AI & scientific discovery · An open research initiative
AI should do more than find answers on an existing map. It could help us reconsider how the map is drawn.
Explore our focusAn oblique view reveals the layered structure of the torus.
Mathematical illustration, not experimental data. A different view is not evidence.
Why parallax?
When variables, tasks and evaluation criteria are already fixed, better predictions do not necessarily change how we understand a phenomenon. We look one step earlier: how do we organize experience, and why do we ask these questions?
Parallax is a methodological metaphor. Let different representations face the same world, exposing assumptions that a single perspective may hide. Then let experiments determine whether those differences matter.
What we care about
Make the premises of scientific inquiry part of the inquiry itself.
Objects, relations, variables and scales need not remain fixed premises. Can a new representation unify anomalies and produce testable predictions that the old one cannot?
A new name is not yet a new concept.
Which experiment would actually change our judgment? A question must specify what to change, what to measure and what competing explanations predict—not merely sound novel.
Disagreement alone is not scientific value.
Instruments, sampling and processing shape what becomes an observation. Does an anomaly belong to the system, or to the way we observe it? The measurement model must be part of the explanation.
Accurate prediction is not correct attribution.
Commit to predictions before running experiments. Record revisions, scope and failures. Let independent evidence decide which explanations survive—and allow our own methods to be refuted.
A compelling story is not a scientific discovery.
Three historical shifts
These breakthroughs did more than borrow a tool. They changed what physicists thought they were looking at.
Botany × statistical physics
The irregular motion of particles in water looked like noise. Einstein stopped asking about each collision and studied many trajectories instead. A measurable pattern in randomness became evidence for invisible molecules in motion.
Read the historyDifferential geometry × gravity
In the Newtonian view, gravity pulls one object toward another. General relativity changed the language: mass shapes spacetime, while matter and light follow its natural paths. The stage itself became part of the physics.
Read the historyCondensed matter × particle physics
The rules of a magnet favor no direction, yet cooling can make all its parts choose one. Particle physics imported that viewpoint: a vacuum need not be blank. It can settle into a state that changes how particles behave.
Read the historyA working hypothesis
Rather than decide what a “good paradigm” is in advance, give ideas a chance to be tested, revised and rejected.
Keep different executable representations in play.
Find experiments where predictions genuinely diverge.
Commit to predictions, then observe and intervene.
Track lineages, including their failures.
Coevolution of paradigms and questions is one candidate implementation, not a demonstrated advantage. Algorithms and application domains may change; testability and independent evidence are non-negotiable.
The limits of our claims
This is a long-term research direction, not an accomplished scientific breakthrough.
Read experiments & auditsExisting explorations demonstrate structural recovery in constrained synthetic worlds. They do not establish the discovery of new natural laws.
A distinctive advantage for coevolution has not been supported. Audits have narrowed earlier claims about the mechanism behind the results.
Previous stop decisions stand. New research requires new questions and independent tests.