Scientific Reasoning Instruction
How AI-assisted teaching changes hypothesis generation, critique, calibration, transfer, and scientific judgment.
RESEARCH & METHODS
Building the Methods for AI-Assisted Science
Santaros Labs studies how AI systems affect the learning and teaching of scientific practice while preserving rigor, traceability, privacy, learner agency, and human accountability.
OUR MISSION
Expand Scientific Capability Without Weakening Scientific Standards
We test whether AI-assisted instruction changes scientific reasoning, literature review, coding, analysis, and research-team learning. Evaluation separates short-term productivity from accuracy, evidential support, calibration, independent transfer, and error detection.
Evidence and plans stay distinct
Completed source syntheses report their scope and limits. Protocols in development and preregistration are never presented as completed intervention evidence.
Four Research Pillars
Reproducible Analysis Training
How learners can preserve the chain from source data and literature to transformations, code, results, and claims.
Evidence-Synthesis Pedagogy
How learners use AI to evaluate citations, disagreement, uncertainty, missing data, and limits of the evidence.
Mentored Research Teams
How shared AI workspaces affect feedback, learning, handoffs, review, expertise, and accountability in scientific groups.
PILLAR 01
Scientific Reasoning Instruction
Measure what learners can explain, revise, and transfer independently.
We study AI-assisted teaching across question formation, hypothesis critique, experimental design, simulation, code generation, statistical interpretation, and scientific argument.
Comparisons separate temporary task support from learning. Better instruction should improve testability, evidence alignment, calibration, error detection, and performance on new problems without the original scaffold.
PILLAR 02
Reproducible Analysis Training
Teach learners to make the path from raw evidence to claim inspectable.
AI-assisted coursework and research can involve transformations that disappear from the final submission: retrieved sources, data cleaning, generated code, model choices, rejected analyses, and rewritten interpretations.
We test teaching methods that make provenance part of the analysis itself, then ask whether an independent learner can reproduce the work.
PILLAR 03
Evidence-Synthesis Pedagogy
Teach learners to test whether a claim is supported, not merely well written.
Scientific synthesis can become unreliable when fluent summaries flatten disagreement, attribute unsupported claims to sources, or hide uncertainty. We design instruction around contradiction detection, source discrimination, evidence certainty, and calibrated refusal.
Evaluation combines citation-level checks, expert scoring, confidence calibration, delayed transfer, and structured error taxonomies. We measure the learner's reasoning separately from the fluency of the AI-assisted text.
PILLAR 04
Mentored Research Teams
Study the lab as both a research system and a learning environment.
Research is collaborative and instructional. We examine how AI changes feedback, review burden, mentoring, handoffs, authorship, responsibility, and opportunities to build expertise across research teams.
The aim is to design shared practices that make reasoning visible, preserve productive disagreement, and help mentors support learning without turning supervision into surveillance.
We test whether transparent assistance supports early-career learning and whether opaque delegation reduces opportunities to build research judgment.
How We Work
Each study pairs scientific ambition with explicit controls for validity, ethics, security, and reproducibility.
Preregister
Specify questions, measures, exclusions, and analysis decisions before outcomes are known whenever the design permits.
Work with domain experts
Evaluation criteria come from the science being studied, not from generic impressions of output quality.
Protect research data
Use data minimization, access controls, and study-specific security practices appropriate to the sensitivity of the work.
Preserve provenance
Record sources, transformations, code, model context, and material human decisions across the workflow.
Replicate
Separate exploratory discovery from confirmatory analysis and include independent reproduction where feasible.
Report limitations
Publish boundaries, null results, and failure modes so future teams know where evidence does and does not apply.
RESEARCH STANDARDS
Methods Need an Audit Trail
Our standards distinguish exploratory from confirmatory work, define outcome status, document model and tool configurations, and require limitations and corrections to remain visible.
Read the research standardsRESEARCH COLLABORATION
Begin With the Question, Then Specify the Test
Bring a research question, an evaluation problem, or a reproducibility challenge. We will start with study design, not marketing claims.
Proposed collaborations proceed only after scope, governance, ethics, and data terms are clear.