The Result
We fed our system 42 measurements of supermassive black holes, just their masses and the average speed of stars in the surrounding galaxy, and asked it a simple question: What's the formula?
The system did not find a formula. Instead, it found something far more important: it proved that a formula is impossible with the known variables, and that a hidden quantity must exist.
Specifically, it proved that the only way to write a physically consistent equation connecting black hole mass (M) to stellar velocity (σ) using Newton's gravitational constant (G) is to introduce a hidden acceleration constant a₀:
This is not a fit. This is not a correlation. This is a mathematical proof that the data requires an unknown acceleration scale to be physically consistent.
That acceleration scale is exactly what physicist Mordehai Milgrom proposed in 1983 as the foundation of MOND (Modified Newtonian Dynamics), the primary rival theory to Dark Matter.
Why This Matters
In 1983, Milgrom noticed that galaxies start behaving strangely at a specific acceleration threshold: about 1.2 × 10⁻¹⁰ m/s². Below this value, gravity appears stronger than Newton predicts. He proposed that either our law of gravity is wrong at low accelerations, or there is a universal acceleration constant, a₀, built into the fabric of spacetime.
For 43 years, the physics community has debated whether Dark Matter or MOND better explains the data. Both sides have compelling arguments. Neither has a definitive proof.
Our AI didn't take sides. It didn't know about Dark Matter. It didn't know about MOND. It didn't know about Milgrom. It was given:
- A table of black hole masses and stellar velocities
- Newton's gravitational constant G
- The speed of light c
From these inputs alone, the system mathematically proved that the existing variables cannot form a complete physical equation. It then determined that a hidden acceleration is required to make the equation work:
This is the M-sigma relation in its most fundamental form. The AI derived it from scratch.
What the AI Actually Did
The system didn't "guess" or "correlate." It performed abductive reasoning, the same type of logical inference that Newton used when he postulated the existence of gravitational force.
Newton's situation in 1687: he had Kepler's data (orbital periods and distances) but no concept of "force." He realized that the data couldn't be explained without inventing a new physical quantity that mediated the relationship between mass and acceleration.
Our system's situation: it had the M-sigma data (black hole masses and velocity dispersions) but no concept of a₀. It realized that the data couldn't be made physically consistent without postulating a new quantity with the dimensions of acceleration.
The logic is identical. The AI automated the "Aha!" moment.
The Broader Implications
The M-sigma relation is one of the tightest empirical correlations in astrophysics. A black hole is the size of our solar system. The galaxy bulge whose stars define σ is 100,000 light-years across. There is no known physical mechanism that should connect them with a simple power law.
The fact that the AI independently concludes that a hidden acceleration scale is needed, without any knowledge of the MOND debate, adds weight to the hypothesis that something fundamental is missing from our understanding of gravity at galactic scales.
Whether that missing piece is a new particle (Dark Matter) or a new law (MOND) remains an open question. But the AI has now independently confirmed that the missing piece has the specific dimensions of acceleration, exactly as Milgrom predicted 43 years ago.
The Data
The analysis used 42 galaxies from the McConnell & Ma (2013) compilation, spanning black hole masses from 10⁶ to 10¹⁰ solar masses and velocity dispersions from 77 to 385 km/s. No fitting was performed. The result is a purely mathematical argument: the only way to connect M, σ, and G into a consistent relationship is to introduce a variable with dimensions of acceleration.
The estimated acceleration scale from the M-sigma data is consistent (within numerical prefactors) with the MOND acceleration a₀ ≈ 1.2 × 10⁻¹⁰ m/s² that appears independently in galaxy rotation curves, the Tully-Fisher relation, and the radial acceleration relation.
What Comes Next
The system that made this discovery is being pointed at other unsolved problems in physics, high-temperature superconductivity, neutron star interiors, and the particle mass hierarchy. In each case, the approach is the same: give the system raw data and let it tell you what's missing.
Sometimes the most important discovery isn't finding the answer. It's proving that the question requires something new.
Want to see what this approach can do in your domain? Contact us or explore our model marketplace.