Background: Approximately 70% of survivors of stroke have problems with arm function. Physiotherapists assess arm functional range of motion (ROM) using either a goniometer or functional questionnaires, which lack objective accuracy and require a skilled physiotherapist. We developed the Track-UL algorithm based on a markerless motion capture system to measure arm ROM. Objective: This study aimed to measure the agreement between our novel Track-UL algorithm and Kinovea software in assessing arm ROM during functional tasks in the laboratory and home settings. Methods: Videos were recorded while 27 survivors of chronic stroke performed 4 functional tasks (forward reaching, arm abduction, moving the hand toward the mouth, and moving the hand toward the head) in the laboratory and at home. The videos were analyzed by 2 independent raters using the Track-UL algorithm and Kinovea software. The limits of agreement and intraclass correlation coefficients were calculated. Results: We found no clinically significant systematic bias in shoulder and elbow angle, with good agreement between the Track-UL algorithm and Kinovea software (assessed via Bland-Altman plots). The 95% limits of agreement were –3.18 to 6.41 degrees for the shoulder joint and −5.35 to 8.78 degrees for the elbow joint in the laboratory setting, and –6.21 to 3.62 degrees for the shoulder joint and −4.06 to 2.53 degrees for the elbow joint in the home setting. There was excellent absolute agreement between the measurement tools across all tasks and joints; intraclass correlation coefficient values ranged from 0.97 (95% CI 0.97-0.99) to 0.99 (95% CI 0.99-0.99; <.001 for both laboratory and home measurements). Conclusions: The novel Track-UL algorithm is an accurate, valid, and easy tool that can be used to assess upper-limb ROM in survivors of stroke at clinics and potentially at home. This will support physiotherapists in remotely monitoring and adapting rehabilitation programs.
Background: Persisting symptoms affect about one-third of youth following concussion. Mental health history, distress, and coping style are key predictors of prolonged recovery. Early and scalable psychological interventions, such as mindfulness-based intervention (MBI) delivered via smartphones, may improve patients’ ability to regulate their emotions and neurophysiologically recover, reducing overall symptom burden. However, no digital therapeutic (DTx) trials in adolescents experiencing concussion exist. Objective: This study primarily aimed to assess the feasibility of conducting a larger randomized controlled trial (RCT) evaluating the effectiveness of a DTx-MBI in adolescents with a concussion compared with an attention-matched sham intervention. Methods: This was a Health Canada-regulated, parallel-group, blinded, single-crossover feasibility RCT. Adolescents aged 12 to <18 years presenting to a Pediatric Emergency Department or interdisciplinary concussion clinic within 7 days of a physician-diagnosed concussion were approached for participation from November 2022 to June 2024. After providing consent, participants were randomized (1:1), stratified by sex, to either the experimental group (DTx-MBI) or the control group (sham, attention-matched math puzzle game). The DTx-MBI was delivered via the AmDTx platform (Mobio Interactive Pte Ltd, Singapore) as a custom-designed 4-to-8-week program of 8 standardized modules for adolescents with concussion, including audio-recorded guided mindfulness exercises, goal setting, journaling, and psychoeducation. The control intervention, delivered through the same interface, excluded mindfulness content and instead featured the open-source game “2048”. Participants in both groups were encouraged to engage with the app for at least 10 minutes/day, at least 4 days/week. Feasibility criteria to support progression to a full-scale RCT included: eligibility rate >40% of those screened; recruitment rate >50% of eligible participants randomized; intervention credibility >70% scoring above the midpoint on the credibility and expectancy questionnaire (CEQ) at 1 week; retention >75% of randomized participants completing 4-week outcomes; and adherence >70% completing 10 minutes of intervention on at least 4 days/week for 4 weeks. Results: A total of 124 out of 195 (63.6%) screened youth met eligibility criteria. Of these, 99/124 (79.8%) consented and were randomized to either the DTx-MBI group (n=49, median [IQR] age=15.28 [13.66‐16.19] years, 30 [61.2%] female) or the Sham group (n=50, median [IQR] age=14.92 [13.32‐16.71] years, 30 [60.0%] female). Credibility was high, with 62/83 (74.7%) of participants scoring above the credibility midpoint (DTx-MBI: 75.0%; Sham: 74.4%). Retention was strong, with 89/99 (89.9%) of participants completing the 4-week outcomes (DTx-MBI: 89.8%; control: 90.0%). Overall adherence was moderate (54/99 [54.5%]; DTx-MBI: 59.2%; control: 50.0%), and a little higher among outcome assessment completers (53/89 [59.6%]; DTx-MBI: 63.6%; Sham: 55.6%). Feasibility indicators were similar between groups. Conclusions: This feasibility trial supports the implementation of a larger RCT, with modifications to enhance adherence, to rigorously evaluate the clinical efficacy of the DTx-MBI. By targeting modifiable psychological risk factors through a scalable digital platform, DTx-MBI could be a low-burden, cost-effective adjunct to pediatric concussion care. Trial Registration: ClinicalTrial.gov NCT05105802; International Registered Report Identifier (IRRID): RR2-10.2196/57226
One of the most challenging aspects of combatting HIV-1 infection is that the virus continually evades neutralizing antibodies. However, one consequence of this is that a small percentage of people with HIV-1 (1-5%) develop rare, broadly neutralizing antibodies (bNAbs) that can neutralize a large fraction of global HIV-1 isolates. These broadly neutralizing antibodies are among the most promising new long-acting HIV treatments, offering the potential to forego traditional daily dose of antiretroviral drugs. Indeed, a recent trial found that participants who received a single dose of two bNAbs maintained a nearly undetectable viral load for up to 20 weeks, and a third did so for about a year.
Despite the known promise of bNAbs, the pathways through which the virus escapes these antibodies remain incompletely understood across diverse HIV-1 strains.
“Knowing how different strains of the virus respond to leading bNAb therapies will greatly improve our ability to anticipate whether a particular therapy will be effective for individual patients,” says Paul Bieniasz, PhD, professor at The Rockefeller University and an HHMI Investigator. “And if we can identify broadly neutralizing antibodies that the majority of strains have great difficulty escaping from, we can create more robust treatments.”
Now scientists have established the most comprehensive view to date of how HIV-1 can escape bNAbs. Using thousands of parallel viral selection experiments combined with bioinformatic analysis and experimental validation, the team discovered viral mutations that make HIV-1 strains resistant to two bNAbs: 3BNC117 and 10-1074.
The researchers sought to investigate the relationship between different HIV-1 strains and bNAbs collected from HIV infected persons. Only a handful of resistance mutations have been identified in a limited number of viral strains. The researchers wanted to expand that number to represent global viral diversity.
“No one has attempted to do this at such a scale before,” said Theodora Hatziioannou, PhD, research professor at The Rockefeller University.
The team developed an approach that would allow them to study the mutational pathways to escape among 15 strains of HIV-1 sourced from around the globe. The goal was to pinpoint the mutations that were contributing to each strain’s propensity to develop resistance.
“We found that most viral strains can escape bNAb neutralization, but there’s substantial variation in the likelihood that they will and the mechanisms that enable it,” says Alex Stabell, MD, PhD, an infectious disease physician and clinical scholar at The Rockefeller.
Stabell devised a pipeline that began by growing large amounts of virus in cell culture. The bulk populations were used to seed thousands of parallel selection experiments with varying concentrations of bNAbs. Viruses that were able to spread in the presence of the bNAbs were isolated and sequenced. Custom bioinformatic processing gave a list of putative resistance mutations, which were subsequently experimentally validated for each viral strain.
Using this method, called RISC (resistance identification via selection and cloning), the team found more than 100 bNAb escape mutations across the 15 viral strains tested, dramatically expanding the known number. Surprisingly, they found that in most cases, a single amino acid change may be enough to confer resistance. That turned out to be true for 12 of the 15 viruses tested against the 3BNC117 antibody and for all nine tested against 10-1074.
“It was striking that it’s actually quite easy for most HIV strains to escape these special antibodies,” Bienasz says. “But it’s not true for all strains—a handful Alex worked with needed multiple amino acid substitutions or unusual ways to replicate in order to escape.”
“The genetic barrier to resistance was higher for these viruses,” Stabell adds. “One of the goals of therapy these days is not simply to have therapies that are transiently effective, but to have this high genetic barrier.”
They also identified a surprising number of mutations occurring outside the epitope on the viral envelope recognized by bNabs that target the CD4 binding site, such as 3BNC117. (10-1074 aims for a more mutable envelope target, which may help explain why it’s easier to escape.) “These were quite prominent and unexpected,” says Hatziioannou. “No one would have predicted these would affect bNAb sensitivity.”
In the future, the team will use Stabell’s method to identify to discover resistance mutations to other bNAbs as well as to combinations of them.
“HIV-1 mutates so fast and the diversity in the population is already quite enormous, so we’ve long known that a multidrug approach is the best course of treatment,” Hatziioannou says. “We hope to identify combinations that potentially raise the genetic barrier to resistance and are therefore more effective.”
Five factors predicting death or serious complications in hypertrophic cardiomyopathy, a heart condition where the heart muscle becomes abnormally thick, have been uncovered in a study led by the University of Virginia.
“Hypertrophic cardiomyopathy, with a prevalence of one in 500 in the U.S., is the most frequent cause of sudden cardiac death in young individuals,” explain lead author Christopher Kramer, MD, a researcher at University of Virginia Health, in JAMA.
“Although some patients remain asymptomatic, others develop effort intolerance, exertional angina, progressive heart failure, atrial and ventricular arrhythmias, and sudden cardiac death.”
There is some disagreement about how best to predict risk in patients diagnosed with this condition, which is inherited in 60% of cases, with different factors used for assessment in different places and current guidelines focusing on sudden cardiac death risk and not other serious adverse events such as the risk for heart failure.
This study enrolled 2,698 hypertrophic cardiomyopathy patients from 44 sites across North America and Europe between 2014 and 2017 and followed them for an average of 6.9 years. The participants underwent wide ranging tests on enrollment including cardiac magnetic resonance imaging with core laboratory analysis, genetic testing of 36 cardiomyopathy genes, blood biomarker analysis, and detailed clinical assessments. Patients with pre-existing implantable cardioverter defibrillators, often prescribed to patients with this condition to avert sudden cardiac death, were excluded.
Patients were reviewed once a year by telephone, with an average follow-up time of around seven years. Records were reviewed if events occurred during the study.
Overall, 117 events—death, nonfatal sustained ventricular arrhythmias requiring cardioversion or defibrillation, left ventricular assist device implant or heart transplant—occurred in 104 participants during the follow up period.
Five factors were significant predictors of a poor outcome. These included the extent of scarring on the heart measured by imaging, heart muscle size, and heart chamber size with all three predicting worse outcomes with greater measures. The other two factors were history of heart failure and higher levels of a blood protein marker of heart stress, NT-proBNP.
“Current risk prediction guidelines for hypertrophic cardiomyopathy are imperfect, as they predict only sudden cardiac death, and not heart failure or other fatal and nonfatal cardiac adverse events,” said Kramer in a press statement. “This study is a major advance in that it provides evidence that incorporating these additional assessment methods better predicts risk of adverse outcomes.”
The team plan to continue this work and to develop a risk score as well as to seek external validation from independent databases and researchers using similar measures of risk.
Researchers at Garvan Institute of Medical Research have identified a previously underappreciated mechanism that may explain why some breast cancers return many years, even decades, after apparently successful treatment.
The study, published in Nature Communications, reveals that certain estrogen receptor-positive (ER+) breast cancer cells survive therapy not by entering complete dormancy, but by continuing to divide at an extraordinarily slow pace. These stealth-like cells can gradually form microscopic secondary tumors that remain undetectable for years before eventually triggering metastatic relapse.
The findings offer new insight into one of the most persistent challenges in breast cancer care: why relapse can occur long after patients are considered cancer-free.
The long shadow of ER-positive breast cancer
ER-positive breast cancer is the most common subtype of breast cancer and is typically treated with hormone therapies designed to block estrogen signaling. These treatments are often highly effective at eliminating actively dividing tumor cells.
However, ER-positive disease has a unique clinical problem: recurrence risk persists for decades.
Even after five to ten years of endocrine therapy, up to 30% of patients can eventually develop metastatic relapse. Once breast cancer spreads to distant organs such as bone, lung, or brain, the disease becomes largely incurable.
Traditionally, relapse has been attributed to dormant cancer cells—cells that enter a state of complete hibernation before later “waking up.” But the new study suggests this may not be the only pathway.
“We have become very good at treating primary breast cancer, but late relapses remain a major challenge,” said Liz Caldon, associate professor and senior author of the study.
Not dormant—just incredibly slow
The researchers discovered that some breast cancer cells never fully stop proliferating during therapy. Instead, they survive by drastically slowing their rate of division.
This subtle distinction may be clinically critical.
Rather than entering complete cellular arrest, these cells continue to grow at an almost imperceptible pace, allowing them to evade therapies that primarily target rapidly dividing cells.
“Instead, they survive by growing extremely slowly in the background, until a tiny speck becomes a pebble,” Caldon explained.
Over many years, these microscopic lesions, known as micrometastases, can gradually expand until they become clinically detectable or disrupt vital organs.
The work challenges a long-standing binary view of cancer persistence in which tumor cells are considered either actively proliferating or fully dormant. Instead, the findings support the existence of an intermediate “slow-cycling” state that may be particularly effective at evading treatment.
Isolating the slowest cancer cells
Studying these rare cells was technically difficult because of their exceptionally slow growth.
The research team spent years isolating and cultivating these populations in the laboratory. Once established, they introduced the cells into preclinical models to determine whether slow proliferation impaired metastatic potential.
It did not.
Despite dividing slowly, the cells retained the ability to migrate throughout the body and colonize distant organs such as bone and lung.
“It took years to isolate these specific cells because they were dividing so slowly, almost in defiance of how we typically expect cancer to behave,” said Kristine Fernandez, first author of the study.
“These cells were migrating to organs like the bone and lungs, proving that speed isn’t everything when it comes to metastasis.”
The findings reinforce a growing understanding in oncology that aggressive cancer behavior is not solely defined by rapid proliferation. Cellular adaptability and survival under therapeutic pressure may be equally important.
Rac1 emerges as a potential therapeutic target
After identifying the slow-growing cells, the researchers investigated what allowed them to survive.
The study pinpointed a signaling pathway centered on Rac1, a protein involved in cell movement, structural organization, and survival. Using advanced biosensor imaging, the team directly visualized Rac1 pathway activation inside live slow-growing cancer cells.
Inhibiting this pathway appeared therapeutically promising.
Experimental Rac1 inhibitors significantly reduced tumor size and tumor number in patient-derived breast cancer models.
This suggests that targeting Rac1-dependent survival programs could potentially eliminate slow-growing residual cancer cells before they evolve into clinically significant metastases.
Rethinking cancer relapse biology
The findings contribute to a broader shift in cancer biology away from viewing residual disease as uniformly dormant.
Instead, tumors may contain multiple survival states, including cells that persist through continuous but ultra-slow proliferation. These populations may be especially dangerous because they remain biologically active while escaping conventional therapeutic detection.
The work also raises important clinical questions about long-term endocrine therapy. Current treatment durations are largely standardized, yet some patients may harbor persistent slow-cycling tumor cells despite years of therapy.
“If we can understand the specific biology of these slow-growing cells, we might eventually be able to offer better ways to track whether a decade of hormone therapy is actually working and ultimately prevent recurrence,” Caldon said.
Toward preventing late relapse
The study’s implications extend beyond breast cancer alone. Slow-cycling drug-tolerant cancer cells have increasingly been identified across multiple tumor types, including melanoma, lung cancer, and leukemia.
By identifying a concrete signaling mechanism underlying this state in ER-positive breast cancer, the research provides a potential therapeutic entry point for preventing relapse before metastatic disease emerges.
The next challenge will be determining whether Rac1 inhibitors, or similar approaches targeting slow-cycling survival programs, can safely and effectively eliminate residual cancer cells in patients.
If successful, such strategies could fundamentally alter how clinicians approach long-term relapse prevention in breast cancer, shifting the focus from simply suppressing visible disease to actively eradicating the hidden cellular reservoirs that remain years after treatment ends.
Researchers at the Medical University of South Carolina (MUSC) have found evidence that the asthma medication formoterol may reverse metabolic dysfunction-associated steatohepatitis (MASH), a progressive fatty liver disease associated with obesity and type 2 diabetes that can lead to fibrosis, cirrhosis, liver failure, and liver transplantation. The research, published in npj Metabolic Health and Disease, arose unexpectedly as a result of findings on the use of formoterol in mouse models of diabetic kidney injury, which also showed that the mice had low levels of liver fat accumulation.
“Kind of unexpectedly, we found that the liver damage also reversed,” said senior author Joshua Lipschutz, MD, division director of nephrology and Arthur Williams Endowed Chair in nephrology at MUSC.
Based on this observation, the MUSC researchers initiated a study to find out whether the beta-2 adrenergic receptor pathway targeted by formoterol could influence metabolic disease in the liver as well as the kidney. According to the researchers, the connection between the diseases lies in shared metabolic dysfunction associated with type 2 diabetes relating to mitochondrial dysfunction and impaired energy metabolism.
To test the hypothesis, the team used a high-fat diet mouse model designed to mimic MASH. Mice fed the diet for 16 weeks developed liver steatosis and were subsequently treated with formoterol for four weeks. Testing of the mice after the four weeks of treatment found that steatosis was largely resolved as a result.
The evidence showed that formoterol increased mitochondrial biogenesis, a process that increases the number and function of mitochondria within cells.
“It looked like formoterol was rescuing the injury by increasing mitochondrial biogenesis,” Lischutz said. “It kind of revs up the mitochondria so they work better.”
The researcher noted that mice treated with formoterol had increased levels of PGC1α (a protein that helps control how cells produce and use energy) and electron transport chain proteins, along with an increase in mitochondrial proteins and lower lipid accumulation in liver tissue. Human HepaRG liver cells exposed to free fatty acids also showed reduced lipid accumulation and increased after formoterol treatment.
“The coordinated induction of oxidative phosphorylation and amino acid metabolism pathways suggests that formoterol may promote metabolic competence through non-lipid sources, including amino acids,” the researchers wrote.
While there were no approved drugs to treat MASH when the MUSC researchers initiated their study, current treatments still remain limited with resmetirom and semaglutide the only current approved therapies for this condition. Both medications have shown only limited efficacy in a subset of patients and have known side effects.
“All the current drugs for diabetic nephropathy only slow progression, but they don’t reverse the damage. This drug actually reversed the damage at the histologic, ultrastructural, and functional levels,” said Lipschutz.
Further, formoterol is already an approved and established medication that has been prescribed for year to treat both asthma and chronic obstructive pulmonary disease (COPD). Because its metabolic effects in humans and its safety profile has been detailed in its approval for these conditions, it could hasten approval for these other therapeutic uses.
“If you can repurpose something that’s approved and already being used safely, that’s kind of our dream as physician-scientists,” Lipschutz added.
Lipschutz and colleagues are currently conducting a clinical trial for the use of formoterol in chronic kidney disease (NCT07022418). Future research will focus on what dosing levels would be appropriate to use as treatment for CKD and MASH, whether inhaled delivery would be effective, and how durable the response to this potential treatment could be.
This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here.
A few months before he was awarded the Nobel Prize in economics in 2024, Daron Acemoglu published a paper that earned him few fans in Silicon Valley. Contrary to what Big Tech CEOs had been promising—an overhaul of all white-collar work—Acemoglu estimated that AI would give only a small boost to US productivity and would not obviate the need for human work. It’s okay at automating certain tasks, he wrote, but some jobs will be perfectly fine.
Two years later, Acemoglu’s measured take has not caught on. Chatter about an AI jobs apocalypse pops up everywhere from Senator Bernie Sanders’s rallies to conversations I overhear in line at the grocery store. Some previously skeptical economists have gotten more open to the idea that something seismic could be coming with AI. A California gubernatorial candidate said last week that he wants to tax corporate AI use and pay victims of “AI-driven layoffs.”
On the one hand, the data is still on Acemoglu’s side; studies repeatedly find that AI is not affecting employment rates or layoffs. But the technology has advanced quite a bit since his cautious predictions. I spoke with him to understand if any of the latest developments in AI have changed his thesis, and to find out what does worry him these days if not imminent AGI.
AI agents
One of the biggest technical leaps in AI since Acemoglu’s paper has been agentic AI, or tools that can go beyond chatbots and operate on their own to complete the goal you give them. Because they can work independently rather than just answering questions, companies are increasingly pitching agents as a one-to-many replacement for human workers.
“I think that’s just a losing proposition,” Acemoglu says. He thinks agents are better thought of as tools to augment particular pieces of someone’s work than something malleable enough to handle a person’s whole job.
One reason has to do with all the various tasks that go into a job, something Acemoglu has been researching in his work on AI since 2018. For example, an x-ray technician juggles 30 different tasks, from taking down patient histories to organizing archives of mammogram images. A worker can naturally switch between formats, databases, and working styles to do this, Acemoglu says, but how many individual tools or protocols would an AI require to do the same?
Whether or not agents will supercharge AI’s impact on jobs will come down to whether they can eventually handle the orchestration between tasks that humans do naturally. AI companies are in heated competition to prove that their AI agents can work independently for ever longer periods without making mistakes, sometimes exaggerating the results—but Acemoglu says many jobs will be spared from an AI takeover if agents can’t fluidly switch between tasks.
The new hiring spree
For years Big Tech has been offering staggering salaries to recruit AI researchers. But I asked Acemoglu about a different hiring spree I’ve noticed: AI companies are all building in-house economics teams.
OpenAI hired Ronnie Chatterji from Duke University in 2024 to be its chief economist and announced last year that Chatterji will work with Jason Furman—Harvard economist and former advisor to Barack Obama—to research AI and jobs. Anthropic has convened a group of 10 leading economists to do similar work. And just last week, Google DeepMind announced it had hired Alex Imas, an economist from the University of Chicago, to be its “director of AGI economics.”
Acemoglu has noticed colleagues getting snatched up for these roles too. “It makes sense,” he says: AI companies are well aware that public skepticism about AI, in large part due to job concerns, is growing. And they have strong incentives to shape the economic narrative around their technology (consider OpenAI’s latest proposal for a new era of industrial policy).
“What I hope we won’t get,” Acemoglu says, “is that they’re interested in economists just to further their viewpoints or further the hype.” That tension hangs over the emerging field of “AI economics”; it’s concerning that some of the most influential research about AI’s impact on work may increasingly come from the companies with the most to gain from favorable conclusions.
AI apps
I don’t think of AI as hard to use; most of us interact with it via chatbots that use plain language. But Acemoglu says we should consider how it compares with the sort of software that kicked off earlier tech transformations, like PowerPoint for slide decks and Word for documents.
“Anybody could install these on their computer and get them to do the things that they want them to do,” he says. They spread accordingly.
“We have not seen the development of apps based on AI that have the same usability,” he says. Even if anyone can chat with an AI model, it tends to take a while for the average worker to get practical and productive use out of it. That’s part of the reason why AI has not yet shown any seismic impact on the job market or the economy. One of the key signals Acemoglu is watching, then, is the creation of apps that make AI easier to use.
But he acknowledges that for a while, we’re going to see all sorts of conflicting evidence about AI: anecdotes that college grads are finding the job market worse and worse, but no noticeable effect of AI on productivity, for example. “There’s a huge amount of uncertainty,” he says. And that’s the most telling thing about the AI economy right now: the certainty of the rhetoric alongside the uncertainty of everything else.
Scientists at Columbia University have developed a brain-controlled hearing technology that allows users to amplify the conversation they are focusing on while reducing other voices. Published today in Nature Neuroscience, this study marks the first time this kind of technology has been tested in humans.
“We have developed a system that acts as a neural extension of the user, leveraging the brain’s natural ability to filter through all the sounds in a complex environment to dynamically isolate the specific conversation they wish to hear,” said Nima Mesgarani, PhD, principal investigator at Columbia’s Zuckerman Institute and associate professor of electrical engineering at Columbia’s Fu Foundation School of Engineering and Applied Science. “This science empowers us to think beyond traditional hearing aids, which simply amplify sound, toward a future where technology can restore the sophisticated, selective hearing of the human brain.”
While modern hearing aids can amplify human speech while suppressing background noise, they cannot separate and enhance specific voices when multiple people are speaking. This can make it difficult for users to concentrate on a specific conversation in everyday scenarios such as restaurants, classrooms, busy workplaces, and family gatherings.
The hearing device developed by Mesgarani’s team mimics the way the human brain can naturally identify and focus on a single speaker out of many within a crowd. Previously, the researchers had found a way of identifying which brain signals are linked to a specific conversation, by matching the timing of peaks and valleys of the brain waves to the sounds and silences of that conversation. They also identified distinct patterns of brain activity that indicate which conversation a person is focusing on and which one they are filtering out.
In the current study, the scientists developed a machine learning algorithm that could examine the user’s brainwaves and identify which conversation they are paying attention to in real time, making that voice louder and others quieter to make it easier to listen to. This system was tested on epilepsy patients who already had electrodes implanted in their brains. The electrodes were used to measure the user’s brain activity as they focused on two overlapping conversations played simultaneously, and the algorithm automatically detected which conversation they were trying to focus on.
“The results mark an important step toward a new generation of brain-controlled hearing technologies that align with the listener’s intent, potentially transforming how people navigate noisy, multi-talker environments,” said Vishal Choudhari, PhD, who led the development and evaluation of the system.
More research will be needed before minimally invasive wearable systems can integrate this kind of brain sensing technology with advanced audio processing capabilities, especially to ensure they can accurately decode conversations in real time and in real-world scenarios where multiple voices can be heard.
“The central unanswered question was whether brain-controlled hearing technology could move beyond incremental advances, towards a prototype that could help someone hear better in real time,” said Choudhari. “For the first time, we have shown that such a system that reads brain signals to selectively enhance conversations can provide a clear real-time benefit. This moves brain-controlled hearing from theory toward practical application.”
In a new study published in Cell titled, “RegVelo: gene-regulatory-informed dynamics of single cells,” researchers from Stowers Institute of Medical Research have developed a new AI model that connects two areas of single-cell biology that have often remained separate: estimating how cells change over time and inferring the gene regulatory networks controlling those changes.
“You can imagine if you had a very early set of cells, having a particular set of instructions could allow you to reproduce, in vitro, some of these cell types in a very natural way. These cells could then be used in cell therapies in regenerative medicine,” said Tatjana Sauka-Spengler, PhD, Stowers Institute Investigator and co-senior author of the study.
While development is often described as a series of static snapshots of cell states, RegVelo models how these fate decisions are encoded in gene regulatory networks over time and space, and what drives cell state transitions. In zebrafish neural crest development, RegVelo identified an early driver of pigment cell formation (tfec) and revealed a previously unknown regulator of pigment cell fate (elf1). The neural crest is a developmental system that gives rise to many different cell types, including pigment cells, craniofacial tissues, and parts of the peripheral nervous system.
CRISPR/Cas9-mediated knockout and single-cell Perturb-seq supported predictions, showing that the model could do more than describe developmental changes and generate biologically meaningful hypotheses that held up in living systems.
Alejandro Sánchez Alvarado, PhD, Stowers President and chief scientific officer says RegVelo’s value “extends well beyond” neural crest cells and is applicable to any system in which cells change over time, from basic developmental biology to modeling tumor trajectories and the cellular outcomes that may inform treatment.
“Sauka-Spengler and her collaborators have developed a meaningfully different way to process this kind of data,” said Sánchez Alvarado. “It allows us to infer the most likely path of each component through space and time, and to use deep learning to predict those dynamics and test them experimentally.”
Single-cell biology research has made it possible to build increasingly detailed maps of development. RNA velocity methods can help researchers estimate how cells move through developmental landscapes, while gene regulatory network approaches can identify relationships among genes. However, these methods have typically been used in parallel rather than together.
“For a long time, cellular dynamics and gene regulation have largely been modeled separately,” said Fabian Theis, PhD, the study’s co-senior author and director of the institute of computational biology at Helmholtz Munich. “RegVelo brings those pieces together, allowing us to ask not only how cells are changing, but which regulatory interactions are helping drive those changes.”
The framework jointly models splicing kinetics and gene regulatory relationships, allowing researchers to map the hidden timeline of cell development, predict how cells shift from one state to another, and test what might happen when specific regulators are perturbed.
The framework can incorporate additional regulatory layers, including chromatin, protein activity, and other multimodal measurements. While the study’s limitations include simplifying assumptions around latent time, regulatory interactions, and computational cost, the results demonstrate a compelling proof of principle.
“When dynamic cell-state modeling is linked directly to gene regulation, it becomes possible to move closer to mechanism and then discovery,” Sauka-Spengler said.
Despite years of digitization, organizations capture less than one-third of the value expected from digital investments, according to McKinsey research. That’s because most big companies begin with technological capabilities and bolt applications onto them, rather than starting with customer needs and working backward to technology solutions. Not prioritizing the customer can create fragmented solutions; disjointed customer experiences; and ultimately, failed transformations.
Organizations that achieve outsized results from AI flip the script. They adopt a “customer-back engineering” mindset, putting customers at the heart of technology transformation.
It’s a strategy in which products and services are developed with the customer experience first in mind, including the customers’ challenges, needs, and expectations. Product development teams then work backward in a nimble and agile way to find the steps necessary to design and build solutions that achieve the desired experience.
“When you get your engineers closer to customers, you get a lot more sideways innovation,” says Ashish Agrawal, managing vice president of business cards and payments tech at Capital One. “That leads to a multiplier effect, because engineers can approach a problem from a different dimension that can be unique to the sales or product perspective.”
The case for customer-centricity in engineering
Engineers are problem-solvers by nature, says Agrawal. When they hear about challenges customers are experiencing, or how they are using products and services in the real world, they can devise ways to efficiently address customer needs, since they are naturally closer to systems and data than many other teams across the company.
“Fostering a customer-centric culture has a motivational effect on engineers when they actually start seeing how the core changes they’re making, or the features they’re adding, are having a direct impact on the lives of customers,” says Agrawal.
It also takes discipline. Agrawal explains that Capital One has set a goal for every engineer in his organization to establish several touchpoints with customers throughout the year in different forms, including:
Digital empathy sessions to observe user journeys and identify where users hit friction
Embedded customer support for periods of time to deepen understanding of servicing needs
Engineering ride-alongs, in which engineers join customer success, sales, and support staff on calls or on-site visits
Hackathon competitions to build solutions around real customer problems
The AI opportunities with customer-centricity
“The biggest challenge engineers within large companies face is a lack of direct access to customers,” says Agrawal. “This can make it harder for technologists to work with customers to identify problems and innovate solutions.”
AI has accelerated the challenges as well as the opportunities. The lifecycle of launching products has become significantly faster. But the good news is that engineers are closer to the data that feeds into AI, so they can more rapidly apply AI-informed data techniques to solve customer problems.
Agrawal outlines a recent scenario: In the customer servicing space, conversations can instantly be summarized and give a customer agent context on the member’s original request and remaining action points. Agentic AI can also be enabled to ask pointed follow-up questions about the interaction that would otherwise take human agents time to read through the entire thread.
“A solution would have been a lot harder in an ecosystem without a lot of high-quality data,” says Agrawal. “But when you combine a rich data ecosystem with agentic tools, you move from incremental fixes to high-velocity transformation.”
By investing in AI data and tools and focusing on rapid experimentation, Agrawal says the cycle of deploying solutions can be accelerated. Teams learn that if they meet customer needs and iterate on a wider range of solutions much faster, then the entire innovation cycle speeds up.
For example, Capital One used customer insights to build a state-of-the-art, multi-agent AI framework called Chat Concierge to enhance the customer experience for car buyers and dealers. In a single conversation, Chat Concierge can perform tasks like comparing vehicles to help car buyers decide on the best choice and scheduling test drives or appointments with salespeople.
Agrawal explains that car buyers can engage with Chat Concierge directly through participating dealer websites. Dealers can access and can take over the chat through Navigator Platform. The AI assistant consists of multiple logical agents that work together to mimic human reasoning, allowing it to provide information and take action based on the customer’s requests.
The elements of an AI-first mindset
According to a recent MIT Technology Review Insights survey, 70% of leaders say their firm uses agentic AI to some degree. Roughly half of executives say agentic AI systems are highly capable of improving fraud detection (56%) and security (51%), reducing cost and increasing efficiency (41%), and improving the customer experience (41%).
Looking into the future, achieving these outcomes looks even more likely. More than half of the banking executives surveyed say they expect to continue to improve fraud detection (75%), security (64%), and the customer experience (51%). Agentic AI use cases that show strong potential to transform the customer experience in financial services include responding to customer services requests, adjusting bill payments to align with regular paychecks, or extracting key terms and conditions from financial agreements.
Placing the customer at the center of a transformation requires an AI-first mindset. Companies must shift from simply augmenting an existing product to fundamentally reimagining the problem and the user’s needs through the lens of AI’s capabilities.
A few best practices that Agrawal recommends include:
Reimagine the core function of AI to solve a user’s problem: “The true value isn’t in chasing the AI hype; it’s in solving meaningful customer problems. By focusing on impact, we ensure that our innovation isn’t just fast; it’s transformative,” says Agrawal.
Start with high-quality, well-governed data as the foundation: “Data readiness and unified information across systems are the non-negotiable foundations of AI. A clean data layer is what orchestrates the agentic loop— enabling the perception, reasoning, and execution required to solve a customer’s problem before they even have to ask,” explains Agrawal.
Rebuild workflows with AI embedded from the start: “People treat models as black boxes, but agentic systems require tremendous rigor and oversight. Having a data ecosystem that is well-governed and responsible AI standards are essential pillars for building trust in these systems,” says Agrawal.
Build a cross-functional team involving data science, engineering, product, design, and other partners: Agrawal advises, “It’s important to be open and nimble to transforming how we work and create impact as AI becomes more integrated into workflows. It’s also important to take a ‘crawl, walk, run approach’ if you are new to AI, as opposed to simply jumping into it.”
In the end, achieving end-to-end transformation depends on empowering engineers and partner teams to start with customer needs and work backward to technology solutions, rather than starting with technological capabilities first and finding applications for them. When organizations make a customer-back approach second nature, they are able to not only reimagine the customer experience from the inside out, but to also place the customer front and center from the very start.
This content was produced by Insights, the custom content arm of MIT Technology Review. It was not written by MIT Technology Review’s editorial staff. It was researched, designed, and written by human writers, editors, analysts, and illustrators. This includes the writing of surveys and collection of data for surveys. AI tools that may have been used were limited to secondary production processes that passed thorough human review.
