02 / PHASE SPACELINKED
First angle × canonical momentum · rad / kg·m²/s
PROJECT 024 · Frontier research concept · Interactive experiment
Alien Instruments
“What if the next scientific revolution requires a new way of perceiving?”
INSTRUMENT 01 / DOUBLE PENDULUMSIMULATED DATA · 36 SECONDS · TWO INITIAL STATES
One reality.
Unfamiliar eyes.
Change the instrument. Follow the same observation into a different way of seeing.
What is actually learned?
01 / OBSERVATION FIELDPhysical space
— A / reference┄ B / perturbedFaint lines: full recorded trajectory
Position of the second mass · metres
Hover a plot to pause and inspect its nearest observation. On touch, drag across it. All views share the same time; the slider also works with arrow keys. Playback starts only when you choose it.
03 / LEARNED SPACEPCA · EXPERIMENTAL
Two learned principal components · feature units
A projection, not a physical law. Colour identifies the trajectory, not confidence.
04 / SEPARATIONLINKED
Distance between second masses · metres
THE INSTRUMENT MUST EXPLAIN ITSELFBeautiful is not
Beautiful is not
the same as true.
The learned view is real PCA fitted to this simulation. It is a transparent baseline for the larger research idea—not a novel discovery engine.
Reconstruction residual on A from 18–36 s, relative to its centred feature energy. Lower is less discarded information, not better prediction.
Across both trajectories, relative to the fixed 29.43 J gravitational energy scale. A numerical diagnostic, not a guarantee of trajectory accuracy.
Open the instrument: assumptions, learned weights and limits
01 / Physical model
Two 1 kg point masses; two rigid, massless 1 m rods; gravity 9.81 m/s². No friction. Second initial angle: 110°. Both initial angular velocities: zero. Fourth-order Runge–Kutta at 1/240 s; display samples every 1/30 s.
Equations and derivation — myPhysicsLab ↗02 / What the model learns
Six features: sin θ₁, cos θ₁, sin θ₂, cos θ₂, ω₁/√g, ω₂/√g (using a 1 m rod scale). The first 540 observations of A determine the mean and covariance. Two principal directions are extracted by power iteration. The same fixed transform is then applied to both trajectories.
| Feature | PC1 | PC2 |
|---|---|---|
| sin θ₁ | 0.070 | 0.051 |
| cos θ₁ | -0.042 | 0.054 |
| sin θ₂ | 0.006 | 0.168 |
| cos θ₂ | -0.024 | -0.168 |
| ω₁/√g | -0.056 | -0.967 |
| ω₂/√g | 0.995 | -0.061 |
03 / Where it can mislead
The test segment comes from the same trajectory: it is not independent evidence of generalisation. PCA maximises retained training variance, not scientific usefulness. Visual crossings need not be physical encounters. The phase angle wraps at ±π. Morphing uses separately normalised view coordinates; intermediate shapes have no physical units. Finite-time separation is not a Lyapunov exponent.
The proposed research platform would add nonlinear representations, out-of-distribution tests and physical experiments. Those capabilities are not implemented here.
THE QUESTION THAT REMAINS
Is the structure in the world,
or in the way we learned to look?
01 / PREMISE
The beginning
Every scientific instrument makes something visible and something else disappear. A telescope extends sight; a coordinate system decides what sight can mean. Alien Instruments begins with a different ambition: to invent instruments that change the representations through which we reason. Could a machine propose a structure that a human would not naturally choose, and could we learn to think through it? Not an image of alien intelligence, but a disciplined encounter with unfamiliar ways of organising the same evidence.
We inherit objects, axes, categories and interfaces, then mistake their familiarity for necessity. A trajectory can look disordered in physical space while revealing useful structure in another representation. Conversely, a compelling cluster can be an artefact of a projection. The bottleneck is not simply the amount of data available. It is our ability to explore alternative descriptions without losing the distinction between what belongs to a system and what our instrument has imposed on it. Better perception must remain accountable to the evidence it transforms.
02 / THE PRODUCT
What it could become
I imagine a research environment in which one physical experiment can be inhabited through several competing instruments. Observations retain their identity as the view changes. A researcher can follow one event across physical space, phase space and a learned representation, inspect what each transformation discards, and ask where the descriptions disagree. The first working experiment here is deliberately bounded: two simulated double pendulums, a common clock, and four linked views. The learned view uses a transparent two-component principal-component projection, not a novel scientific theory. Its limitations are part of the experience. The larger platform would search over candidate representations and help researchers design experiments that distinguish useful structure from persuasive geometry.
For whom
- Scientists investigating nonlinear physical systems
- Research teams evaluating learned representations
- Educators teaching the relationship between models and evidence
- Artists and philosophers of scientific perception
Core capabilities
- Linked observations across competing representations
- Direct manipulation of initial conditions and shared time
- Continuous transitions with observation identity preserved
- Held-out reconstruction tests and inspectable transformation weights
- Numerical energy-drift checks
- Optional, explicitly mapped sonification
- Future research: representation search and discriminating experiments
The project becomes meaningful only when a new technical possibility is translated into a clear human advantage, an experience people can understand, and a system capable of earning trust over time.
03 / FOUNDATIONS
Intelligence and mathematics
The first instrument learns a linear basis from the first half of one simulated trajectory and tests it on the second half. It is a modest but real form of data-adaptive representation, with no language-model service or autonomous discovery claim. A research platform could extend this to nonlinear encoders, symmetry-aware models and competing predictive objectives. Candidate generators would propose representations; independent evaluators would challenge them under held-out conditions, perturbations and measurement noise. Human researchers would decide what physical interpretation, if any, survives. A representation earns authority through useful tests, never through visual elegance.
The experiment integrates the frictionless double-pendulum equations with fourth-order Runge–Kutta at a fixed 1/240-second step. Physical coordinates show the second mass. A phase portrait pairs the first angle, wrapped to one revolution, with its canonical momentum. A six-feature state—sines and cosines of both angles plus scaled angular velocities—is centred using training data and projected onto two principal directions. A held-out residual measures the discarded feature-space information; it is neither a prediction accuracy nor evidence of a new physical law. A separate view tracks the distance between the two second masses. This finite-time separation is not a Lyapunov exponent. Every instrument is an explicit choice about what to preserve.
04 / VENTURE LOGIC
How it might live
Begin with an open, reproducible instrument for nonlinear dynamics and a small partnership with a research or teaching laboratory. The first question is whether linked representations help people formulate better, testable questions than conventional plots. A professional platform could later support private datasets, reproducible transformation pipelines and collaborative model review. The defensible asset would be a body of validated instruments and the evidence describing where each succeeds or fails—not a proprietary catalogue of beautiful pictures. Expansion into other physical domains should follow domain-specific validation.
For me, a venture is more than an interesting technology. It needs a narrow first user, a repeated problem, a distribution path, a credible advantage and a reason to improve as more people use it. I would test those conditions before deciding whether this idea should become a company, a product, an open technology or an ongoing research programme.
05 / DESIGN PRINCIPLES
Rules for making it real
- 01
One observation must remain traceable through every view.
- 02
A learned coordinate is not a discovered physical quantity.
- 03
Expose the information a representation discards.
- 04
Separate visual interest from predictive and experimental usefulness.
- 05
Keep incompatible models available when evidence cannot choose.
- 06
Make uncertainty legible without inventing a probability distribution.
- 07
Let people pause, inspect and navigate without motion or sound.
- 08
The best instrument may reveal that its own assumptions are wrong.
06 / ROADMAP
From question to company
- 01Frame
Build the linked double-pendulum instrument with transparent numerical methods.
- 02Prototype
Compare the experience with conventional plots in a small researcher study.
- 03Prove
Test representations across new initial conditions, noise levels and integrator resolutions.
- 04Build
Introduce nonlinear candidate representations and independent evaluation protocols.
- 05Advance
Connect disagreements between models to feasible physical experiments.
- 06Advance
Publish validated instruments, failure cases and reproducible evaluation data.
07 / HONEST QUESTIONS
What could go wrong
Serious imagination includes the possibility that an idea should change radically—or should not exist. These are the tensions the project would need to resolve:
- A projection creating apparent structures that do not exist in the physical system.
- Temporal correlation making a held-out trajectory segment an optimistic test of generalisation.
- Numerical error being mistaken for physical divergence.
- A persuasive interface concealing weak predictive performance.
- Sonification encouraging associations unsupported by its mapping.
- Representation search rediscovering known mathematics while claiming novelty.
- Compressing unfamiliar reasoning into forms that erase what made it useful.
EVIDENCE / BOUNDARIES
Built from state-of-the-art research, not visual spectacle
Reality Syntax extends active research in declarative visualization, scalable interaction, uncertainty communication, accessible description and immersive analytics. Its more radical claims—especially cross-modal equivalence and proof-carrying representations—would require new standards, controlled studies and co-design with disabled people before they could be treated as reliable.