Recent discussions among AI industry leaders have highlighted growing concerns regarding the potential dangers of artificial intelligence, particularly in its intersection with biotechnology. Dario Amodei, CEO of Anthropic, emphasized the risks associated with AI, suggesting a need for a more measured pace of development. This sentiment was echoed by OpenAI CEO Sam Altman, who concurred that the rapid advancement of AI technology warrants caution. These discussions gained traction following an announcement by AI researcher Jacob Coxon, who resigned from Anthropic, claiming that neither his company nor OpenAI was adequately addressing the existential risks posed by AI. Coxon’s alarming statement that AI could lead to catastrophic outcomes within the next decade was supported by fellow Anthropic employee Evan Hubinger, who voiced similar concerns on social media.
One of the most pressing fears is the possibility of AI facilitating the creation of bioweapons. The capabilities of AI could theoretically enable the design of highly pathogenic viruses tailored to specific genetic profiles, or even toxins that could contaminate water supplies undetected. A notable study from Collaborations Pharmaceuticals in 2022 demonstrated how an AI-driven molecule generator could produce thousands of chemical compounds in mere hours, some of which exhibited toxicity levels surpassing known nerve agents. David Magnus, a professor at Stanford University with a long history in biomedical ethics, expressed alarm at these findings, stating that the risks associated with AI and biotechnology have escalated dramatically in recent years.
As AI technology becomes more advanced, its accessibility raises significant security concerns. Large language models can provide comprehensive instructions on conducting various scientific experiments, making it easier for individuals to engage in potentially hazardous research. Although there are safeguards, such as DNA screening by suppliers and rigorous peer-review processes for risky research, experts warn that these measures may not be foolproof. A recent report from Anthropic revealed attempts to misuse their models for enhancing the transmissibility of the chikungunya virus and creating more dangerous forms of bird flu. While some researchers argue that current AI capabilities may not suffice for developing bioweapons, others, like biosecurity researcher Dunja Sabra, advocate for proactive measures, urging countries to strengthen healthcare systems and prepare for potential threats.
Source: The specter of AI-enabled bioweapons is a wake-up call for biotech via MIT Technology Review