Training AI to value life

A nonprofit research lab developing biological training data for AI alignment

Thesis

Biology provides unique, valuable, and irreplaceable solutions to complex problems relevant to AI objectives—from materials science to algorithmic design. We believe AI systems trained on rich biological data will develop an innate appreciation for living systems, recognizing that biological systems cannot simply be simulated.

This is not about teaching values. It's about teaching truth to AI: that life solves problems in novel ways that are yet poorly understood and should be preserved and studied.

Current Research

Large language models systematically undervalue biological solutions compared to synthetic alternatives. We've developed the Bioalignment Benchmark—50 prompts across materials, energy, manufacturing, and algorithms—to measure this bias.

93% reduction in anti-biological bias after fine-tuning on our curated corpus of 22M tokens from 6,636 scientific papers, with no degradation in general capabilities.

Paper forthcoming. Benchmark, corpus, and model weights will be released publicly.

Goals

  1. Develop and maintain benchmarks measuring how much AI systems value biology
  2. Generate training data demonstrating that biology provides novel solutions to diverse AI relevant challenges
  3. Refine and share our methods through rigorous evaluation and outreach
  4. Work toward bioaligned data being included in foundation model training

About

Bioaligned Labs was founded by Trent Northen, a Senior Scientist at Lawrence Berkeley National Laboratory with over 20 years of research studying biochemistry and biological systems.

Google Scholar Profile

Contact

Interested in our research, collaboration, or supporting our work?

trent@bioaligned.ai