AI Meets Science

SoTS Research Lab

Available: Princeton In-Person | Online

Mentor: Sergey Samsonau, PhD

Seize the Opportunity

Science is the foundation of modern society - medicine, sociology, technology, ecology all depend on it. AI reshapes how scientific research gets done. What implications will this bring - for science and for everyone? Build expertise at the intersection of AI and science - the combination every research lab, tech company, and graduate program wants. Stand out in college applications. Help find answers - while the field is new and your work can shape it.

The Opportunity

A newly formed field - "AI for Science" - has seen explosive growth. The 2024 Nobel Prize in Chemistry went to the creators of AlphaFold. Google launched an "AI co-scientist." Anthropic, OpenAI, and others race to build AI research assistants. Billions flow into AI-powered drug discovery, materials science, and scientific automation.

Tools now claim to read thousands of papers instantly, generate hypotheses, write analysis code, predict protein structures, and design experiments. Every major research institution adopts them.

Understanding lags behind adoption. Where does AI genuinely accelerate discovery? Where does it introduce new errors? Where does it help some fields but fail others? These questions shape the future of research - and the answers remain unclear.

This lab puts you at the center of that conversation.

What You Get

A rare skill combination. Everyone uses AI. Few people understand how it performs in scientific contexts. You develop expertise in both AI capabilities and scientific methodology - and learn where they intersect. Employers, grad programs, and research labs need people with this judgment.

Skills that apply everywhere. How to design rigorous comparisons. How to analyze results. How to communicate findings clearly. How to think critically about big claims. These matter anywhere - not just in science or AI.

Something real for college applications. Not "I used ChatGPT for homework." Research that investigates where AI accelerates scientific discovery and where it falls short. That's a different story to tell.

Early entry to a new field. AI for Science barely existed five years ago. Rigorous understanding of these tools is even newer. You're not behind - you're early.

What Students Do

You pick a scientific area you care about. You investigate how AI tools perform on real tasks in that area. You document what you find. You aim to publish.

Examples of published studies (the kind of work you'd do):

Literature search: Elicit vs. traditional database search - found 39.5% of relevant papers (vs. 94.5% traditional), but higher precision
Scientific code: ChatGPT tested across 9 languages on scientific computation - Fortran and Rust often failed to compile
Statistical analysis: ChatGPT-4 vs. SPSS and R - "highly consistent" for basic analyses, discrepancies in complex tests
Protein structure: AlphaFold vs. experimental crystal structures - often accurate, but doesn't account for ligands or environmental conditions
Image analysis: Deep learning for microscopy - high accuracy on training data, but real-world performance varies
Hypothesis generation: AI-generated hypotheses evaluated by experts - only 6% rated "highly novel"

Different AI claims. Different methods. All need more independent investigation.

Categories you could explore: literature search, scientific code, data analysis, image recognition, hypothesis generation, experimental design, scientific writing, data visualization, pattern detection, anomaly detection, prediction accuracy, reproducibility, tool evolution over time.

Tools you might evaluate:

Literature search: Elicit, Consensus, Scite, PaperQA
Scientific code: GitHub Copilot, ChatGPT, Claude for scientific computation
Data analysis: ChatGPT Advanced Data Analysis, Code Interpreter tools
Structure prediction: AlphaFold, ESMFold, RoseTTAFold
AI scientist platforms: Google AI Co-Scientist, FutureHouse

Pick a tool. Pick a task. Study it carefully. Document what you find.

Note: Some AI tools require paid subscriptions. Costs depend on which tools you choose to evaluate and are not covered by SoTS Labs.

Why This Matters

AI companies make big claims. "Accelerates discovery." "Transforms research." "Like a thousand research assistants."

Science challenges claims. Finds what's true and what isn't.

That's what this lab does - use the scientific method to evaluate tools that claim to do science. Your findings help scientists choose tools, help companies improve them, help everyone see what's real.