Machine
Foresight
Run the world forward
before you
decide.
Foresight Machines builds a self-learning foresight engine for markets, infrastructure, and institutions. Run simulations to predict consequences, find your next best action, and learn from real-world outcomes.
State T₀ Futures T₁ → Tₙ
Manifesto
Every decision contains a model of the future.
An investment memo, a credit file, a board strategy, an operating plan — each contains a model of a world that does not exist yet.
Today, that model is dispersed. It lives across spreadsheets, slide decks, analyst notes, market prices, policy assumptions, and individual judgment. It is revised unevenly, separated from the decision it informs, and rarely measured after the fact.
The conditions around those decisions are changing faster than the institutions built to make them. Capital moves across systems continuously. Supply, energy, information, and policy travel through tighter networks. Rules fragment across jurisdictions. A change in one system can alter the conditions of another before the next planning cycle begins.
Software learned to retrieve, summarize, and answer. It did not learn to maintain a state of the world around a consequential decision — and show how that state may change when an institution acts.
We build foresight machines.
A foresight machine maintains an explicit state of what is known now. It advances that state through evidence, time, and possible interventions. It produces distributions, not declarations. It shows which assumptions matter, which conditions break the decision, and which actions stay sound across futures.
It then commits its claims before the outcome is known.
When the outcome is known, the machine is scored. The record remains. The next decision starts with a better model of the world.
The hardest decisions are not unknowable. They are unrun.
The Machine
Model the world around the decision.
A foresight machine connects real-world evidence to a model of the people, assets, rules, and dependencies around a decision. As the world changes, the model updates.
Introduce an action. Change the conditions. Compare how the consequences could unfold, where the plan could fail, and which alternatives remain viable.
Domains
Where foresight goes to work.
The Record
The machine learns.
The record holds.
Every run preserves the model state, evidence, assumptions, actions tested, and projected consequences in a timestamped cryptographic record.
As new evidence and outcomes arrive, the record grows. The original remains verifiable: what the engine knew, what it considered, and how its judgment changed.
Research
Better models of a changing world.
Systems that learn from time.
Our research asks how machines can represent a complex system, reason through interventions, and improve from real-world outcomes.
We combine language models, time-series forecasting, simulation, and causal inference to build models whose predictions can be tested and whose assumptions can be examined.
Project 1654 brings researchers and builders into this work. Each project defines what it models, what it claims, and how the claim will be evaluated.
Progress measured against the world.