Vue normale

Hier — 11 avril 2025Flux principal

Germany To Create 'Super-High-Tech Ministry' For Research, Technology and Aerospace

Par : msmash
11 avril 2025 à 21:30
Germany will get a new "super-high-tech ministry" responsible for research, technology, and aerospace, according to the coalition agreement published by the incoming government this week. From a report: The announcement is one of several nods to science in the 144-page agreement, unveiled on 9 April following weeks of negotiations between the center-right Christian Democrats (CDU) and its sister party, the Christian Social Union in Bavaria (CSU) -- who together won the most seats in February's federal elections -- and the center-left Social Democrats. The agreement is expected to be formally approved by the three parties by early May, paving the way for CDU leader Friedrich Merz to be elected chancellor. [...] The new agreement lists a number of scientific priorities for the new government, including support for artificial intelligence, quantum technologies, biotechnology, microchip development and production, and fusion energy. "Our goal is that the world's first fusion reactor should be realized in Germany," the text states. It also mentions personalized medicine, oceans research, and sustainability research as "strategic" areas. But the agreement does not include any budget estimates, and observers caution it is unclear where the money for new programs would come from. The agreement does affirm current commitments to increase the budgets of the country's main research organizations by 3% per year through 2030.

Read more of this story at Slashdot.

FDA Plans To Phase Out Animal Testing Requirements

Par : msmash
11 avril 2025 à 07:00
The Food and Drug Administration says it would begin phasing out animal testing requirements for antibody therapies and other drugs and move toward AI-based models and other tools it deems "human-relevant." Axios: The FDA said it would launch a pilot program over the next year allowing select developers of monoclonal antibodies to use a primarily non-animal-based testing strategy. Commissioner Marty Makary in a statement said the shift would improve drug safety, lower research and development costs and address ethical concerns about animal experimentation.

Read more of this story at Slashdot.

À partir d’avant-hierFlux principal

No, the Dire Wolf Has Not Been Brought Back From Extinction

Par : msmash
8 avril 2025 à 04:00
Colossal Biosciences has claimed it "successfully restored" the extinct dire wolf after a "10,000+ year absence," but scientists clarify these are actually genetically modified grey wolves. The U.S. company announced three pups -- males Remus and Romulus born in October, and female Khaleesi born in January -- as dire wolves, but made only 20 genetic edits to grey wolves. Beth Shapiro of Colossal told New Scientist that just 15 modifications were based on dire wolf DNA, primarily targeting size, musculature and ear shape. Five other changes involve mutations known to produce light coats in grey wolves. A 2021 DNA study revealed dire wolves and grey wolves last shared a common ancestor about 6 million years ago, with jackals and African wild dogs more closely related to grey wolves.

Read more of this story at Slashdot.

Bonobos May Combine Words In Ways Previously Thought Unique To Humans

Par : BeauHD
5 avril 2025 à 10:00
A new study shows bonobos can combine vocal calls in ways that mirror human language, producing phrases with meanings beyond the sum of individual sounds. "Human language is not as unique as we thought," said Dr Melissa Berthet, the first author of the research from the University of Zurich. Another author, Dr Simon Townsend, said: "The cognitive building blocks that facilitate this capacity is at least 7m years old. And I think that is a really cool finding." The Guardian reports: Writing in the journal Science, Berthet and colleagues said that in the human language, words were often combined to produce phrases that either had a meaning that was simply the sum of its parts, or a meaning that was related to, but differed from, those of the constituent words. "'Blond dancer' -- it's a person that is both blond and a dancer, you just have to add the meanings. But a 'bad dancer' is not a person that is bad and a dancer," said Berthet. "So bad is really modifying the meaning of dancer here." It was previously thought animals such as birds and chimpanzees were only able to produce the former type of combination, but scientists have found bonobos can create both. The team recorded 700 vocalizations from 30 adult bonobos in the Democratic Republic of the Congo, checking the context of each against a list of 300 possible situations or descriptions. The results reveal bonobos have seven different types of call, used in 19 different combinations. Of these, 15 require further analysis, but four appear to follow the rules of human sentences. Yelps -- thought to mean "'et's do that" -- followed by grunts -- thought to mean "look at what I am doing," were combined to make "yelp-grunt," which appeared to mean "let's do what I'm doing." The combination, the team said, reflected the sum of its parts and was used by bonobos to encourage others to build their night nests. The other three combinations had a meaning apparently related to, but different from, their constituent calls. For example, the team found a peep -- which roughly means "I would like to ..." -- followed by a whistle -- appeared to mean "let's stay together" -- could be combined to create "peep-whistle." This combination was used to smooth over tense social situations, such as during mating or displays of prowess. The team speculated its meaning was akin to "let's find peace." The team said the findings in bonobos, together with the previous work in chimps, had implications for the evolution of language in humans, given all three species showed the ability to combine words or vocalizations to create phrases.

Read more of this story at Slashdot.

Scientists Warn Indonesia's Rice Megaproject Faces Failure

Par : msmash
5 avril 2025 à 05:00
Indonesian President Prabowo Subianto's ambitious plan to create 1 million hectares of new rice farms in eastern Merauke Regency faces strong criticism from scientists who have warned it will fail due to unsuitable soils and climate. Military "food brigades" are currently guarding bulldozers clearing swampy forests in Indonesian New Guinea for the project, which aims to boost food self-sufficiency for the nation's 281 million people. Soil scientists warn that Merauke's conditions could lead to acidic soils unable to support economically viable rice farming, potentially resulting in abandoned fields vulnerable to wildfires. "Farmers will get no profit at all," said Dwi Andreas, a soil scientist at Bogor Agricultural University who tested 12 rice varieties in similar soils with poor results. The initiative mirrors past failed megaprojects, including a 1990s attempt to convert 1 million hectares of Borneo peatlands to rice paddies and a 2020 onion and potato farming expansion in North Sumatra that saw 90% of fields abandoned. A previous 2010 attempt to expand rice farming in Merauke also failed, destroying forests that Indigenous Papuans relied on and increasing childhood malnutrition, according to anthropologist Laksmi Adriani.

Read more of this story at Slashdot.

Publishers Trial Paying Peer Reviewers - What Did They Find?

Par : msmash
31 mars 2025 à 16:02
Two scientific journals that experimented with paying peer reviewers found the practice sped up the review process without compromising quality, according to findings published this month. Critical Care Medicine offered $250 to half of 715 invited reviewers, with 53% accepting compared to 48% of unpaid reviewers. Paid reviews were completed one day faster on average. In a more dramatic result, Biology Open saw reviews completed in 4.6 business days when paying reviewers $284 per review, versus 38 days for unpaid reviews. "For the editors it has been extremely helpful because, prior to this, in some areas it was very difficult to secure reviewers," said Alejandra Clark, managing editor of Biology Open.

Read more of this story at Slashdot.

Did Life on Earth Come from 'Microlightning' Between Charged Water Droplets?

Par : EditorDavid
30 mars 2025 à 01:34
Some scientists believe life on earth originated in organic matter in earth's bodies of water more than 3.5 billion years ago," reports CNN. "But where did that organic material come from...?" Maybe electrical energy sparked the beginnings of life on earth — just like in Frankenstein: Researchers decades ago proposed that lightning caused chemical reactions in ancient Earth's oceans and spontaneously produced the organic molecules. Now, new research published March 14 in the journal Science Advances suggests that fizzes of barely visible "microlightning," generated between charged droplets of water mist, could have been potent enough to cook up amino acids from inorganic material. Amino acids — organic molecules that combine to form proteins — are life's most basic building blocks and would have been the first step toward the evolution of life... For animo acids to form, they need nitrogen atoms that can bond with carbon. Freeing up atoms from nitrogen gas requires severing powerful molecular bonds and takes an enormous amount of energy, according to astrobiologist and geobiologist Dr. Amy J. Williams [an associate professor in the department of geosciences at the University of Florida who was not involved in the research]. "Lightning, or in this case, microlightning, has the energy to break molecular bonds and therefore facilitate the generation of new molecules that are critical to the origin of life on Earth," Williams told CNN in an email... For the new study, scientists revisited the 1953 experiments but directed their attention toward electrical activity on a smaller scale, said senior study author Dr. Richard Zare, the Marguerite Blake Wilbur Professor of Natural Science and professor of chemistry at Stanford University in California. Zare and his colleagues looked at electricity exchange between charged water droplets measuring between 1 micron and 20 microns in diameter. (The width of a human hair is 100 microns....) The researchers mixed ammonia, carbon dioxide, methane and nitrogen in a glass bulb, then sprayed the gases with water mist, using a high-speed camera to capture faint flashes of microlightning in the vapor. When they examined the bulb's contents, they found organic molecules with carbon-nitrogen bonds. These included the amino acid glycine and uracil, a nucleotide base in RNA... "What we have done, for the first time, is we have seen that little droplets, when they're formed from water, actually emit light and get this spark," Zare said. "That's new. And that spark causes all types of chemical transformations...." Even on a volatile Earth billions of years ago, lightning may have been too infrequent to produce amino acids in quantities sufficient for life — a fact that has cast doubt on such theories in the past, Zare said. Water spray, however, would have been more common than lightning. A more likely scenario is that mist-generated microlightning constantly zapped amino acids into existence from pools and puddles, where the molecules could accumulate and form more complex molecules, eventually leading to the evolution of life. "We propose," Zare told CNN, "that this is a new mechanism for the prebiotic synthesis of molecules that constitute the building blocks of life."

Read more of this story at Slashdot.

Giant, Fungus-Like Organism May Be Completely Unknown Branch of Life

Par : BeauHD
29 mars 2025 à 01:10
New research suggests that Prototaxites, once believed to be a giant fungus, may actually represent an entirely extinct and previously unknown branch of complex life, distinct from fungi, plants, animals, and protists. Live Science reports: The researchers studied the fossilized remains of one Prototaxites species named Prototaxites taiti, found preserved in the Rhynie chert, a sedimentary deposit of exceptionally well-preserved fossils of early land plants and animals in Scotland. This species was much smaller than many other species of Prototaxites, only growing up to a few inches tall, but it is still the largest Prototaxites specimen found in this region. Upon examining the internal structure of the fossilized Prototaxites, the researchers found that its interior was made up of a series of tubes, similar to those within a fungus. But these tubes branched off and reconnected in ways very unlike those seen in modern fungi. "We report that Prototaxites taiti was the largest organism in the Rhynie ecosystem and its anatomy was fundamentally distinct from all known extant or extinct fungi," the researchers wrote in the paper. "We therefore conclude that Prototaxites was not a fungus, and instead propose it is best assigned to a now entirely extinct terrestrial lineage." True fungi from the same period have also been preserved in the Rhynie chert, enabling the researchers to chemically compare them to Prototaxites. In addition to their unique structural characteristics, the team found that the Prototaxites fossils left completely different chemical signatures to the fungi fossils, indicating that the Prototaxites did not contain chitin, a major building block of fungal cell walls and a hallmark of the fungal kingdom. The Prototaxites fossils instead appeared to contain chemicals similar to lignin, which is found in the wood and bark of plants. "We conclude that the morphology and molecular fingerprint of P. taiti is clearly distinct from that of the fungi and other organism preserved alongside it in the Rhynie chert, and we suggest that it is best considered a member of a previously undescribed, entirely extinct group of eukaryotes," the researchers wrote. The research has been published on the preprint server bioRxiv.

Read more of this story at Slashdot.

A New Image File Format Efficiently Stores Invisible Light Data

Par : BeauHD
28 mars 2025 à 23:50
An anonymous reader quotes a report from Ars Technica: Imagine working with special cameras that capture light your eyes can't even see -- ultraviolet rays that cause sunburn, infrared heat signatures that reveal hidden writing, or specific wavelengths that plants use for photosynthesis. Or perhaps using a special camera designed to distinguish the subtle visible differences that make paint colors appear just right under specific lighting. Scientists and engineers do this every day, and they're drowning in the resulting data. A new compression format called Spectral JPEG XL might finally solve this growing problem in scientific visualization and computer graphics. Researchers Alban Fichet and Christoph Peters of Intel Corporation detailed the format in a recent paper published in the Journal of Computer Graphics Techniques (JCGT). It tackles a serious bottleneck for industries working with these specialized images. These spectral files can contain 30, 100, or more data points per pixel, causing file sizes to balloon into multi-gigabyte territory -- making them unwieldy to store and analyze. [...] The current standard format for storing this kind of data, OpenEXR, wasn't designed with these massive spectral requirements in mind. Even with built-in lossless compression methods like ZIP, the files remain unwieldy for practical work as these methods struggle with the large number of spectral channels. Spectral JPEG XL utilizes a technique used with human-visible images, a math trick called a discrete cosine transform (DCT), to make these massive files smaller. Instead of storing the exact light intensity at every single wavelength (which creates huge files), it transforms this information into a different form. [...] According to the researchers, the massive file sizes of spectral images have reportedly been a real barrier to adoption in industries that would benefit from their accuracy. Smaller files mean faster transfer times, reduced storage costs, and the ability to work with these images more interactively without specialized hardware. The results reported by the researchers seem impressive -- with their technique, spectral image files shrink by 10 to 60 times compared to standard OpenEXR lossless compression, bringing them down to sizes comparable to regular high-quality photos. They also preserve key OpenEXR features like metadata and high dynamic range support. The report notes that broader adoption "hinges on the continued development and refinement of the software tools that handle JPEG XL encoding and decoding." Some scientific applications may also see JPEG XL's lossy approach as a drawback. "Some researchers working with spectral data might readily accept the trade-off for the practical benefits of smaller files and faster processing," reports Ars. "Others handling particularly sensitive measurements might need to seek alternative methods of storage."

Read more of this story at Slashdot.

Scientists Propose 'Bodyoids' To Address Medical Research and Organ Shortage Challenges

Par : msmash
28 mars 2025 à 19:50
Stanford University researchers have proposed creating "bodyoids" -- ethically sourced human bodies grown from stem cells without neural components for consciousness or pain sensation -- to revolutionize medical research and address organ shortages. In a new opinion piece published in MIT Technology Review, scientists Carsten T. Charlesworth, Henry T. Greely, and Hiromitsu Nakauchi argue that recent advances in biotechnology make this concept increasingly plausible. The approach would combine pluripotent stem cells, artificial uterus technology, and genetic techniques to inhibit brain development. The researchers point to persistent shortages of human biological materials as a major bottleneck in medical progress. More than 100,000 patients currently await solid organ transplants in the US alone, while less than 15% of drugs entering clinical trials receive regulatory approval. These lab-grown bodies could potentially generate patient-specific organs that are perfect immunological matches, eliminate the need for lifelong immunosuppression, and provide personalized drug screening models.

Read more of this story at Slashdot.

Inside arXiv - the Most Transformative Platform in All of Science

Par : msmash
27 mars 2025 à 14:55
Paul Ginsparg, a physics professor at Cornell University, created arXiv nearly 35 years ago as a digital repository where researchers could share their findings before peer review. Today, the platform hosts more than 2.6 million papers, receives 20,000 new submissions monthly, and serves 5 million active users, Wired writes in a profile of the platform. "Just when I thought I was out, they pull me back in!" Ginsparg quotes from The Godfather, reflecting his inability to fully hand over the platform despite numerous attempts. If arXiv stopped functioning, scientists worldwide would face immediate disruption. "Everybody in math and physics uses it," says Scott Aaronson, a computer scientist at the University of Texas at Austin. "I scan it every night." ArXiv revolutionized academic publishing, previously dominated by for-profit giants like Elsevier and Springer, by allowing instant and free access to research. Many significant discoveries, including the "transformers" paper that launched the modern AI boom, first appeared on the platform. Initially a collection of shell scripts on Ginsparg's NeXT machine in 1991, arXiv followed him from Los Alamos National Laboratory to Cornell, where it found an institutional home despite administrative challenges. Recent funding from the Simons Foundation has enabled a hiring spree and long-needed technical updates.

Read more of this story at Slashdot.

'What CERN Does Next Matters For Science and For International Cooperation'

Par : msmash
24 mars 2025 à 20:20
CERN faces a pivotal decision about its future as the Large Hadron Collider approaches the end of its usefulness by the early 2040s. Management proposes building the Future Circular Collider (FCC), a machine with a 90-kilometer circumference that would smash particles at eight times the energy of the LHC. This hugely consequential plan faces significant challenges. Much of the required technology doesn't exist yet, including superconducting magnets strong enough to bend high-energy particle beams. The project also lacks the clear rationale that the LHC had in finding the Higgs boson. The proposal has divided physicists. Critics worry about the decades-long timeline, potential cost overruns, and the risk of sacrificing other valuable CERN activities. Germany, which provides 20% of the lab's budget, has already indicated it won't increase contributions. A council-appointed group is now gathering input from the physics community before making recommendations in December. Nature's editorial board adds: Unless some nations step up with a major infusion of cash, the FCC faces an uncertain prospect of being funded. But waiting too long could mean that there will be a large gap between the new facility opening and the closure of the LHC, and precious expertise could end up being lost. Although physicists might disagree on what CERN should do, they nearly unanimously care about the lab's future. They and their leaders must now make the case for why European taxpayers, who fund most of the lab's yearly budget should care, too. The stakes are beyond science, and even beyond Europe.

Read more of this story at Slashdot.

Researchers Search For More Precise Ways To Measure Pain

Par : msmash
24 mars 2025 à 19:04
Scientists are developing biomarkers to objectively measure pain, addressing a fundamental medical challenge that has contributed to the opioid crisis and led to consistent underestimation of pain in women and minorities. Four research teams funded by the Department of Health and Human Services are developing technologies to quantify pain like other vital signs. Their approaches include a blood test for endometriosis pain, a device measuring nerve response through pupil dilation, microneedle patches sampling interstitial fluid, and a wearable sensor detecting pain markers in sweat. "When patients are told that the pain is all in their head, the implication is that it's imagined, but the irony is that's sort of right," said Adam Kepecs, a neuroscience professor at Washington University. "The pain only exists in your brain. It's neural activity, which is why it's invisible and uniquely personal. But it's still real." These innovations could transform treatment for the nearly 25% of Americans suffering from chronic pain, while potentially saving billions in healthcare costs.

Read more of this story at Slashdot.

Have Humans Passed Peak Brain Power?

Par : msmash
17 mars 2025 à 16:00
Across high-income countries, humans' ability to reason and solve problems appears to have peaked in the early 2010s and declined since. Despite no changes in fundamental brain biology, test scores for both teenagers and adults show deteriorating performance in reading, mathematics and science. In an eye-opening statistic, 25% of adults in high-income countries now struggle to "use mathematical reasoning when reviewing statements" -- rising to 35% in the US. This cognitive decline coincides with a fundamental shift in our relationship with information. Americans reading books has fallen below 50%, while difficulty thinking and concentrating among 18-year-olds has climbed sharply since the mid-2010s. The timing points to our changing digital habits: a transition from finite web pages to infinite feeds, from active browsing to passive consumption, and from focused attention to constant context-switching. Research shows that intentional use of digital technologies can be beneficial, but the passive consumption dominating recent years impairs verbal processing, attention, working memory and self-regulation. Some of the cited research in the story: New PIAAC results show declining literacy and increasing inequality in many European countries â" Better adult learning is necessary; Have attention spans been declining?; Short- and long-term effects of passive and active screen time on young children's phonological memory; Efficient, helpful, or distracting? A literature review of media multitasking in relation to academic performance.

Read more of this story at Slashdot.

Man Survives With Titanium Heart For 100 Days - a World First

Par : msmash
13 mars 2025 à 17:03
An Australian man in his forties has become the first person in the world to leave hospital with an artificial heart made of titanium. From a report: The device is used as a stopgap for people with heart failure who are waiting for a donor heart, and previous recipients of this type of artificial heart had remained in US hospitals while it was in place. The man lived with the device for more than three months until he underwent surgery to receive a donated human heart. The man is recovering well, according to a statement from St Vincent's Hospital in Sydney, Australia, where the operations were conducted. The Australian is the sixth person globally to receive the device, known as BiVACOR, but the first to live with it for more than a month.

Read more of this story at Slashdot.

Preprint Sites bioRxiv and medRxiv Launch New Era of Independence

Par : msmash
11 mars 2025 à 22:40
A new chapter has begun for two of the world's most popular preprint platforms, bioRxiv and medRxiv, with the launch of a non-profit organization that will manage them, their co-founders announced today. From a report: The servers allow researchers to share manuscripts for free before peer review and have become an integral part of publishing biology and medical research. Until now, they had been managed by Cold Spring Harbor Laboratory (CSHL) in New York. The new organization, named openRxiv, will have a board of directors and a scientific and medical advisory board. It is supported by a fresh US$16-million grant from the Chan Zuckerberg Initiative (CZI), the projects' main financial backer. "It's just exciting to see this key piece of infrastructure really get the attention that it deserves as a dedicated initiative," says Katie Corker, executive director of ASAPbio, a scientist-driven non-profit organization, which is based in San Francisco, California. Preprints are "the backbone of the scientific publishing ecosystem, maybe especially at the current moment, when there's a lot of worries about who has control of information." The launch of openRxiv "reflects a maturation of the projects," which started as an experiment at CSHL, says Richard Sever, a co-founder of both servers and chief science and strategy officer at openRxiv. It has "become so important that they should have their own organization running them, which is focused on the long-term sustainability of the servers, as opposed to being a side project within a big research institution," says Sever.

Read more of this story at Slashdot.

Are Microplastics Bad For Your Health? More Rigorous Science is Needed

Par : msmash
10 mars 2025 à 18:51
An anonymous reader shares a Nature story: In March last year, researchers found that among a group of nearly 300 participants, people who had higher concentrations of plastics in deposits of fat in their arteries (arterial plaques) were more likely to experience heart attacks or strokes, and more likely to die as a result, than those in whom plastics were not detected. Since it was published, the New England Journal of Medicine study has been mentioned more than 6,600 times on social media and more than 800 times in news articles and blogs. The issue of whether plastics are entering human tissues and what impacts they might have on health is understandably of great interest to scientists, industry and society. Indeed, for the past few years there have been news stories almost every month about peer-reviewed articles that have reported findings of plastic particles in all sorts of human tissues and bodily fluids -- including the lungs, heart, penis, placenta and breast milk. And in multiple countries, policymakers are being urged to implement measures to limit people's exposure to nanoplastics and microplastics. Many of the studies conducted so far, however, rely on small sample sizes (typically 20-50 samples) and lack appropriate controls. Modern laboratories are themselves hotspots of nanoplastic and microplastic pollution, and the approaches that are being used to detect plastics make it hard to rule out the possibility of contamination, or prove definitively that plastics are in a sample. Also, many findings are not biologically plausible based on what is known -- mainly from nanomedicine -- about the movement of tiny particles within the human body. For an emerging area of research, such problems are unsurprising. But without more rigorous standards, transparency and collaboration -- among researchers, policymakers and industrial stakeholders -- a cycle of misinformation and ineffective regulation could undermine efforts to protect both human health and the environment.

Read more of this story at Slashdot.

Mice Give First Aid

Par : EditorDavid
9 mars 2025 à 16:34
Slashdot reader databasecowgirl writes: The Times is reporting an interesting study published in Science in which mice demonstrated doing first aid. In the replicated study, an anaesthetised mouse is exposed to another mouse who recognises the distress and clears airway to revive the unconscious mouse. The mice had never seen an unconscious animal before, so the behaviour is thought to be instinctive. From the Times: Large social mammals have previously been documented lending assistance to each other. Chimpanzees have been seen tending to wounded companions, dolphins are known to push distressed pod members to the surface to help them breathe, and elephants have been observed assisting their ailing relatives. Never before, however, has such a meticulous, paws-on approach to first aid been recorded in a creature as small as a mouse.

Read more of this story at Slashdot.

America Lost 22% of Its Butterflies Within Two Decades

Par : EditorDavid
8 mars 2025 à 19:34
Butterflies "are vanishing from U.S. landscapes at an alarming rate," reports CBS News: A comprehensive study, published Thursday in the journal Science, found that 22% of butterflies in the United States disappeared between 2000 and 2020... The researchers behind the Science study used data from more than 12.6 million butterflies spanning 342 individual species, drawing from 76,000 surveys across 35 nationwide monitoring programs. Funded by the U.S. Geological Survey, the study was the first to integrate such a vast dataset, its authors said. The findings revealed that 33% of butterfly species have experienced significant population declines over the past two decades, with 107 out of the 342 species examined losing more than half of their population — including 22 species that declined by more than 90%. Meanwhile, only 3% of species showed population increases... Ultimately, the butterfly decline is part of a larger global trend of insect population loss, with insects declining by about 1-2% annually, the study's authors said. Butterflies play an essential role in ecosystems, pollinating flowers, crops, and other plants. Their decline could have far-reaching impacts on plant reproduction and the health of ecosystems. Just three months ago the U.S. Fish and Wildlife Service said America's western migratory populations of monarch butterflies had declined by more than 95% since the 1980s, putting them "at greater than 99% chance of extinction by 2080." (America's eastern migratory population were estimated to have declined by approximately 80%.) This latest study found that one factor is climate change, according to CBS News, which reduces food sources, disrupts breeding cycles and increases habitat stress. (Another factor is pesticide use, which fortunately can be adjusted with various policy interventions and farming practices.) And one of the study's co-authors tells CBS News that "the things we do in our own backyards actually make a difference." They recommend allowing backyard to "grow wild" with native plants (and reducing pesticide use) — even creating "habitat spaces" for insects like small piles of brush. "Even simple actions — like leaving a strip of wildflowers or planting species that support pollinators — can provide crucial resources for butterflies and other insects."

Read more of this story at Slashdot.

Intelligence artificielle, genre grammatical féminin

À l’occasion de la Journée internationale des droits des femmes et pour la paix internationale, on va parler d’un sujet « tendance » vu sous un angle un peu spécifique : celui des femmes qui la font. Parce que, bien que l’on soit amené à reprocher aux intelligences artificielles une forme de sexisme due à leurs jeux de données, les femmes n’ont pas été, ne sont pas étrangères à leur conception. C’est donc l’occasion de donner les portraits de Karen Spärck Jones (1935 – 2007) et de Chloé-Agathe Azencott qui ont contribué, et contribuent, à leur existence, et de rappeler les « petits » défauts des IA.

Sommaire

Karen Spärck Jones (1935 – 2007) entre moteur de recherche et traitement automatique du langage

Karen Spärck Jones fait ses études à Cambridge. Elle commencera sa carrière en 1953 à l’unité de recherche linguistique de l’université avec la linguiste Margaret Mastermann, elle-même pionnière dans le domaine de la linguistique informatique. Ses recherches porteront sur les moteurs de recherche et du traitement du langage.

Margaret Mastermann lui confie la mission de programmer un ordinateur qui devait comprendre des mots polysémiques, elle génèrera un thésaurus. Elle entame également une collaboration avec l’informaticien Roger Needham qu’elle épousera en 1958.

En 1964, Karen Spärck Jones publie un article d’une importance capitale Synonymy and semantic classification (Synonymie et classification sémantique), considéré comme un document fondamental dans le domaine du traitement du langage naturel. Une importance qui s’accroîtra avec l’arrivée du World Wide Web.

À partir de 1994, ses travaux portent sur les outils de recherche d’information, notamment les applications vocales, les interrogations de bases de données, la modélisation des utilisateurs et des agents, le résumé et l’évaluation des systèmes d’informations et des systèmes linguistiques.

Elle est élue en 1995 membre de la British Academy (académie des sciences humaines et sociales du Royaume-Uni) dont elle sera vice-présidente de 2000 à 2002. Elle obtiendra aussi plusieurs prix : le Gerard Salton Award en 1988 (un prix de l’ACM et du SIGIR (en), deux associations états-uniennes en informatique), le prix de l’ACL en 2004 (une société savante américaine spécialisée dans le traitement des langues) et la médaille Lovelace de la British Computer Society en 2007 (sept ans après Linus Torvalds).

Elle dira, dans un entretien suite à la réception de la médaille Lovelace :

J’étais sidérée. J’ai regardé la liste des précédents récipiendaires et j’ai pensé : « Qu’est-ce que je viens faire dans ce groupe de gens ? » Mais j’étais particulièrement enchantée de voir que j’étais la première femme à l’obtenir. Très agréable, j’ai vraiment apprécié.
Je pense qu’il est très important de faire en sorte qu’il y ait plus de femmes en informatique. Mon slogan est _« l’informatique est trop importante pour être laissée aux

Chloé-Agathe Azencott, spécialiste de l’apprentissage automatique

On change de génération avec Chloé-Agathe Azencott, elle aurait pu être une petite fille de Karen Spärck Jones.

Chloé-Agathe Azencott est professeure à l’École des Mines de Paris et à l’Institut Curie où elle enseigne l’apprentissage automatique ou apprentissage statistique ou encore apprentissage machine, en anglais machine learning. Elle a fait ses études à l’IMT Atlantique (ENST Bretagne à son époque, et, plus familièrement « Télécoms Bretagne ») et à l’Université de Californie à Irvine (UC Irvine).

Elle est récipiendaire, en 2021, du premier prix de la Jeune ingénieure en intelligence artificielle, organisé par le cabinet de conseil en communication Tilder en partenariat avec France Digitale (une association de startups et de VCs) et le magazine Challenges. Elle est l’autrice d’un livre sur l’apprentissage automatique : Introduction au machine learning chez Dunod, deuxième édition février 2022. On peut en télécharger une version PDF gratuitement mais sans les exercices. La version papier est en réimpression.

Comment définit-elle l’apprentissage automatique qui est l’un des sous-domaines de l’intelligence artificielle ? Elle commence par définir l’apprentissage qui est le fait d’acquérir une compétence par l’expérience et la pratique. Dans une conférence donnée le 25 novembre 2021 à l’Institut Henri Poincaré elle ajoute :

j’aime cette définition parce que je peux l’appliquer à ce qui se passe avec des humains, donc un enfant qui apprend à marcher en essayant de marcher et plus il s’entraine à marcher, plus il marche. Ça s’applique à mes étudiants et mes étudiantes qui, à force de résoudre des problèmes de maths, acquièrent l’expérience et la compétence de savoir faire des stats et des probas et puis ça s’applique aussi aux ordinateurs à condition de, peut-être, détourner un peu le sens de « compétences » et d’« expérience ».

Pour une machine la compétence est un algorithme donc

un nouvel algorithme capable de faire des choses que l’ordinateur n’était pas capable de faire avant et l’expérience ou la pratique ça va être des exemples ou des données.

Définition qui peut être complétée par celle qu’elle donne dans l’introduction de son livre :

Dans le cas d’un programme informatique, […], on parle d’apprentissage automatique, ou machine learning, quand ce programme a la capacité de se modifier lui-même sans que cette modification ne soit explicitement programmée. Cette définition est celle donnée par Arthur Samuel (1959). On peut ainsi opposer un programme classique, qui utilise une procédure et les données qu’il reçoit en entrée pour produire en sortie des réponses, à un programme d’apprentissage automatique, qui utilise les données et les réponses afin de produire la procédure qui permet d’obtenir les secondes à partir des premières.
[…]
Ce point de vue informatique sur l’apprentissage automatique justifie que l’on considère qu’il s’agit d’un domaine différent de celui de la statistique. Cependant, nous aurons l’occasion de voir que la frontière entre inférence statistique et apprentissage est souvent mince. Il s’agit ici, fondamentalement, de modéliser un phénomène à partir de données considérées comme autant d’observations de celui-ci.

Elle pense toutefois qu’il convient de garder un esprit critique vis-à-vis de l’IA notamment parce que :

l’on y injecte souvent des connaissances déjà établies (lois de la physique, notions de linguistique, connexions entre concepts), ces modèles restent essentiellement statistiques et ne mènent aucun raisonnement. L’intelligence artificielle ne remplacera pas les scientifiques, Chloé-Azencott, La Croix, 15 avril 2024

Chloé-Agathe Azencott considère, en outre, qu’il est extrêmement important :

de donner plus de visibilité aux femmes scientifiques, et notamment à celles qui travaillent dans le domaine du machine learning et de la science des données (elles ne représentent que 2% des scientifiques dans ce domaine), mais aussi à toutes les identités, afin de refléter la diversité dans tous ces aspects, y compris social. Chloé-Agathe Azencott, mathématiques et machine learning au service de la recherche médicale, Institut Henri Poincaré, [sd].

Une nécessité qui se démontre ci-après.

Sexiste, raciste l’IA ?

Avant tout chose, une précision. Le sexisme et le racisme ce sont à la fois des opinions et des manifestations. Si les intelligences artificielles n’ont pas d’opinions, en revanche ce qui en sort peut être manifestement raciste ou sexiste et c’est cet aspect-là qu’on va voir à travers une série d’articles de diverses origines parus entre 2017 et 2024. Les articles sont présentés dans l’ordre chronologique.

Il est intéressant de voir, à partir de cette sélection, les questions que pose l’IA et de relever l’impact extrêmement important de cette technologie sur la société, qu’il s’agisse d’emploi (tri des candidatures), de santé, de droits d’auteurs ou de justice, entre autres.

L’intelligence artificielle reproduit aussi le sexisme et le racisme des humains, Morgane Tual, 15 avril 2017, Le Monde.

L’article se fait le relais d’une étude de la vue Science (en) du 14 avril 2017 et commence ainsi :

Les femmes associées aux arts et au foyer, les hommes aux professions scientifiques… Ces stéréotypes ont tellement la vie dure qu’ils se retrouvent reproduits dans des programmes d’intelligence artificielle (IA).

Un problème qui :

ne se situe pas seulement au niveau du langage. Quand un programme d’IA est devenu jury d’un concours de beauté, en septembre 2016, il a éliminé la plupart des candidats noirs.

L’article signale que ce ne sont pas les IA qui ont des préjugés, mais bien nous qui leur donnons les nôtres et relève que cela concerne la sélection des CV, la justice, les assurances. Au niveau des pistes pour redresser la barre, il est suggéré une meilleure diversité au niveau des personnes qui conçoivent les IA (une diversité très mise à mal par la nouvelle présidence des États-Unis et des patrons des GAFAM). Une autre piste évidente : travailler sur les données. L’article conclut que la solution du problème serait de modifier les humains.

L’intelligence artificielle, aussi raciste et sexiste que nous, Fabien Goubet, 4 mai 2017, Le temps.ch.

L’article est basé sur la même étude que celle citée plus haut et il commence assez fort :

Les androïdes rêvent-ils de moutons noirs expulsés par des moutons blancs ? Avec leurs capacités de raisonnement froides, basées sur des calculs complexes, on imagine les intelligences artificielles dénuées de tout préjugé. C’est tout le contraire, comme vient de le confirmer une étude parue en avril dans la revue « Science ».

Il explique que le logiciel, GloVe, utilisé pour l’étude :

s’est prêté au jeu d’association d’idées. Ce programme est une IA basée sur le «machine learning», c’est-à-dire capable d’apprendre, à partir de nombreux exemples, à classer des informations selon des critères exigés par un humain. C’est sur ce type d’apprentissage que reposent notamment les algorithmes de reconnaissance d’images utilisés par Facebook ou Google. Pour entraîner GloVe, Aylin Caliskan l’a donc « nourri » avec un gigantesque corpus de 840 milliards de mots (en) issus du Web, en 40 langues différentes. Ses réponses laissent songeur. Comme un être humain, le programme a associé des noms de fleurs à des connotations positives, tandis que des noms d’insectes, par exemple, ont été catégorisés plutôt négativement.

Il ajoute que ces « biais plutôt innocents » ont été reproduits plus problématiquement : aux prénoms féminins les associations avec la famille, aux prénoms masculins celles avec la carrière, et un meilleur traitement était réservé aux noms à consonance européenne. Comportement qu’un spécialiste des réseaux de neurones artificiels et de la théorie neuronale de la cognition, Claude Touzet, explique :

Les machines capables d’apprentissage sont un miroir du comportement humain. En les nourrissant avec un discours humain forcément biaisé, il est naturel qu’elles le reproduisent.

Avec des idées de solutions possibles, par exemple imposer des lois aux IA, ce que Sébastien Konieczny, directeur de recherche au CNRS, trouve difficile car :

on ne sait pas encore vraiment comment réguler ces algorithmes avec des règles éthiques et morales, pas plus – et c’est tout aussi inquiétant – qu’on ne comprend comment la machine a pris sa décision.

Une solution possible :

serait d’associer ces algorithmes à d’autres méthodes permettant, elles, de rendre compte du raisonnement.

Comment une IA peut devenir raciste ou sexiste, Anne Cagan, 25 juin 2020, Journal du geek.

La base de l’article est une interview de Stéphane d'Ascoli, qui deviendra docteur en intelligence artificielle en 2022 et venait de publier une livre de vulgarisation « Comprendre la révolution de l’intelligence artificielle » aux éditions First. Stéphane d’Ascoli donne l’exemple des recrutements biaisés par les IA :

On a tendance à s’imaginer que les IA sont froides, objectives et parfaitement rationnelles mais ce n’est pas le cas. Elles apprennent de nos données et nos données sont biaisées. Si, pendant dix ans, les femmes ont été défavorisées lors du processus de recrutement d’une entreprise et que celle-ci utilise ces données pour entraîner une IA, il y a des chances que l’IA déduise que les CV de femmes sont moins pertinents pour cette entreprise et qu’elle continue de les défavoriser. Les intelligences artificielles n’ont pas notre esprit critique.

À la question : « comment éviter ces dérives ». Il répond qu’une piste faisable serait d’assurer que :

les jeux de données sur lesquels on va entraîner l’IA sont équilibrés et diversifiés.

Et qu’il faut, évidemment, tester l’IA pour vérifier qu’elle traite tout le monde de façon identique.

L’IA serait-elle raciste ? C’est ce qu’affirme une étude, Daniel Ichbiah, 18 novembre 2023, Futura.

L’étude en question, datée de juillet 2023 a été menée par une équipe plurinationale : Shangbin Feng et Yulia Tsvetkov de l’Université de Washington (USA), Chan Young Park de l’Université privée Carnegie Mellon (USA) et Yuhan Liu de l’Université Jiaotong de Xi'an (Chine).

À chaque fois, il a été noté que les outils d’IA générative manifestaient des biais sociaux et politiques particuliers, en relation avec le lieu où le corpus de données avait été collecté.

L’article relève les inquiétudes de la Cnil anglaise qui estime que l’usage de l’IA pourrait aboutir à « des conséquences dommageables pour la vie des gens ».

ChatGPT et misogynie : l’intelligence artificielle est-elle sexiste ?, Nadine Jürgensen, 11 février 2024, TDG (Tribune de Genève).

D’entrée de jeu, la question est posée :

Deepfakes sur Taylor Swift et Sibel Arslan, représentations suggestives du corps des femmes: que faire contre une IA parfois machiste ?

L’autrice explique qu’elle a testé ChatGPT et qu’elle a été déçue : réponses maladroites, insatisfaisantes, voire fausses. Elle ajoute :

Jusqu’à présent, l’IA ne semble pas exercer une grande attraction sur le sexe féminin. En effet, seuls 30% des utilisatrices et des utilisateurs actifs sont des femmes. Elles sont critiques à l’égard des résultats de l’IA et ne les perçoivent pas comme justes. Oui, elles ont l’impression de tricher lorsqu’elles utilisent l’IA au quotidien. Elle serait pratique pour les hommes, tandis que les femmes peuvent avoir l’impression d’être moins qualifiées parce qu’elles la sollicitent.

Elle reprend la question des sources de données des IA et aborde un point intéressant qui est celui de la réglementation, la Suisse n’en disposant pas. Elle évoque la question de la propriété intellectuelle :

Les artistes et les professionnels des médias de notre pays demandent une meilleure protection de leurs droits d’auteur. Et tant d’autres questions, par exemple où et comment l’intelligence artificielle peut se «servir» de contenus créés par l’homme ou comment protéger nos données personnelles. En outre, il est essentiel de savoir si un contenu a été créé avec l’IA ou non.

Elle conclut, après avoir indiqué qu’elle avait recommencé à jouer avec l’IA, qu’elle continuera à écrire sa chronique elle-même.

Pourquoi les IA génératives sont-elles sexistes, racistes et homophobes ?, Justine Havelange, 29 juillet 2024, EJO.

Cet article est issu d’une rencontre avec Anne Jobin, chercheuse au département informatique de l’Université de Fribourg (Suisse) présidente de la commission fédérale des médias et spécialiste des technologies digitales.

« La technologie n’est ni bonne, ni mauvaise, ni neutre », cette citation de l’historien des sciences Melvin Kranzberg est pour Anna Jobin un guide « pour se rendre compte de la vitesse des changements et de la cohabitation nécessaire entre nous et la technologie. »

L’IA n’est pas neutre, car elle reproduit les stéréotypes de notre société (comme on l’a déjà vu plus haut).

Les bases de données, même gigantesques, sont parfois la source d’un « sous-apprentissage ». Comprenez par là qu’il n’existe pas assez de données sur certains types de personne.
Ce constat a mobilisé l’UNESCO mais également « Numeum », le syndicat [français] de l’industrie du Numérique. L’une des pistes de solution trouvées par ces organisations est de diversifier les équipes de développeurs et d’ingénieurs ou de faire appel à des sociologues.

À la question des solutions possibles : l’ajustement des biais, modèles, bases de données et algorithmes, est une réponse.

▶ Pour finir, et occuper vos futures longues soirée de printemps, d’été, d’automne et d’hiver, la lecture du blog Entretien avec un vampire d’un professeur des universités en informatique qui a fait un assez triste constat.

Depuis deux ans, les IA génératives ont déferlé absolument partout, et donc aussi dans l’enseignement. Plus spécifiquement, les étudiant·e·s s’en servent quotidiennement pour résoudre les exercices que je leur donne, je le constate, iels me le disent. J’ai beau prévenir qu’en faisant ainsi, l’objet même des exercices disparaît (on ne s’exerce plus), le rouleau compresseur marketing les convainc que ça peut les aider et on me dit même comment telle ou telle IA est interrogée pour expliquer le programme qu’elle propose, et les concepts qui vont avec, utilisée comme une vraie auxiliaire de travail en somme.

L’idée du blog étant d’évaluer l’IA comme il le fait avec ses étudiants et de documenter ce travail.

Commentaires : voir le flux Atom ouvrir dans le navigateur

❌
❌