Every day, one story that matters, broken down into three numbered scenarios, each with its own probability. Never fixed — it shifts as the situation changes.
Scénario
Friday, science
Dark Energy: CEA's AI Maps 100 Million Galaxies
Does the CEA's AI method applied to 100M galaxies from DES yield the tightest constraints ever on dark matter and dark energy, and reconcile weak lensing with Planck?
Publié le 16 octobre 2026
Share:
The facts
Our observable Universe conceals a dizzying enigma that modern science has been attempting to solve for decades. Ordinary matter—the stuff that makes up stars, planets, interstellar gas, and the human body—accounts for only about 5% of the total composition of the cosmos. The vast remaining 95% is split between two invisible components: dark matter*, which acts as a gravitational glue pulling galaxies together, and dark energy*, a mysterious force accelerating the expansion of space on large scales. Understanding these invisible forces is the premier challenge of modern astrophysics.
To map this invisible cosmic architecture, astronomers rely on a technique known as weak gravitational lensing*. When a beam of light travels across the Universe over billions of years, its path is slightly deflected by the mass of dark matter encountered along the way. This subtly distorts the apparent shape of background galaxies, much like looking at a landscape through an imperfect window pane or the bottom of a glass of water. By meticulously measuring these minuscule distortions, researchers can map the presence and density of invisible masses scattered throughout the cosmos.
Understanding it
Gravitational lensing works like a giant rippled glass pane.
Imagine looking at a street lamp through a textured shower door. The image of the light is distorted, even though the street lamp itself hasn't changed. In space, dark matter behaves like that glass: its mass curves space and subtly stretches the shapes of distant galaxies behind it.
On October 9, 2026, the CEA's Institute of Research into the Fundamental Laws of the Universe (CEA-Irfu), in collaboration with Imperial College London and University College London, announced a major methodological breakthrough. The team developed a novel hybrid analysis tool combining classic statistical calculations with advanced artificial intelligence algorithms. This method was applied to data from the third year of the Dark Energy Survey, a massive international astronomical survey captured from Chile, to extract information previously inaccessible to standard computational tools.
This unprecedented algorithmic processing scrutinized a catalog of 100 million galaxies spread across a gargantuan swath of the sky, equivalent to 5,000 times the surface area of the full Moon. Conventional statistical approaches relied primarily on simple correlations between two points in the sky, missing complex structures like filaments or dense clusters. By training deep neural networks on numerical simulations of the Universe, researchers managed to capture these non-linear patterns and drastically improve the precision of density measurements.
Understanding it
AI detects complex patterns invisible to traditional statistics.
Basic statistics only measure the average distance between two grains of sand on a beach. Artificial intelligence, however, can recognize the overall shape of sandcastles and dunes. It extracts vital details about the clumping of dark matter that simple mathematical formulas overlooked.
The resulting findings provide the tightest cosmological constraints ever computed solely from weak gravitational lensing. Remarkably, these measurements are fully consistent with benchmark data from the Planck space telescope on the cosmic microwave background*, which maps the earliest light emitted by the Universe 13.8 billion years ago. This historic agreement helps ease a persistent tension dividing the scientific community between observations of the early and late Universe.
This technological breakthrough comes at a strategic juncture for European space science, as ESA's Euclid space telescope, launched in 2023 with strong CEA involvement, prepares its first major results expected around 2027. Future observatories will increase astronomical data volumes a hundredfold in the coming years. The CEA's method demonstrates that artificial intelligence has become essential to managing this avalanche of information, though its absolute robustness against instrumental biases must now be verified by the global astrophysics community.
Number of galaxies analyzed by the CEA's AI method100 million Dark Energy Survey Year 3 data
Sky area covered (full Moon equivalents)5,000 times Dark Energy Survey footprint
Can artificial intelligence unveil the dark side of the Universe?
What we're assessing
FavorableL'IA valide le modèle cosmologique standard avec une précision record.
StableLes données progressent mais le mystère de l'énergie sombre reste ouvert.
DégradéDe nouveaux désaccords observationnels imposent de réécrire les lois de la gravité.
Favorable
40%
Likely
Validation of the Standard Model and Confirmation of the Cosmological Constant
In this first scenario, the analysis method developed by the CEA and its British partners is rapidly adopted and validated by the entire international scientific community. Cross-checks confirm that the neural networks introduce no artificial bias in estimating invisible mass. Future ultra-precise data from the Euclid space telescope and the Vera C. Rubin Observatory on the ground then fully confirm that dark energy behaves precisely as an immutable cosmological constant*, resoundingly validating the standard ΛCDM model*.
This path would put an end to decades of theoretical speculation by demonstrating that Albert Einstein's general relativity holds flawlessly across the entire observable cosmos. Unlike the stable scenario where doubt lingers due to a lack of mathematical consensus, this outcome would definitively confirm the parameters of the Universe with an error margin below 1%. Fundamental physics would thus solidify its theoretical foundations, even if the ultimate microscopic nature of quantum vacuum energy may still spark fresh experimental debates.
Indicators affected
Nombre de galaxies analysées par la méthode IA1,5 milliard↑100 millions (DES Y3)
Surface du ciel couverte (équivalent pleine Lune)30 000 fois↑5 000 fois
The France angleRecognition of the CEA as a global leader in AI-driven cosmological analysis. ↑ Rather favorable for France.
Stable
45%
Likely
Persistent Methodological Uncertainties and Extended Theoretical Debates
In this second outlook, the hybrid method combining AI and statistics is widely praised for its ingenuity but faces technical pushback over instrumental biases and the modeling of complex astrophysical effects. The poorly understood influence of galactic winds and supermassive black holes on ordinary matter complicates the pure interpretation of weak lensing. The resulting measurements remain compatible both with a fixed cosmological constant and with rival theories in which dark energy evolves slightly over billions of years of cosmic history.
Compared to the favorable scenario that would settle the debate immediately, this path prolongs indecision among astrophysicists for another five to ten years. Massive data delivered by the Euclid mission around 2027 and ground-based telescopes will provide valuable clues, but without offering the decisive signal required to conclusively rule out alternative models. Science would thus advance in cautious steps amid intense methodological debate, sustaining healthy competition across international research teams.
Indicators affected
Nombre de galaxies analysées par la méthode IA500 millions↑100 millions (DES Y3)
Surface du ciel couverte (équivalent pleine Lune)15 000 fois↑5 000 fois
The France angleContinued leading scientific role for French laboratories. ↑ Rather favorable for France.
Degraded
15%
Unlikely
Resurgence of Observational Tensions and Questioning General Relativity
In this third scenario, expanding AI-driven analyses to much larger datasets revives irreconcilable discrepancies between different cosmic observation methods. While consistency with Planck appeared settled on the DES dataset, sharp divergences resurface when weak gravitational lensing is cross-referenced with distant supernovae or galaxy cluster studies. The accumulation of data does not narrow the gaps; instead, it proves that the current theoretical framework is incomplete.
In contrast to the favorable scenario where the standard model emerges reinforced, this reproducibility crisis would force modern physics to confront a major paradigm shift. Researchers would need to abandon the simple cosmological constant to explore modified gravity theories, a fifth fundamental force, or exotic interactions within the dark sector. The Universe would turn out to be far more complex and baffling than any equations formulated to date.
Indicators affected
Nombre de galaxies analysées par la méthode IA100 millions→100 millions (DES Y3)
Surface du ciel couverte (équivalent pleine Lune)5 000 fois→5 000 fois
The France angleNeed to realign national research programs and theoretical models. ↓ Rather unfavorable for France.
Ordres de grandeur indicatifs pour les 3 scénarios ci-dessus, estimés avec l'information disponible à la publication et réévalués si la situation change — jamais des prévisions garanties. Learn more about our method → · How to read a probability →
Key takeaways
Can the artificial intelligence method developed by the CEA definitively pin down the properties of dark energy and resolve cosmological debates?
CEA-Irfu analyzed 100 million galaxies from the Dark Energy Survey using a novel hybrid method blending statistics with deep neural networks, achieving record constraints consistent with data from the Planck satellite.
Not immediately: our most likely scenario (45%) suggests methodological uncertainties will keep the debate open for several years, compared with a complete validation of the standard model (40%) or a theoretical overhaul (15%).
The key signal to watch will be the release of the first major cosmological results from ESA's Euclid space telescope, expected around March 2027.
Moderately positive
Our assessment of the impact for France: moderately positive.French research positions itself at the heart of future space discoveries, with an 85% cumulative probability across scenarios strengthening its leadership in cosmology and artificial intelligence.
Si tu devais retenir 1 chose
Le CEA a analysé 100 millions de galaxies avec l'IA pour obtenir les mesures les plus précises sur l'énergie sombre.
Don't miss the next edition:
For those new to the topic
Quick glossary
dark matter
Hypothetical form of invisible matter that neither scatters nor emits light, detectable only through its gravitational attraction on galaxies.
dark energy
Mysterious, invisible component of the Universe responsible for the accelerating expansion of the cosmos on large scales.
weak gravitational lensing
Subtle distortion of the apparent shape of distant galaxies caused by the bending of space around intervening mass.
cosmic microwave background
Relic radiation emitted approximately 380,000 years after the Big Bang, constituting the oldest thermal image of the observable Universe.
cosmological constant
Mathematical parameter introduced by Albert Einstein representing a homogeneous and unchanging energy density inherent to empty space.
standard ΛCDM model
Benchmark theory in cosmology describing a Universe dominated by the cosmological constant Lambda and cold dark matter.
