AlphaScience
Claim calibration for AI-assisted scientific exploration, centered on evidence ledgers, non-claims, and manuscript-facing records.
AI-assisted science · scientific agents · claim calibration
AlphaScience Lab
We study how AI-assisted scientific exploration can be routed into evidence-ledger states before it becomes manuscript-facing claims. The aim is to prevent premature promotion: weak, ordinary, incomplete, or negative outputs being converted into claims stronger than the evidence supports.
Mission
Claim calibration before narrative.
AlphaScience Lab is a methodological research effort led by Hongmin Li, a computational biologist and machine-learning researcher working on AI-assisted science, scientific agents, and claim calibration.
The current work asks how AI-generated hypotheses, code, figures, analyses, and manuscript-like arguments can be routed through auditable evidence states before they are allowed to become public scientific claims.
Featured project
AlphaScience: Claim Calibration for AI-Assisted Scientific Exploration
AlphaScience studies a bottleneck that appears after AI systems generate bounded work products: what is each output responsibly allowed to claim?
The public version tells the method story without disclosing unpublished case work, active repositories, raw data, code, logs, transcripts, internal communications, or case-specific biological claims.
Read the project pageSelected methodological projects
A research program around evidence-led AI-assisted science.
The project list is intentionally ordered by methodological priority. Biology-facing work remains useful, but the public frame is claim discipline, evidence routing, and release-safe scientific accountability.
Autonomous Claim Calibration
Evidence-ledger discipline for scientific agents when no human workflow supervisor is continuously steering interpretation.
Scientific-map fidelity
Testing whether visual claims from UMAP, t-SNE, and related maps have high-dimensional support.
Cell-state representation
Studying bounded transfer across disease, cohort, platform, and virtual-cell evaluation settings.
ID3 and mRNA design
Computational biology and sequence-design work that informs how AI-assisted claims should be bounded and validated.
Public-data storylines
Release-safe field notes from public datasets, written as questions and boundaries rather than finished biological discoveries.
Principles
Public materials are release-safe by design.
- Personal methodological research AlphaScience is presented as a claim-calibration framework, not as a release of unpublished case work.
- Evidence before narrative Outputs are useful only when they can be checked, repeated, and bounded.
- Explicit release boundary Public pages do not disclose raw data, code, runtime logs, AI transcripts, internal communications, active repositories, or case-specific biological claims.
Stories
Follow the lab notebook as it becomes a field report.
Subscribe for AlphaScience stories: research notes, case histories, failed turns, methods, figures, and the practical craft of expert-guided AI-assisted discovery.
Read the current field notes in the AlphaScience Lab blog.
Contact
alphascience-lab.com
AlphaScience Lab is preparing its first public research outputs. For collaboration or correspondence, contact the lab by email.
info@alphascience-lab.com