Biological AI models are trained on biological data (sequence, structure, images, etc.) and are increasing in both size and performance. To safely use these models to cure disease and unlock novel biological applications, novel risk mitigation approaches are needed.
About the project
This project explores risk assessment, technical and policy risk mitigations, and methods for the safe integration of bio models with frontier AI systems including LLMs, agents, and autonomous robotics.
Relevant references: https://www.darioamodei.com/essay/machines-of-loving-grace https://www.science.org/doi/abs/10.1126/science.adq1977 https://www.nature.com/articles/d41586-024-03815-2 https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1012975
Theory of change
One of the most important transformations AI could bring about is improved health. To achieve this, organizations are generating vast quantities of AI-ready biological data and creating bio foundation models to understand and engineer life. Among the first biological constructs we will successfully engineer will be pathogens, including viruses, due to their small and easily engineered genomes. Approaches to reducing both the accident and misuse risks of this are needed.
Your role
Mentees will receive direct mentorship regarding the most important, neglected and tractable problems within securing AI for biology. This can be challenging to determine without mentors who are active in the field. Mentees will then have the autonomy to seek out novel ways of solving these specific problems. In short: specific mentor guidance re: problems that need to be solved + relative mentee autonomy on how to solve them.
Prerequisites
- Knowledge of biology concepts (e.g. structure and function of RNA)
- Knowledge of AI concepts (e.g. RNN vs. transformer)
- Knowledge of biological AI concepts (e.g. genomic language model)
- Knowledge of AI safety methods (e.g. RLHF)
- Knowledge of dual-use biology risks (e.g. immune evasion)
Location preference
Eastern and Pacific time zone availability
Application question(s)
- How will biological AI models be used in the next 5-10 years?
- What are the advantages and drawbacks to integrating AI agents with biological AI models?
- What policies currently exist globally for the oversight of biological AI models?
About the mentor
Dr. Pannu is a Senior Scholar at the Johns Hopkins Center for Health Security and an Assistant Professor in the Department of Environmental Health and Engineering at the Johns Hopkins Bloomberg School of Public Health. At the Center, Dr. Pannu’s primary areas of research include global health security and biosecurity, pandemic prevention and preparedness, and emerging technology security and governance. Dr. Pannu previously worked on AI-enabled diagnostics at Google and has served as a subject-matter expert for the Chan-Zuckerberg Initiative, the National Academies of Sciences, and the Bipartisan Commission on Biodefense, among others. She serves on the board of Blueprint Biosecurity, a philanthropically funded nonprofit dedicated to achieving breakthroughs in humanity's ability to prevent pandemics.
Dr. Pannu regularly briefs government officials and policymakers on emerging technologies with security implications, including artificial intelligence and synthetic biology. She applies her medical training and high-tech industry experience towards timely, scientifically rigorous research to help inform policymakers, public health practitioners and the general public regarding actionable biosecurity policy approaches.