Moonshot Thesis
The Future of Science is Executable
Humanity stores science in documents. Documents are readable but not executable. This is the deepest misunderstanding in how we do research.
What Humanity Has Misunderstood
For centuries, we have treated scientific papers as the final product of research. We write them, read them, cite them, and store them in databases. But papers are descriptions of knowledge, not the knowledge itself.
A paper might say "cool roofs reduce urban temperatures." But what happens when you combine cool roofs with tree canopy? What happens if the AC feedback loop is stronger than we think? The paper cannot answer these questions. Only a causal model can.
We have built extraordinary tools for finding papers (Google Scholar), reading papers (PubMed), and even summarizing papers (AI chatbots). But we have built almost nothing for making science executable — for turning the relationships described in papers into structures we can simulate, test, and reason about.
This is not a search problem. This is not a summarization problem. This is a representation problem. Science needs a new substrate — not documents, but causal worlds.
Why Current Tooling Is Insufficient
Search engines find papers
But papers describe relationships, they don't represent them. You still have to read, interpret, and mentally construct the causal model.
AI chatbots answer questions
But they answer from training data, not from explicit causal structure. They can't simulate interventions or detect contradictions in real-time.
Literature reviews synthesize
But they're narrative, not executable. You can't run a counterfactual on a literature review. You can't compute information gain.
Reference managers organize
But they organize documents, not knowledge. The causal relationships between papers remain invisible and disconnected.
The First-Principles Insight
Science is not a collection of facts. It is a web of causal relationships between variables.
If you can represent that web explicitly — with nodes, edges, signs, strengths, uncertainties, and evidence sources — you can do something no paper can do: simulate what happens when you intervene.
This is the core insight of CausalForge: the transition from descriptive science to executable science requires a new representation layer — a causal world model that sits between the literature and the experiment.
With this layer, you can ask: "What would happen if...?" and get an answer grounded in the actual literature. You can ask "Where do scientists disagree?" and get a map of contradictions. You can ask "What should we measure next?" and get a ranked list of experiments sorted by expected information gain.
System Architecture
Ingestion Layer
PDF parsing, text extraction, structured note ingestion
Extraction Engine
NVIDIA NIM-powered scientific claim extraction with schema validation
Causal Graph Builder
Variable normalization, edge construction, evidence weighting
Analysis Engine
Contradiction detection, uncertainty scoring, conflict mapping
Simulation Core
Counterfactual propagation, confidence propagation, sensitivity analysis
Discovery Engine
Information-gain ranking, experiment recommendation, frontier identification
Technology Stack
Long-term Vision
Imagine a world where every research paper is automatically ingested into a causal world model. Where contradictions are detected within hours of publication, not years later during a literature review. Where the next best experiment is computed, not guessed.
CausalForge is a first prototype. The full vision is a Scientific Operating System — a layer of executable science that sits on top of humanity's accumulated knowledge and helps us reason about it, simulate it, and discover from it at the speed of AI.
This would transform how we approach climate change, drug discovery, materials science, and every field where the gap between understanding and action is limited by our ability to reason about complex causal systems.
Faster science. Less wasted experimentation. Better intervention discovery. This is what executable science enables.
CausalForge — Moonshot Hackathon 2026 · Executable Science for the Next Decade