Pond Scum, Pharaohs and One Xenon Flash

The 2026 Nobel Prize in Physiology or Medicine goes to Peter Hegemann, Georg Nagel and Karl Deisseroth for developing optogenetics, which uses light to switch on or silence nerve cells chosen by the genes they express.

The problem it solved

Neuroscientists had mapped which genes are active in which populations of neurons. They could also delete those neurons genetically and watch what changed. That method has limits. The brain can adapt to losing cells, and removing cells early in development can change how their partner neurons grow. The better test is to turn specific neurons on and off in an otherwise intact brain.

Nerve impulses depend on ion channels, membrane proteins that let charged atoms through only under certain conditions. Earlier attempts to take control of them never caught on. They needed several inserted genes, special chemicals, or lasers strong enough to damage cells.

From algae to mice

The answer came from Chlamydomonas, a single-celled alga with an eye spot that steers it toward light. At Humboldt University in Berlin, Hegemann attached an electrode to the alga and found that a flash of light set off a very fast influx of ions. He found two genes that resembled a light-activated ion pump from archaea. Blocking them with RNA interference reduced the light-triggered current, which tied them to the effect.

Working with Nagel, an ion-flux expert at Würzburg, Hegemann showed that one protein lets only protons through while the other passes a range of positive ions. The two respond to somewhat different wavelengths. Both sense light with a vitamin A relative, as the receptors in our eyes do. The class is now called . Nagel showed they work in frog embryos and cultured human cells, and that shining light on worms carrying the genes changed their behavior.

On September 1, the LUX-ZEPLIN (LZ) collaboration reported one event in its detector that a dark matter particle could have caused. Ordinary known physics could also explain it. In the weeks since, dozens of papers have appeared on arXiv.org trying to explain it. These are preprints, not peer-reviewed work, and the signal itself is unconfirmed.

What LZ saw, and why the higgsino fits

LZ sits at the Sanford Underground Research Facility in South Dakota. It is a large tank of liquid xenon, and it watches for a dark matter particle hitting a xenon nucleus and knocking it away. This event's recoil had unusually high energy. Most dark matter candidates should cause many low-energy recoils as well as high-energy ones, but LZ saw none.

That gap is why many theorists turned to the higgsino, the proposed superpartner of the Higgs boson in supersymmetry. Supersymmetry says every known particle has a heavier partner. A higgsino would have two mass states, and it could only scatter off a nucleus when the collision carried enough energy to lift it from the lighter state to the heavier one. Low-energy recoils would therefore never happen. Katherine Freese (UT Austin) co-wrote a higgsino paper within 24 hours of the announcement.

The problems

Two objections came quickly. First, higgsinos would get trapped in the sun and annihilate each other there, making high-energy neutrinos. Harikrishnan Ramani and a colleague worked out that the IceCube detector at the South Pole should have seen those neutrinos, and it has not. Second, Tracy Slatyer (MIT) and colleagues spotted a plot in an appendix to LZ's paper covering higher energies. If the event was a higgsino, more events should show up at those energies, and none appear. LZ spokesperson Rick Gaitskell says the detector wasn't calibrated for that range, and the team made no claims about it either way.

Egyptologists have long disagreed about whether the earliest pharaohs, the kings of the First Dynasty, were buried with servants and officials who were killed to serve them in the afterlife. This practice is known as retainer sacrifice. There has been little physical evidence either way. Many of the earliest royal tombs were plundered, hastily excavated, or both.

A new paper in Frontiers in Environmental Archaeology goes back to the remains. Roselyn Campbell, a bioarchaeologist at Purdue University, re-examined a collection of skulls taken from these tombs in the 1920s. She found that more than one-third show evidence of blunt-force trauma. She concludes that this fits with the people having been killed as retainer sacrifices, but it doesn't prove it.

Why look at the bones again

Campbell says the debate has hardened into two camps. "Either retainer sacrifice definitely happened, end of discussion, or it definitely didn't happen, end of discussion," she told Ars. "But nobody's actually looked at the bones in a very long time." Forensic methods have improved a lot in the century since these skulls were dug up. That improvement is the main reason to look again. Modern trauma analysis can pick up injuries that excavators in the 1920s weren't looking for or couldn't interpret.

The age of the collection also cuts the other way. Older sets of remains are often incomplete, and some were altered to preserve them. Both problems destroy forensic information. This limits how much any re-analysis can show.

Campbell's wider research is about how cultures decide which kinds of violence are acceptable. She warns against reading modern ideas into ancient practice. In her view, Egyptian archaeology carries a legacy of European colonial excavators who "came in with very specific ideas about what this culture was," and old excavation data has to be put in that context.

Every decision comes with a sense of how likely it is to be right. Confidence was long a side issue in psychology. Now it is studied in its own right, from where the brain represents it to how it breaks down in psychiatric disorders. In an essay for The Transmitter, neuroscientist Ariel Zylberberg asks what the brain is actually computing when it reports confidence.

Two competing pictures

The dominant view is hierarchical. The brain first commits to a choice, even if only provisionally, and then estimates confidence by reading out features of how that choice was reached: how much evidence piled up for each option, how long the decision took, how far the winner pulled ahead, and how often the decider changed their mind. This account has done well at explaining choices, response times and confidence ratings across many tasks.

Zylberberg sets out an alternative that comes from neurobiology, the intentional framework developed by Michael Shadlen and colleagues. In a classic experiment, a monkey judges whether a cloud of dots drifts left or right and reports its answer with an eye movement. Eye-movement-planning neurons in the lateral intraparietal area (LIP) track the developing decision. Their activity doesn't encode an abstract statement like "the motion is rightward." It encodes a growing intention to look at the right-hand target. On this view, deciding and preparing to act are the same process.

Now give the monkey four targets: left-high confidence, left-low, right-low and right-high. The hierarchical view says the brain picks a direction, then rates its confidence, then chooses a target. The intentional view says the four actions are themselves the options. Four accumulators run in parallel, each fed by the sensory evidence, and whichever reaches threshold first sets both the choice and the confidence at once. Strong rightward evidence favors right-high; weak rightward evidence favors right-low.

On September 8, 2026, OpenAI announced that its models had solved one of mathematics' most famous problems. It followed a summer in which AI systems produced proofs of decades-old conjectures almost every week. The article doesn't name the problem. Its subject is the reaction. Scott Aaronson of UT Austin wrote that human mathematicians are "forevermore dethroned" as the main producers of theorems. At a Berkeley talk two days later, about 150 students and researchers confronted Ken Ono, a mathematician who left academia for the start-up Axiom Math. A 50-minute session ran past two hours.

Proofs without understanding

Cepelewicz argues that the problem is not correctness. The AI proofs are checked by a system that verifies their logic, so mathematicians know they are true. They just can't tell why. The proofs are close to unreadable. They skip key steps, spend pages on irrelevant ideas and don't link the new result to existing work. In pure math, she argues, the main payoff has always been the understanding, tools and side discoveries that come from struggling with a problem. AI goes straight to the answer.

The bigger damage is to incentives. Hiring, promotion and credit all depend on proving theorems, an arrangement that worked as long as solving a problem and understanding it went together. Caltech doctoral student Tasmin Chu says AI "breaks our system."

The article lists concrete effects:

• Secrecy: OpenAI says it is holding back more than 100 important results, which can leave mathematicians working on questions already answered.
• journals and arXiv are filling with AI-assisted proofs no one can properly check. arXiv now limits authors to .