[Paper Review] Artificial intelligence and biological misuse: Differentiating risks of language models and biological design tools

Jonas B. Sandbrink|arXiv (Cornell University)|Jun 24, 2023

Law, AI, and Intellectual Property22 citations

TL;DR

The paper distinguishes risks from two AI tool classes—large language models (LLMs) and biological design tools (BDTs)—and discusses how their convergence could elevate biosecurity risks, with proposed interventions.

ABSTRACT

As advancements in artificial intelligence (AI) propel progress in the life sciences, they may also enable the weaponisation and misuse of biological agents. This article differentiates two classes of AI tools that could pose such biosecurity risks: large language models (LLMs) and biological design tools (BDTs). LLMs, such as GPT-4 and its successors, might provide dual-use information and thus remove some barriers encountered by historical biological weapons efforts. As LLMs are turned into multi-modal lab assistants and autonomous science tools, this will increase their ability to support non-experts in performing laboratory work. Thus, LLMs may in particular lower barriers to biological misuse. In contrast, BDTs will expand the capabilities of sophisticated actors. Concretely, BDTs may enable the creation of pandemic pathogens substantially worse than anything seen to date and could enable forms of more predictable and targeted biological weapons. In combination, the convergence of LLMs and BDTs could raise the ceiling of harm from biological agents and could make them broadly accessible. A range of interventions would help to manage risks. Independent pre-release evaluations could help understand the capabilities of models and the effectiveness of safeguards. Options for differentiated access to such tools should be carefully weighed with the benefits of openly releasing systems. Lastly, essential for mitigating risks will be universal and enhanced screening of gene synthesis products.

Motivation & Objective

• Differentiate the biosecurity risks posed by LLMs versus biological design tools (BDTs).
• Explain how LLMs could lower barriers to misuse when used as lab assistants or autonomous science tools.
• Describe how BDTs could enable creation of more dangerous pathogens and targeted weapons.
• Suggest interventions to manage risks, including pre-release evaluations and screening of gene synthesis products.

Proposed method

• Compare risk profiles of LLMs and BDTs based on capabilities and potential misuse scenarios.
• Argue for independent pre-release evaluations to assess model capabilities and safeguards effectiveness.
• Discuss policy options including differentiated access to tools and openness considerations.
• Advocate universal and enhanced screening of gene synthesis products as a mitigation.
• Analyze potential convergence effects when LLMs and BDTs are used together.

Experimental results

Research questions

• RQ1 What are the distinct biosecurity risks associated with large language models compared to biological design tools?
• RQ2 How could convergence of LLMs and BDTs increase the potential harm from biological agents?
• RQ3 What interventions can effectively mitigate risks without unduly hindering beneficial AI progress?
• RQ4 What are suitable governance options for access to powerful AI tools used in biology, including pre-release evaluations and screening procedures?

Key findings

• LLMs may provide dual-use information and lower barriers to biological misuse when turned into lab assistants or autonomous science tools.
• BDTs could enable creation of pandemic pathogens far worse than those seen to date and support more predictable, targeted biological weapons.
• Convergence of LLMs and BDTs could raise the ceiling of harm and broaden accessibility of dangerous capabilities.
• Independent pre-release evaluations can help assess model capabilities and safeguards; differentiated access policies should weigh benefits of open releases.
• Enhanced screening of gene synthesis products is essential for mitigating risks.