Harvard and MIT Experiment Raises Concerns over AI’s Role in Bioweapon Development
01 October 2023
In a recent experiment, scientists at Harvard University and the Massachusetts Institute of Technology (MIT) explored the capabilities of artificial intelligence (AI) in orchestrating bioweapon development. The study, conducted last year, involved a group of students with no specialised training in the life sciences. These participants used AI tools, including OpenAI's ChatGPT-4, to draft plans for initiating a pandemic. Remarkably, within an hour, the students learned how to procure and synthesise deadly pathogens such as smallpox, circumventing current biosecurity measures.
Although AI technology is not yet capable of producing a national security crisis independently, the experiment highlighted the increasing potential for misapplication as biotechnology advances. This experiment has prompted policymakers to scrutinise and regulate the burgeoning AI industry more strictly.
The balance between regulation and innovation is delicate. While addressing catastrophic risks is crucial, overregulation could stifle the beneficial applications of AI technology. For instance, AI has shown promise in numerous fields; AlphaFold and RFdiffusion, for example, have made significant progress in designing new proteins, which can be utilised in medical treatments.
The dual-use nature of AI becomes evident when considering malicious applications. A potential bioweapon identified with AI's assistance could enable a bad-faith actor to order a custom strand of DNA from a commercial provider. This synthetic DNA, manufactured in a lab and mailed to the actor, could code for a toxin or enhance a pathogen's virulence. Many scientists order synthesised DNA for legitimate purposes, such as cancer or infectious disease research, but not all providers rigorously screen orders or verify clients, highlighting a critical loophole in biosecurity.
While closing these loopholes is necessary, it may not entirely mitigate the risk. Strengthening defences through investment in AI-enabled early-detection systems appears prudent. Currently, the Centers for Disease Control and Prevention (CDC) employs a Traveller-based Genomic Surveillance program in cooperation with airports across the nation. This initiative collects and analyses wastewater and nasal swab samples to detect pathogens entering the borders. Additionally, there are systems in place to monitor specific pathogens within cities and communities. However, existing detection methods may prove inadequate against novel agents designed with AI.
The U.S. intelligence community has already begun investing in AI-driven measures to counter next-generation threats. The Intelligence Advanced Research Projects Activity (IARPA) has developed the FELIX program in partnership with private biotech firms. This initiative has produced AI technology capable of distinguishing genetically engineered threats from natural ones and identifying specific alterations.
The untapped potential of AI in detecting and defending against biological threats remains considerable. In the event of a new infectious disease outbreak, AI systems could ascertain how and when a pathogen has mutated, facilitating the rapid development of customised vaccines and treatments. Furthermore, AI could forecast the pathogen’s spread, enhancing containment and mitigation efforts.
For these technologies to fulfil their protective role, global leaders, including those in Washington, must take concerted steps to fortify AI defences. The aim should not be to eliminate AI but rather to leverage 'good AI' to counter 'bad AI', ensuring that AI serves as a tool to enhance, not threaten, global safety and security.
-ENDS-
Source: Noah Wire Services