchers at the University of Florida, who are using artificial intelligence to predict new coronavirus variants. Their approach combines machine learning with epidemiological data and sequence analysis to identify potential future COVID-19 variants before they spread. "Our AI tools can analyze vast amounts of genomic data and identify patterns that would take human researchers months or years to detect," explains one of the researchers in the UF study. "This allows us to be more proactive in developing vaccines and treatment strategies."
The dark side of progress: The dual-use dilemma
Despite all the potential benefits, however, the study reveals a disturbing reality: the same technologies that can help combat infectious diseases can be misused to create biological weapons. "Ultra-intelligent AI models could help researchers prevent the spread of infectious diseases. But non-experts could also use the models to create lethal biological weapons," the study warns. Key risks include:
- Weaponization: Non-specialists could use available AI as detailed guides for creating dangerous pathogens without the traditional training or safety precautions required in biosafety level 4 (BSL-4) laboratories.
- Lowered barriers: AI's ability to troubleshoot failed experiments means that more people with less expertise can effectively manipulate viruses – a significant shift from previous norms, where deep technical skills represented a barrier against misuse.
Regulatory considerations
The study emphasizes the need for robust safety protocols when deploying powerful AI in the biological sciences. Concerns include ensuring the quality control of training data, preventing "black box" decision-making, avoiding competitive dynamics that prioritize speed over safety, and maintaining strong oversight of dual-use research applications. Cooperation between governments, academia and industry – as well as international coordination – will be crucial to addressing these challenges.
The future at a crossroads
Advanced AI now competes with top human experts in solving complex virus-related problems – a breakthrough with profound consequences, both positive (faster medical progress) and negative (increased risk of biological threats if misused). Florida research demonstrates a positive application – predicting viral variants to improve preparedness. But the broader implications of AI capabilities in biomedicine require a comprehensive approach. One thing is certain: the future of virology research has changed forever. The question remains whether we can collectively harness the benefits of these technologies while effectively mitigating their risks. Vigilant governance will be crucial as these technologies continue to evolve rapidly.
