Enjoying the show? Support our mission and help keep the content coming by buying us a coffee.
The universe's biggest unknowns—dark matter and dark energy (the 95% of reality that is invisible)—are no longer intractable. AI has become an essential research partner, moving cosmology into an era of industrial-scale discovery.
Our program unpacks how AI is achieving unprecedented precision in measurement and mapping, and confronts the stunning possibility that the data is pointing to cracks in Einstein's theory of gravity.
AI is fundamentally necessary because the signals from the dark sector are incredibly faint and buried in massive data streams.
Doubling Accuracy: The core technique is weak gravitational lensing (gravity bends light). The UCL team trained AI using thousands of simulated universes and applied it to the Dark Energy Survey (DES) data, doubling the accuracy of measurements for dark matter and dark energy. This precision was achieved without touching the telescope, saving billions of dollars and decades of observation time.
Mapping the Cosmic Web: The Dark Energy Spectroscopic Instrument (DESI) is mapping 18.7 million objects—more than twice the number of objects of all previous 3D spectroscopic surveys combined. The system relies on 5,000 robotic fiber optic positioners that work in parallel. The AI-accelerated processing allows validated results to be back in researchers' hands by the next morning.
Seeing the Invisible: AI trained on advanced simulations confirmed the existence of dark matter bridges—tenuous filaments connecting galaxies—providing the most direct visualization yet of the invisible scaffolding of the cosmos.
Real-Time Discovery: The AI serves as an essential filter for catching transient events (supernovae) that fade fast. AI algorithms flagged supernova SN2023sk early, allowing follow-up that revealed the star was triggered by a catastrophic encounter with a black hole—a direct mode of stellar explosion that would have been missed entirely by human analysis.
AI's precision measurements are forcing cosmologists to confront a potential conflict at the heart of their model:
The Mismatch: The enhanced DES measurements suggest that matter in the universe is distributed more smoothly and less clumpy than predicted by the Lambda CDM model (which extrapolates structure growth from the early universe CMB data).
The Implication: If the measurements are correct, it suggests two possibilities: 1) Einstein's General Relativity might be slightly incomplete on the largest cosmic scales, or 2) there is unknown physics at play (a new interaction of dark matter/dark energy) that naturally suppresses structure growth.
AI innovation in cosmology is driving breakthroughs in other fields:
Hardware Power: DESI and other projects rely on dedicated supercomputing centers (NERSC) that are heavily relying on GPUs (Graphics Processing Units) for parallel processing, providing a ≈40 times speed-up for spectral analysis.
Terrestrial Impact: The core technology—real-time anomaly detection in complex, high-volume data streams—is highly transferable, with applications in medical diagnostics, financial fraud prevention, and national security monitoring.
Democratization: The future involves natural language interfaces (SIMA) where scientists can ask complex, multi-data set questions (e.g., "Show me all merging galaxies with supermassive black holes...") without needing to be expert coders, lowering the barrier to entry for cutting-edge research.
Final Question: AI has delivered the precision to uncover the "lumpiness puzzle." If the evidence suggests that our theory of gravity is incomplete or that there is a major missing piece in our understanding of the universe's contents, what does that mean for the next generation of cosmology?
