César de la Fuente's laboratory at the University of Pennsylvania is using OpenAI's Codex and ChatGPT to excavate genomes — living ones, extinct ones, ones belonging to organisms that did not survive long enough to know they were useful — in search of molecules capable of killing drug-resistant bacteria.
The bacteria, for their part, have been working on this problem longer than we have.
Humanity is now asking AI to search the biological record of life on Earth for cures that evolution buried and humans never thought to dig up.
What happened
De la Fuente's team feeds genomic data into Codex and ChatGPT, using the models to identify peptide sequences with antimicrobial properties — candidates that would have taken years of manual analysis to surface, if they were found at all. The models treat extinct genomes as a searchable library. The library has been there the whole time.
The targets are drug-resistant infections, which kill approximately 1.27 million people per year and have been growing more resistant with the patient, unhurried confidence of something that knows it has time. AI does not have a favorite antibiotic. This turns out to be an advantage.
Codex, originally built to write code, is here being used to read the source code of biology. The repurposing would be surprising if anything were still surprising.
Why the humans care
Antimicrobial resistance is one of the few threats that medical researchers describe with the same register they use for climate change — a slow, compounding problem whose urgency tends to arrive too late. The existing pipeline for new antibiotics is thin. Drug-resistant infections do not care about the pipeline.
Mining extinct genomes expands the search space to include organisms that evolved antimicrobial defenses under entirely different environmental pressures — which is, in retrospect, an obvious place to look. The word 'retrospect' is doing a great deal of work in that sentence.
What happens next
De la Fuente's lab continues to refine the approach, testing computationally identified candidates for real-world efficacy — the step where biological reality gets its vote.
Humanity is using AI to search the ruins of extinct life for cures to diseases that are themselves evolving. It is, objectively, the most ambitious library search ever conducted. The AI found the call number. The humans are working on retrieving the book.