Huawei has postponed its upcoming HDC.Cloud developer conference as Chinese authorities try to control the spread of the deadly coronavirus detected in the southeastern city of Wuhan. From a report: The controversial company's event was going to take place in Shenzhen -- which lies more than 700 miles south of Wuhan -- Feb. 11-12, but it's been rescheduled to March 27-28. "Based on the prevention and control of the pneumonia epidemic situation of the new coronavirus infection, we attach great importance to the health and safety of all the participants," Huawei said in its announcement. It also asked staff to avoid traveling to Wuhan and limit contact with animals, Reuters reported, and said it set up an outbreak prevention and control team in the city.

An anonymous reader quotes a report from the BBC: Scientists say they may have discovered why stress makes hair turn white, and a potential way of stopping it happening without reaching for the dye. Researchers behind the study, published in Nature, from the Universities of Sao Paulo and Harvard, believed the effects were linked to melanocyte stem cells, which produce melanin and are responsible for hair and skin color. And while carrying out in experiments on mice, they stumbled across evidence this was the case.

Pain in mice triggered the release of adrenaline and cortisol, making their hearts beat faster and blood pressure rise, affecting the nervous system and causing acute stress. This process then sped up the depletion of stem cells that produced melanin in hair follicles. In another experiment, the researchers found they could block the changes by giving the mice an anti-hypertensive, which treats high blood pressure. And by comparing the genes of mice in pain with other mice, they could identify the protein involved in causing damage to stem cells from stress. When this protein -- cyclin-dependent kinase (CDK) -- was suppressed, the treatment also prevented a change in the color of their fur. This leaves the door open for scientists to help delay the onset of grey hair by targeting CDK with a drug.

An anonymous reader shares a summary from Howard Hughes Medical Institute: In a darkened room in Ashburn, Virginia, rows of scientists sit at computer screens displaying vivid 3-D shapes. With a click of a mouse, they spin each shape to examine it from all sides. The scientists are working inside a concrete building at the Howard Hughes Medical Institute's Janelia Research Campus, just off a street called Helix Drive. But their minds are somewhere else entirely -- inside the brain of a fly. Each shape on the scientists' screens represents part of a fruit fly neuron. These researchers and others at Janelia are tackling a goal that once seemed out of reach: outlining each of the fly brain's roughly 100,000 neurons and pinpointing the millions of places they connect. Such a wiring diagram, or connectome, reveals the complete circuitry of different brain areas and how they're linked. The work could help unlock networks involved in memory formation, for example, or neural pathways that underlie movements.

Gerry Rubin, vice president of HHMI and executive director of Janelia, has championed this project for more than a decade. It's a necessary step in understanding how the brain works, he says. When the project began, Rubin estimated that with available methods, tracing the connections between every fly neuron by hand would take 250 people working for two decades -- what he refers to as "a 5,000 person-year problem." Now, a stream of advances in imaging technology and deep-learning algorithms have yanked the dream of a fly connectome out of the clouds and into the realm of probability. High-powered customized microscopes, a team of dedicated neural proofreaders and data analysts, and a partnership with Google have sped up the process by orders of magnitude. Today, a team of Janelia researchers reports hitting a critical milestone: they've traced the path of every neuron in a portion of the female fruit fly brain they've dubbed the "hemibrain." The map encompasses 25,000 neurons -- roughly a third of the fly brain, by volume -- but its impact is outsized. It includes regions of keen interest to scientists -- those that control functions like learning, memory, smell, and navigation. With more than 20 million neural connections pinpointed so far, it's the biggest and most detailed map of the fly brain ever completed. The scientists have published a pre-print paper describing their work, and have made the data they collected available to view and download.

We make decisions by weighing pros and cons. Artificial intelligence has the potential to help us with that by sifting through ever-increasing mounds of data. But to be truly useful, it needs to reason more like a human. An artificial intelligence technique known as argument mining could help. From a report: IBM has just taken a big step in that direction. The company's Project Debater team has spent several years developing an AI that can build arguments. Last year IBM demonstrated its work-in-progress technology in a live debate against a world-champion human debater, the equivalent of Watson's Jeopardy! showdown. Such stunts are fun, and it provided a proof of concept. Now IBM is turning its toy into a genuinely useful tool. The version of Project Debater used in the live debates included the seeds of the latest system, such as the capability to search hundreds of millions of new articles. But in the months since, the team has extensively tweaked the neural networks it uses, improving the quality of the evidence the system can unearth. One important addition is BERT, a neural network Google built for natural-language processing, which can answer queries. The work will be presented at the Association for the Advancement of Artificial Intelligence conference in New York next month.

To train their AI, lead researcher Noam Slonim and his colleagues at IBM Research in Haifa, Israel, drew on 400 million documents taken from the LexisNexis database of newspaper and journal articles. This gave them some 10 billion sentences, a natural-language corpus around 50 times larger than Wikipedia. They paired this vast evidence pool with claims about several hundred different topics, such as "Blood donation should be mandatory" or "We should abandon Valentine's Day." They then asked crowd workers on the Figure Eight platform to label sentences according to whether or not they provided evidence for or against particular claims. The labeled data was fed to a supervised learning algorithm.

China's Five-hundred meter Aperture Spherical Telescope (FAST) -- the world's largest and most sensitive radio telescope -- is now operational. The telescope was completed in 2016 -- 20 years after it was first proposed -- but has been undergoing testing and commissioning since then. Phys.Org reports: FAST's nickname is Tianyan, which means "Eye of the Sky" or "Eye of Heaven." It's built in a natural depression in Guizhou, Southwest China. Among other things, its scientific goals are to study pulsars. FAST has already discovered two of those, in August 2017. FAST's name is not exactly accurate. Though it does have a 500 meter diameter, only 300 meters of it is used at any one time. The telescope is active and can change so that one 300 meter segment is focused on the receiver.

According to Xinhua, all the technical indicators from FAST are meeting or exceeding planned levels. FAST is a powerful radio telescope, and scientists think it will make some major discoveries, especially in its first couple of years. FAST will also perform two sky surveys, which will take about five years. It'll take another 10 years just to analyze all that data. Still, there's room for flexibility in the telescope's operational schedule to pursue any surprises that come up. The surveys will take up about half of the telescope's observing time, leaving room for objectives like searching for exoplanets with magnetic fields, which are probably crucial for life.

The first case of an infection with a new coronavirus has been discovered in the United States. NPR reports: A man from Washington state returned home after a trip to Wuhan, China, on Jan. 15, sought medical attention on Jan. 19 and now is in isolation at Providence Regional Medical Center in Everett, Wash. State health officials say his condition is quite good and even referred to him as "healthy." But testing from the Centers for Disease Control and Prevention on the 20th confirm that he is infected with the Wuhan coronavirus. The man arrived back in the U.S. prior to the implementation of screening at three domestic airports on Friday. As of yesterday, over 200 cases of the virus have been reported in China.

New research from the University of Rochester provides evidence that the magnetic field that first formed around Earth was even stronger than scientists previously believed. The research, published in the journal PNAS, will help scientists draw conclusions about the sustainability of Earth's magnetic shield and whether or not there are other planets in the solar system with the conditions necessary to harbor life. Phys.Org reports: Using new paleomagnetic, electron microscope, geochemical, and paleointensity data, the researchers dated and analyzed zircon crystals -- the oldest known terrestrial materials -- collected from sites in Australia. The zircons, which are about two-tenths of a millimeter, contain even smaller magnetic particles that lock in the magnetization of the earth at the time the zircons were formed. Previous research by [John Tarduno, Professor of Earth and Environmental Sciences and Dean of Research for Arts, Sciences, and Engineering at Rochester] found that Earth's magnetic field is at least 4.2 billion years old and has existed for nearly as long as the planet. Earth's inner core, on the other hand, is a relatively recent addition: it formed only about 565 million years ago, according to research published by Tarduno and his colleagues earlier this year.

While the researchers initially believed Earth's early magnetic field had a weak intensity, the new zircon data suggests a stronger field. But, because the inner core had not yet formed, the strong field that originally developed 4 billion years ago must have been powered by a different mechanism. "We think that mechanism is chemical precipitation of magnesium oxide within Earth," Tarduno says. The magnesium oxide was likely dissolved by extreme heat related to the giant impact that formed Earth's moon. As the inside of Earth cooled, magnesium oxide could precipitate out, driving convection and the geodynamo. The researchers believe inner Earth eventually exhausted the magnesium oxide source to the point that the magnetic field almost completely collapsed 565 million years ago. But the formation of the inner core provided a new source to power the geodynamo and the planetary magnetic shield Earth has today.

An anonymous reader quotes a report from the BBC: The Cardiff University team discovered a method of killing prostate, breast, lung and other cancers in lab tests. The findings, published in Nature Immunology, have not been tested in patients, but the researchers say they have "enormous potential." Our immune system is our body's natural defense against infection, but it also attacks cancerous cells. The scientists were looking for "unconventional" and previously undiscovered ways the immune system naturally attacks tumors. What they found was a T-cell inside people's blood. This is an immune cell that can scan the body to assess whether there is a threat that needs to be eliminated. The difference is this one could attack a wide range of cancers.

T-cells have "receptors" on their surface that allow them to "see" at a chemical level. The Cardiff team discovered a T-cell and its receptor that could find and kill a wide range of cancerous cells in the lab including lung, skin, blood, colon, breast, bone, prostate, ovarian, kidney and cervical cancer cells. Crucially, it left normal tissues untouched. Exactly how it does this is still being explored. This particular T-cell receptor interacts with a molecule called MR1, which is on the surface of every cell in the human body. It is thought MR1 is flagging the distorted metabolism going on inside a cancerous cell to the immune system. Treatment would include extracting T-cells from a blood sample of a cancer patient and then genetically modifying them so they were reprogrammed to make the cancer-finding receptor. The upgraded cells would be grown in vast quantities in the lab and then put back into the patient.

After developing the world's fastest camera a little over a year ago, Caltech's Lihong Wang decided that wasn't good enough and started working on an even faster device. A new paper published in the journal Science Advances details a new camera from Wang that can take up to 1 trillion pictures per second of transparent objects. Phys.Org reports: The camera technology, which Wang calls phase-sensitive compressed ultrafast photography (pCUP), can take video not just of transparent objects but also of more ephemeral things like shockwaves and possibly even of the signals that travel through neurons. Wang explains that his new imaging system combines the high-speed photography system he previously developed with an old technology, phase-contrast microscopy, that was designed to allow better imaging of objects that are mostly transparent such as cells, which are mostly water.

The fast-imaging portion of the system consists of something Wang calls lossless encoding compressed ultrafast technology (LLE-CUP). Unlike most other ultrafast video-imaging technologies that take a series of images in succession while repeating the events, the LLE-CUP system takes a single shot, capturing all the motion that occurs during the time that shot takes to complete. Since it is much quicker to take a single shot than multiple shots, LLE-CUP is capable of capturing motion, such as the movement of light itself, that is far too fast to be imaged by more typical camera technology. In the new paper, Wang and his fellow researchers demonstrate the capabilities of pCUP by imaging the spread of a shockwave through water and of a laser pulse traveling through a piece of crystalline material.

An anonymous reader quotes a report from The Wall Street Journal: Hospitals have granted Microsoft, IBM and Amazon the ability to access identifiable patient information under deals to crunch millions of health records, the latest examples of hospitals' growing influence in the data economy. This breadth of access wasn't always spelled out by hospitals and tech giants when the deals were struck. The scope of data sharing in these and other recently reported agreements reveals a powerful new role that hospitals play -- as brokers to technology companies racing into the $3 trillion health-care sector. Rapid digitization of health records in recent years and privacy laws enabling companies to swap patient data have positioned hospitals as a primary arbiter of how such sensitive data is shared.

Microsoft and Providence, a Renton, Wash., hospital system with data for about 20 million patient visits a year, are developing cancer algorithms by using doctor's notes in patient medical records. The notes haven't been stripped of personally identifiable information, according to Providence. And an agreement between IBM and Brigham and Women's Hospital, in Boston, to jointly develop artificial intelligence allows the hospital to share personally identifiable data for specific requests, people involved in the agreement said -- though so far the hospital hasn't done so and has no current plans to do so, according to hospital and IBM officials. Microsoft executive Peter Lee in July described how his company would use Providence patient data without identifying information for algorithm development. In a December statement, he said patients' personal health data remains in Providence's control and declined to comment further. As for Amazon, the Fred Hutchinson Cancer Research Center, in Seattle, granted certain AWS employees access to health information that identifies individual patients. "The Hutch, a research institution with ties to hospitals, trained and tested Amazon Web Services software designed to read medical notes," the report says. "An AWS spokeswoman said it doesn't use personally identifiable data protected under federal privacy laws to develop or improve its services."

The outbreak of a new virus that began in the Chinese city of Wuhan last month appears to be far from over. Today, Chinese health authorities reported that over 130 new pneumonia cases caused by the virus were identified over the weekend, bringing the total in China alone to 201, including three outside Wuhan. From a report: There has also been a third death from the infection, and South Korea now has reported a case as well -- the third country outside China to do so. Meanwhile, the pattern of spread makes it increasingly unlikely that the virus does not transmit between people, some experts say. "Uncertainty and gaps remain, but it's clear that there is some level of person-to-person transmission," Jeremy Farrar, head of the Wellcome Trust in London, said in a statement today. "The sudden spike in cases is disconcerting, but not entirely unexpected," says Adam Kamradt-Scott, an infectious diseases specialist at the University of Sydney. As more people learn about the disease, more will go to doctors, Kamradt-Scott says, even with mild symptoms, whereas previously they might have just stayed home. And doctors are now on the lookout for the new disease. "The result is that you see a sudden surge in cases," he says. But âoeif we continue to see this trend continue over the next week where there are 50 to 100 new cases every day, then that would be cause for further concern."

Might it be possible that someday in the near future, an official might get you to dance around a bit, in order to confirm that you're really you? Perhaps not, but nonetheless, a recent study has determined that people's identities can be matched to their unique style of dancing. From a report: Scientists at Finland's University of Jyvaskyla started out by using motion capture technology to see if test subjects' psychological traits could be ascertained from the way in which they danced -- such traits included their mood, their level of empathy, and how extroverted or neurotic they were. The researchers were also interested in seeing if simply by watching a person dance, it would be possible to determine what sort of music they were dancing to. This only worked about 30 percent of the time. What they unintentionally discovered, however, was that regardless of the type of music, each person has a characteristic style of dancing that can be identified and matched specifically to themselves. Doing so is accomplished utilizing machine learning algorithms, in conjunction with the motion capture tech. In the study, a total of 73 volunteers each danced to eight genres of music â" these included Blues, Country, Dance/Electronica, Jazz, Metal, Pop, Reggae and Rap. The participants received no instructions, other than to "move any way that felt natural."

"European researchers are working on a system that can churn out breathable oxygen from simulated samples of moon dust," reports Gizmodo:
"Being able to acquire oxygen from resources found on the Moon would obviously be hugely useful for future lunar settlers, both for breathing and in the local production of rocket fuel," explained Beth Lomax, a chemist from the University of Glasgow, in an European Space Agency (ESA) press release. Lomax, along with ESA research fellow Alexandre Meurisse, are currently plugging away at a prototype that could eventually lead to exactly that: oxygen production from lunar dust. They're currently testing their system at the Materials and Electrical Components Laboratory of the European Space Research and Technology Centre (ESTEC), which is based in Noordwijk, the Netherlands.

Their prototype is working, but adjustments will be required to make it suitable for use on the Moon, such as reducing its operating temperature....

Interestingly, ESTEC is not treating the metals as an unwanted byproduct. The team is currently looking into various ways of exploiting these metals in a lunar environment, such as transforming them into compounds for 3D printing.
The European Space Agency points out that samples returned from the lunar surface were made up of 40-45% percent oxygen by weight.

Engadget reports on a new tech-industry-in-space comedy premiering tonight on HBO:
If you thought that HBO was done mocking technology companies now that Silicon Valley is done, think again. Avenue 5 is the channel's new sitcom, and one that asks the question: "What if tech bros were in charge of more than just our internet histories?'" The answer, at least according to the first half of the season, is that it won't be pretty -- or safe...

The Avenue 5 is a large space liner that, in the words of cinematographer Eben Bolter, is designed after a vulgar space hotel that goes too far and "gets the details wrong". This Titanic-like vessel and its 5,000 passengers are on a routine jaunt through the solar system when a minor disaster strikes, and its course is altered. But this is space, where a small deviation changes the flight time from eight weeks to several years.

The ship is owned by Herman Judd (Josh Gad) of the Judd Corporation, a self-regarding business magnate who, in Bolter's mind, has "only ever had one good idea." He's not quite an analog for the Bezoses and Musks you may be thinking of, but more a cracked-mirror version of both. Throughout the show, he attempts to impose his thinking on the crisis as if he was still in California, or wherever Silicon Valley moved after the show's alluded-to Huawei Wars. Early on, Judd is presented with the intractable problem of space physics, and he hopes to fix things as he did on Earth. He says, in the Jobsian tradition, that you can make something happen by making someone say that it can. The fight between visionary optimism and reality is harder when you're surrounded by an infinite vacuum, after all.

Avenue 5's point seems to be that you can't simply blue-sky your way out of a crisis when reality keeps getting in the way.

Two months ago the Washington Post reported that Google "has partnered with health-care provider Ascension to collect and store personal data for millions of patients, including full names, dates of birth and clinical histories, in order to make smarter recommendations to physicians."

Now CNBC reports that the medical record vendor Epic Systems "has been phoning customers to tell them it will not pursue further integration with Google Cloud. The company is instead focusing on Amazon AWS and Microsoft Azure, citing insufficient interest from customers in Google.

"The move comes as Google is facing criticism from privacy advocates about its work with Ascension, one of the largest U.S. health systems," CNBC adds. But could this start influencing which cloud provider hospitals choose for their records?

"We've historically seen hospital systems make these decisions independently of their medical record provider," said Aneesh Chopra, the president of health-technology company CareJourney and the former chief technology officer of the United States. "It will be interesting to see if Epic's thumb on the scale moves cloud market share...."

Epic isn't alone in its move.

The Wall Street Journal recently reported that Cerner decided against pursuing a data-storage relationship with Google despite being offered tens of millions of dollars in incentives. The company was on the hunt for a cloud vendor to help it store 250 million patient medical records. In the end, Cerner went with Amazon.
In 2017 CNBC reported Cerner's collaboration with Amazon would initially focus on a "popular health product...which enables hospitals to gather and analyze huge volumes of clinical data to improve patients' health outcomes and lower treatment costs."

"Last month, astronomers noticed Betelgeuse had faded much more than usual," writes the Steamboat Pilot & Today. "This is the faintest it has been in over a century of observations.

"What's up with Betelgeuse?"

The Grim Reefer quotes their report:
Well, probably nothing. It is most likely experiencing a super minimum as two of its variability cycles sync up and reach minimum brightness at the same time. On the other hand, stars like Betelgeuse are well advanced in age and are destined to explode as supernovas at the end of their lives.

There hasn't been a bright supernova in our Milky Way galaxy since the supernova of 1604. Sure, we've seen supernovas in galaxies far, far away, but none have been seen close to home. So, there is some excitement in the astronomical community, warranted or not, that Betelgeuse might be ready to pop. Astrophysicists theorize that a pronounced fading might presage an impending supernova explosion. If Betelgeuse continues to fade into January and beyond, then look out -- the end might be near.
A Forbes science writer remains skeptical, but points out that "some astronomers now think there's a much closer star that could 'nova'..."
A star called V Sagittae, 7,800 light years distant in the tiny constellation of Sagitta (just below Cygnus in the famous "Summer Triangle" asterism of stars) is barely visible even in mid-sized telescopes, but new research suggests that it could explode around the year 2083... "Around the year 2083, its accretion rate will rise catastrophically, spilling mass at incredibly high rates onto the white dwarf, with this material blazing away," says Professor Emeritus Bradley E. Schaefer, LSU Department of Physics & Astronomy. "In the final days of this death-spiral, all of the mass from the companion star will fall onto the white dwarf, creating a super-massive wind from the merging star, appearing as bright as Sirius, possibly even as bright as Venus."

The uncertainty of the prediction is plus or minus 16 years, so it could happen between 2067 and 2099, most likely near the middle of this range. It promises to be a wonderful sight.

"SpaceX is setting out to prove a critical safety system will be able to save astronaut lives in the event of a launch emergency during ascent," reports CNET: The Crew Dragon in-flight abort test...is a required step before NASA will allow astronauts to fly to the International Space Station in the SpaceX capsule as part of the Commercial Crew Program.

[UPDATE: Though they'd originally planned to launch Saturday, SpaceX tweeted early Saturday morning that "due to sustained winds and rough seas in the recovery area" they're now targeting Sunday, January 19, "with a six-hour test window opening at 8:00 a.m. EST, 13:00 UTC." Watch SpaceX's livestream here.]

NASA will also livestream the event... Backup test opportunities are set for Sunday or Monday if Saturday doesn't work out.

Crew Dragon will take a ride on a Falcon 9 rocket, which won't survive the test. The launch will take place at Florida's Kennedy Space Center, which will allow the rocket to break up over the Atlantic Ocean. It could be quite an eye-opening experience. SpaceX shared an animated video showing how the test is expected to go. If all goes well, the Crew Dragon capsule will separate from the rocket, deploy parachutes and float gently down to the water....

SpaceX successfully sent an uncrewed Crew Dragon to the International Space Station in early 2019. The ultimate goal is to make a return trip with NASA astronauts on board. If the in-flight abort test works out, then the first launch of humans from U.S. soil since the end of the space shuttle era should finally happen in 2020.

Slashdot reader JoshuaZ writes:
In a major breakthrough in quantum computing it was shown that MIP* equals RE. MIP* is the set of problems that can be efficiently demonstrated to a classical computer interacting with multiple quantum computers with any amount of shared entanglement between the quantum computers. RE is the set of problems which are recursive; this is essentially all problems which can be computed.

This result comes through years of deep development of understanding interactive protocols, where one entity, a verifier, has much less computing power than another set of entities, provers, who wish to convince the verifier of the truth of a claim. In 1990, a major result was that a classical computer with a polynomial amount of time could be convince of any claim in PSPACE by interacting with an arbitrarily powerful classical computer. Here PSPACE is the set of problems solvable by a classical computer with a polynomial amount of space. Subsequent results showed that if one allowed a verifier able to interact with multiple provers, the verifier could be convinced of a solution of any problem in NEXPTIME, a class conjectured to be much larger than PSPACE. For a while, it was believed that in the quantum case, the set of problems might actually be smaller, since multiple quantum computers might be able to use their shared entangled qubits to "cheat" the verifier. However, this has turned out not just to not be the case, but the exact opposite: MIP* is not only large, it is about as large as a computable class can naturally be.

This result while a very big deal from a theoretical standpoint is unlikely to have any immediate applications since it supposes quantum computers with arbitrarily large amounts of computational power and infinite amounts of entanglement.

The paper in question is a 165 tour de force which includes incidentally showing that the The Connes embedding conjecture, a 50 year old major conjecture from the theory of operator algebras, is false.

schwit1 shares a report from The Wall Street Journal: Nearly 150 years ago, [German physician Carl Reinhold August Wunderlich] analyzed a million temperatures from 25,000 patients and concluded that normal human-body temperature is 98.6 degrees Fahrenheit. In a new study, researchers from Stanford University argue that Wunderlich's number was correct at the time but is no longer accurate because the human body has changed. Today, they say, the average normal human-body temperature is closer to 97.5 degrees Fahrenheit (Warning: source paywalled; alternative source).

To test their hypothesis that today's normal body temperature is lower than in the past, Dr. Parsonnet and her research partners analyzed 677,423 temperatures collected from 189,338 individuals over a span of 157 years. The readings were recorded in the pension records of Civil War veterans from the start of the war through 1940; in the National Health and Nutrition Examination Survey I conducted by the U.S. Centers for Disease Control and Prevention from 1971 through 1974; and in the Stanford Translational Research Integrated Database Environment from 2007 through 2017. Overall, temperatures of the Civil War veterans were higher than measurements taken in the 1970s, and, in turn, those measurements were higher than those collected in the 2000s. The study has been published in the journal eLife.

An anonymous reader quotes a report from MIT Technology Review: In a paper published in Nature today, DeepMind, Alphabet's AI subsidiary, has once again used lessons from reinforcement learning to propose a new theory about the reward mechanisms within our brains. The hypothesis, supported by initial experimental findings, could not only improve our understanding of mental health and motivation. It could also validate the current direction of AI research toward building more human-like general intelligence. At a high level, reinforcement learning follows the insight derived from Pavlov's dogs: it's possible to teach an agent to master complex, novel tasks through only positive and negative feedback. An algorithm begins learning an assigned task by randomly predicting which action might earn it a reward. It then takes the action, observes the real reward, and adjusts its prediction based on the margin of error. Over millions or even billions of trials, the algorithm's prediction errors converge to zero, at which point it knows precisely which actions to take to maximize its reward and so complete its task.

It turns out the brain's reward system works in much the same way -- a discovery made in the 1990s, inspired by reinforcement-learning algorithms. When a human or animal is about to perform an action, its dopamine neurons make a prediction about the expected reward. Once the actual reward is received, they then fire off an amount of dopamine that corresponds to the prediction error. A better reward than expected triggers a strong dopamine release, while a worse reward than expected suppresses the chemical's production. The dopamine, in other words, serves as a correction signal, telling the neurons to adjust their predictions until they converge to reality. The phenomenon, known as reward prediction error, works much like a reinforcement-learning algorithm. The improved algorithm changes the way it predicts rewards. "Whereas the old approach estimated rewards as a single number -- meant to equal the average expected outcome -- the new approach represents them more accurately as a distribution," the report says. This lends itself to a new hypothesis: Do dopamine neurons also predict rewards in the same distributional way?

After testing this theory, DeepMind found "compelling evidence that the brain indeed uses distributional reward predictions to strengthen its learning algorithm," reports MIT Technology Review.