What AI can and cannot do in biosecurity

As AI continues to gain momentum in recent weeks, opinions are divided over its potential impact on humanity. An AI researcher once estimated a 10% chance of humanity being wiped out by 2030; however, the veracity of such claims remains uncertain. This debate has shed light on a critical public health question: How does biosecurity risk intersect with superintelligence?

To delve deeper into this topic, I reached out to my friend Dr. Claire Dillavou, an experienced applied epidemiologist and AI enthusiast. Together, we explored the feasibility, challenges, and solutions necessary from a health perspective. We shared these insights with the YLE community to foster a broader discussion on the matter.

Biosecurity encompasses a broad spectrum of threats, ranging from natural occurrences like viruses and antibiotic-resistant bugs to engineered dangers such as toxin modifications. Despite the varying levels of probability and feasibility, the physical aspects of creating these threats will persist, necessitating a hybrid approach combining AI and traditional techniques.

Some AI-driven threats have surfaced, with Anthropic recently exposing five case studies demonstrating its potential misuse for biological weapons development. These studies explored techniques like enhancing virus transmissibility and evading immune responses. While Anthropic intervened to safeguard its models, questions still linger about the extent of AI's impact on biosecurity.

Conversely, AI also plays a crucial role in mitigating biosecurity risks. It can accelerate the understanding of novel viruses, aid in vaccine development, and enable early disease diagnosis. However, the extent to which AI outweighs the risks in biosecurity remains unclear. To address this, several initiatives are underway to bridge the gap between concern and knowledge, with a particular focus on protecting sensitive data.