OpenAI, one of the leading firms in this area. A group from OpenAI said they had solved the Navier-Stokes problem, one of seven “Millennium Problems” chosen in 2000 by the Clay Mathematics Institute as the hardest and most important in the field. OpenAI thus seems to have gazumped Tristan Buckmaster (pictured) and Levent Alpöge, two mathematicians who had been labouring on that task, alongside some AI, and were thought close to a solution. The firm released a 166-page paper describing its work, which employed models not yet publicly available. Despite its length, however, this paper contained little of the explanatory detail which mathematicians typically provide when sharing a new discovery. At first sight, this spat sounds like a highfalutin version of the pearl- clutching that accompanied the spread of electronic calculators (loss of mental-arithmetic skills), or even of ball-point pens (that a lack of sensory impact on the brain might have adverse consequences for writing style). Those concerns, real enough at the time, now sound as laughable as Socrates’s—for the written word has augmented memory, not replaced it. But there is a difference. The letter writers do not seem to be trying to stop the use of AI in their discipline. Indeed one of them, Terence Tao, argued of it, in a video posted by OpenAI on X, a social-media platform, in May, that “It allows me to experiment; I will try crazier things.” He then listed AI’s helpful capabilities and his hopes of how it would be especially useful if the work was shared, as is mathematical culture and tradition. Their concern, rather, is that definitive answers to problems are but one aspect of maths. What is often more valuable is that the process of arriving at important proofs leads to the asking of yet more questions and the creation of new areas of study. This, they fret, risks being lost if AI does the grunt work—and doubly so if that grunt work is done by organisations that then fail to release sufficient details to the outside world for human beings to read and understand. As Hugo Duminil-Copin, another of the signatories, puts it: airdropping someone on the summit of Mount Everest is rather different from climbing it.

If this sounds a self-serving argument by professionals fearful for their jobs, it no doubt is. But that does not make it entirely meritless. Mathematicians may start to work less openly, in order to avoid the fate of Dr Buckmaster and Dr Alpöge. Another, more subtle, danger is that leaning on the crutch of AI will erode habits of mind and make human mathematicians less capable practitioners of their field. Here, there is a historical analogy: a phenomenon called cognitive offloading, in which, by reducing the load on human memory, information technology causes the mind’s skills to atrophy in the sorts of ways that bothered Socrates. Work published as long ago as 2011, by Betsy Sparrow of Columbia University and her colleagues, confirmed the common-sense idea that people turn to Google-searching answers when faced with tough questions, instead of attempting to recall what they know. A more recent study, by Michael Gerlich of the SBS Swiss Business School, similarly found “a negative correlation between the frequent use of AI tools and critical-thinking abilities”. The row therefore encapsulates wider fears about the spread of AI. OpenAI’s executives have said in the past that their version of the future is one in which AIs work with mathematicians on “figuring out what problems are actually important to tackle”. That points to a world where cognitive mathematical effort may indeed be offloaded onto software agents. It is the details of how this happens which will determine whether that is a good thing or a bad one and so whether, in this case, Socrates may have been right after all. ■ This article was downloaded by zlibrary from https://www.economist.com/science-and-technology/2026/09/11/top-mathematicians-are- outraged-by-openais-methods

Science & technology | Xenocortication Researchers have created a mouse with a partly human brain It will help the development of drugs for neurological illnesses Sep 17th 2026 Xenotransplantation is the insertion of an organ from one species into another. When the organ is an animal’s and the recipient human (as with the transplant of pig kidneys into people whose own are failing), the idea is to prolong life. But what about the other way round—transplants from human to animal? With one crucial distinction, that is what Sergiu Pasca of Stanford University and his colleagues have been up to. The distinction is that they have been transplanting, into mice, not actual human organs (which would be both unethical and physically impossible) but “organoids” grown from stem cells that derive, in turn, from human skin. The organoids in question are made of brain cells.

Xenocortication, as Dr Pasca calls this process, and which he and his colleagues describe in a paper just published digitally by Nature, results in rodents that have brains which are, literally, part human. The trick is to make room for the human neurons by engineering a mouse’s genes to stop most of the cells of its cerebral cortex and hippocampus developing. A human brain organoid composed of appropriate cells is then transplanted, in the first few days after the animal’s birth, into the void thus created. It makes itself at home and connects up with the outside world as if it were mouse tissue. The purpose of this is to study early development of human brains, which is both practically and ethically impossible if the tissue in question is still in a human, but which is crucial to understanding the roots of many neurological illnesses. Developmentally, neurons march to the beat of an internal drum, so for the first few months of a mouse’s life they still behave like fetal cells, even as the animal itself matures. Armed with their new tool, Dr Pasca and his colleagues plan to look into several conditions, starting with hypoxia-induced cerebral palsy (usually caused by a difficult birth), to which human neurons are particularly susceptible. Xenocortical mice deprived of oxygen suffer damage to their human neurons but not their murine ones. Comparing their behaviour with that of both unengineered controls and those lacking a cortex and hippocampus, but without compensating human cells, confirms the varied effect. Other conditions the researchers have in their sights include fronto-temporal dementia, epilepsy and autism. Their hope in these cases is that modified animals might act as test beds for drugs. Testing neurological pharmaceuticals in unmodified animals has a notorious record of highlighting molecules that then turn out to be duds. The idea of wiring up human neural tissue to help run another animal is, of course, something to be approached with caution. The team has consulted a range of experts, including neuroscientists, legal scholars and patient advocates, over and above the normal ethical oversight that would be applied to work like this. Any risk of some uniquely human quality emerging in these animals, however, is greatly mitigated by the mismatch between the developmental pace of human beings and mice, the small

number of human neurons that will fit into a mouse cranium, and the evolutionary distance between mice and humans, which last had a common ancestor about 70m years ago. Were someone to try something similar in, say, a monkey, that calculation would change. Such an experiment would certainly not pass ethical scrutiny. Whether it might nevertheless happen somewhere, sometime, is rather a different question. ■ This article was downloaded by zlibrary from https://www.economist.com/science-and-technology/2026/09/16/researchers-have- created-a-mouse-with-a-partly-human-brain

Science & technology | Actions and words A new device helps severely paralysed patients communicate Eventually they might be able to control robots or exoskeletons Sep 17th 2026 Brain-computer interfaces have come a long way quickly. The first successful decoding of vowel sounds from a neural implant was reported in 2009. In 2021 a team from Edward Chang’s lab at the University of California, San Francisco decoded the words and sentences of a patient who could not speak. A new paper by Dr Chang’s group, published in Nature Neuroscience, describes the passing of another milestone. He and his colleagues have enabled two severely paralysed people to communicate with them, by means of an avatar, using both words and gestures. That is an auspicious result for the several million individuals around the world paralysed by strokes, traumatic injuries and degenerative illnesses such as motor-neuron disease.

The group did this by implanting, through a small incision in the skull of each patient, an array of sensors that covered a large part of the motor cortex —the brain region responsible for controlling bodily motion. This procedure took over two hours, in part because the process for connecting the sensor array to a data-carrying cable is fiddly. But the team are working on a short- cut, using a wireless device, that would reduce this to 40 minutes. The implants themselves allowed Dr Chang and his team to collect signals as the patients attempted to say particular words and make particular gestures. They then used those data to train two machine-learning models for each patient—one to decode the words and the other the gestures—allowing them to reconstruct both actions concurrently with approximately two-thirds accuracy. It is the first occasion that such simultaneous decoding has been achieved, allowing patients to communicate more expressively. There is further to go. These two initial individuals were restricted to a vocabulary of about ten words and phrases, such as “hello” and “nice to meet you”, and a similar number of gestures, such as a clap or a shrug. But Dr Chang and his collaborators are working on raising this to more than 100 actions. They also hope that the sensor array’s wide reach across the cortex means it may be possible to monitor not only intentions to communicate but also actions intended for the whole body. That might permit paralysed individuals to operate devices such as robots and exoskeletons, which really would be life-changing. ■ This article was downloaded by zlibrary from https://www.economist.com/science-and-technology/2026/09/14/a-new-device-helps- severely-paralysed-patients-communicate