Research program
Artificial intelligence can expand discovery while also increasing the burden of interpretation, error detection, synthesis, and governance. My research examines how scientific institutions and technical infrastructure can preserve credibility when production scales faster than verification and accumulation.
Core thesis
A candidate claim becomes useful knowledge only after it is represented faithfully, independently checked, connected to alternatives, and incorporated into a current knowledge state. AI changes the relative cost and throughput of these functions, but it does not remove any of them.
01
Produces candidate questions, hypotheses, designs, analyses, findings, code, and artifacts.
Abundance risk: plausible output expands faster than warranted knowledge.
02
Tests representation fidelity, artifact integrity, measurement, identification, reproduction, transport, and rival explanations.
Current focus: making this layer accurate, reusable, and independently auditable.
03
Preserves, relates, versions, contests, updates, synthesizes, and retrieves knowledge through time.
Abundance risk: even verified findings fail to become a coherent, current state.
Four linked workstreams
Measure how AI changes lifecycle costs, bottlenecks, incentives, and coordination problems in scientific knowledge production.
Foundation: Knowledge Abundance ParadoxSpecify alternative scientific objects and transitions—including explicit DGP hypotheses, finding-level records, provenance, and bounded updates.
Foundation: DGP Multiverse ScienceBuild and test protocols, benchmarks, and human–AI workflows that can detect consequential scientific errors accurately and efficiently.
Current priority: structured peer-review experimentConnect verified findings into updateable systems for synthesis, research prioritization, and decision support.
Advancement condition: empirical benefit must exceed representation and governance costsResearch discipline
The program does not assume that structured or AI-native systems are better. Any successor to paper-centered coordination should advance only when it improves quality-adjusted verification and accumulation, preserves context and pluralism, exposes dependence and provenance, and remains feasible for the people who bear its costs.
Read the foundations