Latest news with #MicrosoftResearch
Mint
3 days ago
- Science
- Mint
Does AI make you stupid?
AS ANYBODY WHO has ever taken a standardised test will know, racing to answer an expansive essay question in 20 minutes or less takes serious brain power. Having unfettered access to artificial intelligence (AI) would certainly lighten the mental load. But as a recent study by researchers at the Massachusetts Institute of Technology (MIT) suggests, that help may come at a cost. Over the course of a series of essay-writing sessions, students working with as well as without ChatGPT were hooked up to electroencephalograms (EEGs) to measure their brain activity as they toiled. Across the board, the AI users exhibited markedly lower neural activity in parts of the brain associated with creative functions and attention. Students who wrote with the chatbot's help also found it much harder to provide an accurate quote from the paper that they had just produced. The findings are part of a growing body of work on the potentially detrimental effects of AI use for creativity and learning. This work points to important questions about whether the impressive short-term gains afforded by generative AI may incur a hidden long-term debt. The MIT study augments the findings of two other high-profile studies on the relationship between AI use and critical thinking. The first, by researchers at Microsoft Research, surveyed 319 knowledge workers who used generative AI at least once a week. The respondents described undertaking more than 900 tasks, from summarising lengthy documents to designing a marketing campaign, with the help of AI. According to participants' self-assessments, only 555 of these tasks required critical thinking, such as having to review an AI output closely before passing it to a client, or revising a prompt after the AI generated an inadequate result on the first go. The rest of the tasks were deemed essentially mindless. Overall, a majority of workers reported needing either less or much less cognitive effort to complete tasks with generative-AI tools such as ChatGPT, Google Gemini or Microsoft's own Copilot AI assistant, compared with doing those tasks without AI. Another study, by Michael Gerlich, a professor at SBS Swiss Business School, asked 666 individuals in Britain how often they used AI and how much they trusted it, before posing them questions based on a widely used critical-thinking assessment. Participants who made more use of AI scored lower across the board. Dr Gerlich says that after the study was published he was contacted by hundreds of high-school and university teachers dealing with growing AI adoption among their students who, he says, 'felt that it addresses exactly what they currently experience". Whether AI will leave people's brains flabby and weak in the long term remains an open question. Researchers for all three studies have stressed that further work is needed to establish a definitive causal link between elevated AI use and weakened brains. In Dr Gerlich's study, for example, it is possible that people with greater critical-thinking prowess are just less likely to lean on AI. The MIT study, meanwhile, had a tiny sample size (54 participants in all) and focused on a single narrow task. Moreover, generative-AI tools explicitly seek to lighten people's mental loads, as many other technologies do. As long ago as the 5th century BC, Socrates was quoted as grumbling that writing is not 'a potion for remembering, but for reminding". Calculators spare cashiers from computing a bill. Navigation apps remove the need for map-reading. And yet few would argue that people are less capable as a result. There is little evidence to suggest that allowing machines to do users' mental bidding alters the brain's inherent capacity for thinking, says Evan Risko, a professor of psychology at the University of Waterloo who, along with a colleague, Sam Gilbert, coined the term 'cognitive offloading" to describe how people shrug off difficult or tedious mental tasks to external aids. The worry is that, as Dr Risko puts it, generative AI allows one to 'offload a much more complex set of processes". Offloading some mental arithmetic, which has only a narrow set of applications, is not the same as offloading a thought process like writing or problem-solving. And once the brain has developed a taste for offloading, it can be a hard habit to kick. The tendency to seek the least effortful way to solve a problem, known as 'cognitive miserliness", could create what Dr Gerlich describes as a feedback loop. As AI-reliant individuals find it harder to think critically, their brains may become more miserly, which will lead to further offloading. One participant in Dr Gerlich's study, a heavy user of generative AI, lamented 'I rely so much on AI that I don't think I'd know how to solve certain problems without it." Many companies are looking forward to the possible productivity gains from greater adoption of ai. But there could be a sting in the tail. 'Long-term critical-thinking decay would likely result in reduced competitiveness," says Barbara Larson, a professor of management at Northeastern University. Prolonged AI use could also make employees less creative. In a study at the University of Toronto, 460 participants were instructed to propose imaginative uses for a series of everyday objects, such as a car tyre or a pair of trousers. Those who had been exposed to ideas generated by AI tended to produce answers deemed less creative and diverse than a control group who worked unaided. When it came to the trousers, for instance, the chatbot proposed stuffing a pair with hay to make half of a scarecrow—in effect suggesting trousers be reused as trousers. An unaided participant, by contrast, proposed sticking nuts in the pockets to make a novelty bird feeder. There are ways to keep the brain fit. Dr Larson suggests that the smartest way to get ahead with AI is to limit its role to that of 'an enthusiastic but somewhat naive assistant". Dr Gerlich recommends that, rather than asking a chatbot to generate the final desired output, one should prompt it at each step on the path to the solution. Instead of asking it 'Where should I go for a sunny holiday?", for instance, one could start by asking where it rains the least, and proceed from there. Members of the Microsoft team have also been testing AI assistants that interrupt users with 'provocations" to prompt deeper thought. In a similar vein, a team from Emory and Stanford Universities have proposed rewiring chatbots to serve as 'thinking assistants" that ask users probing questions, rather than simply providing answers. One imagines that Socrates might heartily approve. Get with the program Such strategies might not be all that useful in practice, even in the unlikely event that model-builders tweaked their interfaces to make chatbots clunkier, or slower. They could even come at a cost. A study by Abilene Christian University in Texas found that AI assistants which repeatedly jumped in with provocations degraded the performance of weaker coders on a simple programming task. Other potential measures to keep people's brains active are more straightforward, if also rather more bossy. Overeager users of generative AI could be required to come up with their own answer to a query, or simply wait a few minutes, before they're allowed to access the AI. Such 'cognitive forcing" may lead users to perform better, according to Zana Buçinca, a researcher at Microsoft who studies these techniques, but will be less popular. 'People do not like to be pushed to engage," she says. Demand for workarounds would therefore probably be high. In a demographically representative survey conducted in 16 countries by Oliver Wyman, a consultancy, 47% of respondents said they would use generative-AI tools even if their employer forbade it. The technology is so young that, for many tasks, the human brain remains the sharpest tool in the toolkit. But in time both the consumers of generative ai and its regulators will have to assess whether its wider benefits outweigh any cognitive costs. If stronger evidence emerges that ai makes people stupid, will they care?
International Business Times
12-07-2025
- Science
- International Business Times
What is BioEmu? Microsoft's New AI Breakthrough Decodes Protein Movements in Hours
Microsoft has introduced BioEmu, a powerful artificial intelligence tool that could revolutionize the way scientists study protein movement and structure. Traditionally, this kind of molecular analysis takes years of heavycomputation. Now, thanks to BioEmu, the same tasks can be done in a matter of hours. Announcing the innovation on X (formerly Twitter), Microsoft CEO Satya Nadella said the AI system can emulate the shifting forms proteins take—known as structural ensembles—which are key to understanding how diseases function and how drugs can be created to treat them. Developed by Microsoft Research's AI for Science team, BioEmu version 1.1 demonstrates remarkable accuracy. It closely reflects real-world experimental data with prediction errors of less than 1 kcal/mol and correlation scores above 0.6 across large datasets. The AI was trained using more than 200 milliseconds of molecular simulation data, alongside half a million protein stability experiments and extensive structural information. What makes BioEmu exceptional is its ability to detect subtle and rare changes in proteins—especially "cryptic binding pockets," which are hidden areas that could become future drug targets. This deep insight into protein behavior could be a game-changer for pharmaceutical development. Unlike older systems that demand long-term GPU usage and intensive computing power, BioEmu provides rapid results while dramatically cutting costs and processing time. Microsoft Research highlighted that BioEmu mimics complex molecular movements such as domain shifts and local unfolding—both of which are essential for understanding how proteins work inside living organisms. The results were published in the journal Science, presenting BioEmu as a generative deep learning model capable of replicating protein behavior both in labs and inside the human body. Experts believe this new tool could transform fields like drug design, synthetic biology, and medical research, giving scientists a major head start in identifying new treatments for diseases.
Mint
11-07-2025
- Science
- Mint
Satya Nadella announces BioEmu, Microsoft's AI to fast-track drug discovery: All about it
Microsoft has launched BioEmu, an artificial intelligence system designed to speed up the process of understanding how proteins behave in the human body, work that traditionally takes years of complex computer simulations. Announcing the breakthrough on X, Microsoft Chairman and CEO Satya Nadella said, 'Understanding protein motion is essential to understanding biology and advancing drug discovery. Today we're introducing BioEmu, an AI system that emulates the structural ensembles proteins adopt, delivering insights in hours that would otherwise require years of simulation.' You may be interested in The AI system, developed by Microsoft Research's AI for Science team, is capable of predicting the different shapes and movements (or "conformational changes") that proteins can take as they function inside living organisms. This ability is crucial for understanding how diseases work and how new drugs can be designed to target specific proteins. In a detailed explanation on X, Microsoft Research stated that BioEmu version 1.1 can closely match real-world experimental protein stability data, with prediction errors of less than 1 kcal/mol and strong correlation scores above 0.6 on large test datasets. The AI was trained using more than 200 milliseconds of molecular dynamics simulations, data from over 500,000 protein stability experiments, and vast structural information. Unlike traditional methods that require extensive GPU usage over years, BioEmu can now complete the same simulations within hours, saving enormous computing time and cost. One of BioEmu's standout features is its ability to predict hard-to-detect changes in protein structure, including the formation of so-called 'cryptic' binding pockets, hidden spots on proteins that could be targeted by future drugs. 'BioEmu can emulate equilibrium distributions of millisecond-timescale molecular dynamics simulations at many orders of magnitude faster speeds,' Microsoft Research noted. 'It also predicts functionally important movements, like large domain shifts and local unfolding, which are often central to how a protein works.' The research has been published in the journalScience, showcasing BioEmu as a generative deep learning model designed to replicate the structural ensembles of proteins in lab settings or within the human body. These protein ensembles are vital to understanding how proteins perform their roles, especially since many proteins constantly shift between different forms. The launch of BioEmu is expected to have significant impact on fields such as drug development, disease research, and synthetic biology, potentially allowing scientists to discover and test new therapies far faster than ever before.
NDTV
11-07-2025
- Science
- NDTV
Microsoft's New AI Tool To Help Advance Drug Discovery: Satya Nadella
New Delhi: Microsoft's new AI system BioEmu will help decode protein motion and help in faster discovery of drugs, said CEO Satya Nadella on Friday. Biomolecular Emulator-1 (BioEmu-1) is a deep learning model that can generate thousands of protein structures per hour on a single graphics processing unit (GPU). "Understanding protein motion is essential to understanding biology and advancing drug discovery," said Mr Nadella, in a post on social media platform X. Understanding protein motion is essential to understanding biology and advancing drug discovery. Today we're introducing BioEmu, an AI system that emulates the structural ensembles proteins adopt, delivering insights in hours that would otherwise require years of simulation. — Satya Nadella (@satyanadella) July 10, 2025 Sharing a research paper on the model, Mr Nadella added that "today we're introducing BioEmu, an AI system that emulates the structural ensembles proteins adopt, delivering insights in hours that would otherwise require years of simulation". Proteins play an essential role -- from forming muscle fibers to protecting against diseases -- in almost all biological processes in both humans and other life forms. While recent years have seen progress in better understanding of the protein structures, predicting a single protein structure from its amino acid sequence was not feasible. But, with BioEmu-1, scientists can get a glimpse into the rich world of different structures each protein can adopt, or structural ensembles. This enables them to get a deeper understanding of how proteins work -- critical for designing more effective drugs. "BioEmu integrates over 200 milliseconds of molecular dynamics (MD) simulations, static structures, and experimental protein stabilities using novel training algorithms. It captures diverse functional motions --including cryptic pocket formation, local unfolding, and domain rearrangements -- and predicts relative free energies with 1 kcal/mol accuracy compared to millisecond-scale MD and experimental data," revealed scientists from AI for Science at Microsoft Research, in the paper published in the journal Science. The team noted that BioEmu provides "mechanistic insights by jointly modelling structural ensembles and thermodynamic properties". The approach pays off the cost of MD and experimental data generation, demonstrating a scalable path toward understanding and designing protein function, they added.
Geek Wire
08-07-2025
- Science
- Geek Wire
Concrete goes green: UW and Microsoft use seaweed to create novel carbon-trapping cement
Sustainability: News about the rapidly growing climate tech sector and other areas of innovation to protect our planet. SEE MORE Meng-Yen Lin, a member of Eleftheria Roumeli's lab at the University of Washington, mixes up seaweed-containing batter for making cement. (UW Photo) Researchers have developed a novel solution for trapping carbon in concrete by blending a sustainable, easy-to-grow green seaweed into the industrial batter that makes cement, all without reducing its strength. Scientists from the University of Washington and Microsoft Research used machine learning to expedite their experimentation, coming up with a solution that lowered the cement's global warming impacts by 21% A study on their work is publishing online today in Matter, a science journal focused on materials science. The research was led by Eleftheria Roumeli, a UW assistant professor in materials science and engineering, and Kristen Severson, a senior researcher with Microsoft Research. Roumeli previously developed an algae-based plastic that biodegrades in nature as quick as a banana peel. The scientists wanted to tackle concrete given that it's responsible for between 8-11% of global carbon emissions. Cement, the key component of concrete, contributes to nearly all of that climate burden. There are efforts worldwide to curb concrete's carbon footprint by using clean energy in generating the heat needed to product cement and by swapping in different ingredients to bind the cement such as industrial wastes like fly ash and furnace slag. Scientists with the University of Washington and Microsoft Research used dehydrated seaweed to make a high-performing, lower-carbon cement. (UW Photo) Roumeli and Severson turned to the seaweed given that it pulls carbon from the air and locks it away during photosynthesis. Others have tested the use of smaller algae, but the researchers were the first to choose a macroalgae called Ulva as their material because it has a more robust cellular structure they suspected could help reinforce the cement. The scientists dehydrated the seaweed before incorporating it and were able to successfully use much higher amounts in the cement than previous demonstrations. One of the difficulties in developing lower-carbon cement is the time required to make sure it's strong enough to be used in the construction of buildings, bridges and other infrastructure. Concrete gains strength over time, leading researchers to typically test the material after 28 days. To speed up that process, the scientists developed a machine learning model that could predict how much stronger a sample would get over four weeks, allowing them to abandon underperforming strategies earlier in the process. The approach shaved 112 days off of experiment time. The model is adaptable and can be applied to other solutions for making greener cement. '[T]his work establishes a framework with the potential to accelerate the design of sustainable cement with feasible experimental resources while satisfying critical performance requirements,' the authors wrote. Other authors of the paper were Meng-Yen Lin and Paul Grandgeorge, who did the research as a graduate student and a post-doc, respectively, in Roumeli's lab. The publication is titled 'Closed-loop optimization using machine learning for the accelerated design of sustainable cements incorporating algal biomatter.'
