Using a classic visual illusion, 91��ɫ researchers identify a history-dependent computation present in primate vision that current AI models fail to reproduce.

TORONTO, Oct. 2, 2026 – Our eyes do not always tell us exactly where things are – and that may be a feature of how biological vision works, rather than simply a flaw. A new study by 91��ɫ researchers uses a common illusion to ask if artificial intelligence is meant to see more like us, should it make some of the same systematic perceptual “mistakes”?

For example, after staring at something moving steadily in one direction, a stationary object viewed immediately afterward can appear slightly displaced in the opposite direction. This well-known visual illusion, called a motion aftereffect, gives scientists an unusual window into the computations underlying perception: the image itself has not moved, but our experience of where it is has changed.

“Today’s AI vision systems are impressive, but they still do not always see the world the way we do. This study captures the promise of NeuroAI and what it can do when neuroscience and artificial intelligence are brought together. By using smart experiments to reveal the computations biological vision uses and AI still lacks, we can use those insights to build better, more brain-like artificial systems,” says senior author 91��ɫ Assistant Professor Kohitij Kar, the Canada Research Chair in Visual Neuroscience and a member of 91��ɫ’s Centre for Vision Research and Centre for Integrative and Applied Neuroscience.

Current AI vision systems can often determine where an object is accurately, but they generally do not reproduce the way recent visual experience can reshape that answer. In humans, staring at motion can make a subsequently viewed stationary object appear displaced even though its pixels have not moved. The researchers found a corresponding change in the primate visual cortex – but not in the AI models they tested.

To gain better insight into the issue, the researchers examined whether artificial neural network models capture the same history-dependent changes in spatial representations seen in biological vision, or whether their position representations primarily reflect the physical properties of the image. The researchers, including the paper’s first author and 91��ɫ graduate student Elizaveta Yakubovskaya, used precise measurements to find out where AI differs from biological vision and how that gap could be bridged.

“Combining recordings from primate visual cortex with human perception experiments, we used motion adaptation to induce a visual illusion and make a stationary object appear slightly shifted in position, then asked whether the brain and AI showed the same effect. Human observers reported the illusion, and neural representations of position in the primate inferior temporal (IT) cortex shifted in the same direction, even though the image itself had not changed,” says Yakubovskaya.

The researchers leveraged motion adaptation to show where perceived and pixel-based positions diverge, allowing them to test the behavioral relevance of IT codes. The findings not only further our understanding of IT’s role in spatial information encoding but provide a new benchmark to evaluate dynamic vision models.

“There is a growing question in AI about whether increasingly capable systems will become more like us or increasingly different from us,” says Kar of the Faculty of Science and a member of the 91��ɫ-led Connected Minds. “If we want AI that works with humans and understands the world in more human-compatible ways, we cannot focus only on whether it gets the right answer. We also need to understand the computations that produce human perception and behavior. Neuroscience gives us a way to discover those computations and, potentially, build them into AI,” adds Kar.

The study titled, , is published today in Current Biology.

The post Visual illusion reveals what today’s AI vision is missing – 91��ɫ study appeared first on News@91��ɫ.

TORONTO, ON, July 28, 2026 – 91��ɫ’s Connected Minds: Neural and Machine Systems for a Healthy, Just Society and the (CIGI) are launching a 16-month pilot partnership to design and build new governance frameworks for artificial intelligence (AI) and neurotechnology.

The initiative combines Connected Minds’ university-based interdisciplinary research excellence with CIGI’s applied policy expertise to deliver a coordinated, multi-level engagement model that connects United Nations standard setting, Canadian legislative processes, and applied regulatory guidance.

“Intelligent technologies, including artificial intelligence and neurotechnologies, are transforming society at an unprecedented pace, making it more important than ever that public policy is informed by rigorous, interdisciplinary research,” says Professor Shayna Rosenbaum, scientific director of Connected Minds. “Our partnership with CIGI creates a bridge between leading researchers and policy experts, helping to ensure that technological innovation advances in ways that strengthen society, protect human rights, and contribute to a healthy, just future.”

The initiative will help domestic and international legislators better address critical gaps in human rights laws created by rapidly advancing technologies. Without coordinated action, individuals, businesses, and governments face growing legal and ethical uncertainty, eroded public trust, and a decline in privacy.

“Privacy is a fundamental human right,” says Aaron Shull, research director, digitalization, security & democracy at CIGI. “Canada has an opportunity and a responsibility to lead globally in creating and shaping safeguards needed to protect human rights and strengthen public trust in emerging technologies.”

Through a phased approach, this partnership will:

• establish a replicable framework for similar technology and human rights initiatives in other domains; 
• position Canada as a global leader in the responsible governance of neurotechnology and AI; and
• produce tangible policy outputs and communications tools with immediate relevance to global policymakers, regulators, and industry leaders.

About Connected Minds:

Connected Minds is a visionary, first-of-its-kind research program led by 91��ɫ in partnership with Queen’s University. Supported by $318 million in funding, including $105.7 million from the Canada First Research Excellence Fund (CFREF), it is the largest 91��ɫ-led research initiative to date.

Bringing together 300+ scholars across fields including science, arts, neuroscience, social sciences, law, and artificial intelligence, Connected Minds explores the risks and benefits of modern technology on society, both now and in the future. With more than 90 partners and collaborators from diverse sectors, the program aims to create a healthier, more just society through interdisciplinary collaboration and innovation.

About CIGI:

The Centre for International Governance Innovation (CIGI) is an independent, non-partisan think tank whose peer-reviewed research, foresight and trusted analysis influence policy makers to innovate. With the engagement of a global network of experts and contributors, CIGI tackles the governance challenges and opportunities of data and transformative technologies, including AI, and their impact on the economy, security, democracy and, ultimately, societies. For more information, please visit .

Media contacts:

Sandra McLean
91��ɫ
sandramc@yorku.ca

Richia McCutcheon
Centre for International Governance Innovation
226-988-2951
rmccutcheon@cigionline.org

The post 91��ɫ’s Connected Minds and CIGI partner to strengthen global AI and neurotechnology governance appeared first on News@91��ɫ.

The wiring and rewiring of the brain never ends. Neural pathways are constantly being reshaped as we interact with the world and learn new things. At 91��ɫ and MIT’s McGovern Institute, scientists are combining detailed analysis of brain activity with computational modelling to better understand that change.

91��ɫ Assistant Professor , McGovern Institute Postdoctoral Fellow , an affiliate member of Kar’s lab and a member of the Centre for Integrative and Applied Neuroscience at 91��ɫ, and Investigator  worked together to compare what happened when monkeys and an artificial neural network with brain-like architecture were trained to visually identify the same objects. As the model’s performance improved, it reorganized itself in ways that closely paralleled changes the team detected in the brains of monkeys.

Their research, reported  , shows how changes in visual processing support animals’ ability to learn to discriminate new kinds of objects. By modelling these changes, the researchers hope to better predict how training reshapes perception, which could one day inform educational strategies for a wide range of learners.

“Our visual brain does not undergo a drastic reconfiguration when we learn a new object. Instead, it subtly reshapes how visual information is represented, making the distinctions that matter easier for the rest of the brain to read out. What is powerful about this study is that artificial neural networks could predict these subtle brain changes, giving us a concrete bridge between neuroscience and AI,” says Kar. “By translating inferences from animal neuroscience into computational models that can generate human-testable predictions, this framework makes translation much more feasible. It allows us to move from understanding how learning changes the macaque visual system to perhaps asking in the near future: why some children may struggle to form stable, generalizable representations – such as recognizing letters across fonts, linking words to objects, distinguishing faces, or applying knowledge in new situations.”

Subtle changes

Learning about a new object calls on many parts of the brain. Visual-processing areas work together to make sense of information taken in through the eyes, then communicate with other brain areas to give the visual information meaning and guide behavior. Multiple parts of this system likely change during learning, and the research team wanted a clearer understanding of how that change is distributed.

Neuroscientists have debated how much change occurs in the brain’s visual-processing areas when an animal learns to recognize new objects. Some suspected that visual-processing pathways remain largely unchanged during learning to avoid broadly disrupting visual perception, but others have reported changes in activity within dedicated visual-processing areas with this kind of learning in humans and other primates. 

To take a closer look, the team focused on neural activity in a key component of the brain’s visual object-processing network, the inferior temporal (IT) cortex. By the time visual information reaches the IT cortex, key object features are clearly represented – so much so that it’s possible to “decode” what object a monkey is seeing and even predict what errors it’s likely to make in identifying it, simply by analyzing patterns of neural activity there.

The team recorded neural activity in the IT cortex from two groups of monkeys as the animals looked at and identified images of objects. Some of the monkeys were untrained, so the images they saw had little meaning to them. Others had already learned to identify similar objects, so they could usually discriminate between elephants, chairs and other select objects, even when those objects were presented at different sizes, from different angles, or against different backgrounds than the ones they had seen before.

The broad pattern of activity in the IT cortex was largely similar in trained and untrained monkeys, suggesting that learning had not dramatically rewritten this high-level visual representation. Still, the group found subtle but reliable differences in the way neurons in the IT cortex responded to images in monkeys that had learned to recognize the kinds of objects they were shown, compared to the untrained monkeys.

Modelling learning

The group turned to computational models to investigate how those modest changes might contribute to learning. Sörensen trained a suite of artificial neural networks whose internal components had been mapped to monkey IT cortex to identify the same categories of objects the monkeys had seen. The models were designed to learn using gradient descent, meaning they continually improved their accuracy by adjusting their parameters in response to errors.

Only some of the primate-like models showed learning behavior that matched that of the monkeys. In those that did, the IT-like stage changed in ways that resembled the learning-related changes the researchers had observed in the IT cortex of trained monkeys.

While gradient descent is commonly used to train artificial intelligence, it is generally considered biologically implausible as a direct model of how the brain learns. The researchers say the strong match in learning effects between the monkeys and their model demonstrates that these kinds of artificial neural networks can offer insights into biological learning at a useful level of abstraction, even if the brain does not learn in the same way.

“This shows that you can actually build in silico versions of future experiments,” Sörensen says. “I think that gives us this playground of asking ‘what if’ questions – and potentially predicting new things that go beyond the experimenter’s intuition.”

Most of the changes that allowed for learning in the model occurred outside of the IT cortex. “This tells us that there is a lot between the area we recorded from and the final behavioral readout that needs to change during this process,” Kar says. He adds that the team’s model will be useful as researchers look more deeply into how downstream brain areas contribute to learning.

The researchers stress that their study allowed more granular measurements of brain activity than would be possible in humans, and because monkeys’ brains are organized similarly to our own, their experiments have direct relevance to human learning. They say understanding the impact of plasticity in monkeys’ IT cortex could help researchers design new learning strategies for humans.

“Our prior conceptual working model of you – or a monkey – learning new objects was that your brain makes changes to synaptic connections that are largely downstream of your visual system, so you don’t destroy your visual system,” says DiCarlo, who is also the Peter de Florez Professor of Brain and Cognitive Sciences and director of the MIT Siegel Family Quest for Intelligence. “You wouldn’t want your whole visual system to become an elephant detector [just because you’ve learned to identify an elephant]. But this study went beyond that to say actually, when you learn ‘elephant,’ your IT does change a little bit to make it a little more relevant to elephants.”

That likely has consequences for recognizing other visual features, too. Subtle changes in the IT cortex that support elephant recognition might also make you better at identifying things other than elephants, DiCarlo says. Likewise, the same changes might make it a little harder to identify something else.

These kinds of consequences may be difficult to predict intuitively, but become obvious with computational modelling. For instance, the team’s models revealed that after learning to recognize new objects, the IT cortex contained more information about objects’ locations. By providing insights like these, models could aid the design of more effective training strategies for visual tasks, including for people with altered sensory processing, who may learn from visual information in atypical ways.

With files from MIT's McGovern Institute

The post Learning to identify new objects reshapes parts of the brain, research finds appeared first on News@91��ɫ.

New national initiative to instruct and mentor 91 highly qualified trainees in equitable, data-driven, public health decision-making

91��ɫ Distinguished Research Professor is spearheading a national initiative to ensure Canada remains a world leader in using artificial intelligence (AI) and mathematical modelling to advance health equity nationally and globally, and protect vulnerable communities.

Supported by a recent $1.65 million grant and bolstered by about $3.65 million in partner contributions, the Mathematical Innovations for Precision Public Health (CREATE-MIPPH) program will train 91 undergraduates, master’s and doctoral trainees over six years to transform massive volumes of complex data into actionable, life-saving public health strategies.

The program responds to a growing challenge in public health: a critical shortage of professionals trained to integrate an increasingly huge amount of health, behavioural, environmental and pathogen data into actionable strategies for disease prevention, health-care planning and emergency preparedness.

“Canada has extraordinary data resources and strong public health institutions, but data alone do not improve outcomes,” says Wu. “We need highly trained experts who can integrate AI, mathematical modelling and established decision-making systems to help governments, healthcare systems and industry respond more effectively to emerging health challenges.”

Supported by a network of researchers from six Canadian universities and partnerships with government agencies, hospitals, health technology companies and pharmaceutical organizations, trainees will gain hands-on experience through research placements, interdisciplinary collaborations and real-world projects. The program will equip trainees with expertise in data science, mathematical modelling, AI-driven data analytics and health economics. Professional and leadership development will focus on science communication, policy engagement and knowledge mobilization.

Grounded in team-based science and equity-focused practice, research will focus on quantifying emerging diseases, the influence of behavioural responses and socioeconomic factors on outbreaks and optimizing clinical trials. Going beyond technical modelling, trainees will examine how environmental factors, systemic barriers, individual and collective behaviours and public health interventions shape health outcomes. 

“This program is about building a new generation of highly skilled scientists and professionals who can bridge advanced analytics with public health needs. Our trainees will not only develop technical expertise, but also learn how to work across disciplines and sectors to translate complex data into evidence-based decisions that improve health outcomes and strengthen preparedness,” says , a co-applicant and professor of computational epidemiology and vaccine science at 91��ɫ.

CREATE-MIPPH will strengthen Canada’s emergency preparedness capacity by combining behavioural science, modelling and simulation-based training. The program will equip graduates with the crisis communication skills required to lead multi-agency coordination and advise decision-makers during an outbreak, while fostering inclusive, equitable environments. As the trainees enter the workforce, they will serve as a valuable national resource linking academic innovation directly to the front lines of public health.

Other co-applicants from 91��ɫ include , professor of disaster and emergency management, director of CIFAL 91��ɫ and executive director of the Victor Phillip Dahdaleh Advanced Disaster, Emergency and Rapid Response Simulation (ADERSIM); Shayna Rosenbaum, distinguished research professor, psychology and principal investigator of the Rosenbaum Memory Lab; and Woldegebriel Assefa Woldegerima, assistant professor of mathematical biology. Co-applicant institutions include University of Guelph, University of Toronto, Western University, University of Saskatchewan and University of Regina.

About Jianhong Wu

A 91��ɫ Research Chair in industrial and applied mathematics, Jianhong Wu is recognized for his fundamental and applied research in disease modelling, with leading roles in high-impact interdisciplinary projects, including the National COVID-19 Modelling Rapid Response Task Force, the NSERC-EIDM network of MfPH and the ADERSIM facility. He is an elected fellow of the Royal Society of Canada and a fellow of the Canadian Academy of Health Sciences. His efforts in building collaborative capacity across industry, government and academia have been recognized by the CAIMS-Fields Industrial Mathematics Prize (2019) and the NSERC Synergy Award (2024).

The post 91��ɫ profs awarded $1.65M NSERC CREATE grant to train the next generation of experts in precision public health appeared first on News@91��ɫ.

The new Centre of Excellence will bring together multiple disciplines across the University to develop and deploy artificial intelligence systems to improve health care

TORONTO, April 9, 2026 – As artificial intelligence (AI) becomes increasingly important, especially in the health-care field, 91��ɫ continues to play an outsized role. Today, the University announced the launch of a new Centre of Excellence – Artificial Intelligence for Public Health Advancement (AIPHA) funded through an Ontario Research Fund – Research Excellence program.

Director General, Applied Public Health Sciences Pamela Ponic of the Science and Policy Integration Branch, Public Health Agency of Canada, Chief Medical Officer of Health for the Ontario Ministry of Health Kieran Moore, and MPP for Whitby Lorne Coe, Parliamentary Assistant to the Minister of Colleges, Universities, Research Excellence and Security all spoke at the lunch-time event.

As a key player in supporting health-care decision making, AIPHA will strengthen external partnerships and accelerate the transfer of knowledge from research to policy and practice, where hospitals, medical practitioners, policymakers and leaders can use it. It will also help bridge the gap between health analytics and real-world socioeconomic conditions further positioning 91��ɫ as a national and global leader in AI-integrated public health solutions through research and innovation.

"The launch of AIPHA marks a defining moment for 91��ɫ and for the future of public health in Canada. By bringing together expertise across disciplines, from mathematical modelling and AI to health policy and social equity, we are creating something truly transformative: a hub where research doesn't just advance knowledge but directly shapes the decisions that protect and improve people's lives,” says 91��ɫ Interim President and Vice-Chancellor Lisa Philipps. “91��ɫ has long been committed to addressing society's most pressing challenges, and AIPHA reflects that mission at its fullest. We are proud to be building the next generation of AI-adept public health leaders right here, and to be positioning Canada as a global force in equitable, evidence-informed health innovation."

As a dedicated, multi-disciplinary and national hub, AIPHA will bring together expertise from across faculties, including in advanced mathematical and computational modelling, precision analytics and multi-source databases. The goal is to integrate epidemiological, clinical, environmental, climate and socioeconomic indicators in newly created AI models, while training the next generation.

“Artificial intelligence tools are increasingly being used in health-care settings but a coordinated, ethical and equitable approach to ensure the tools use integrated data sources, and that they are being properly tested and deployed for patient good is lacking. The current speed of newly developed AI models is at times outpacing governance,” says Faculty of Science Dean Maydianne Andrade. “As a new Centre of Excellence in the Faculty of Science, AIPHA will lead the way toward better integration of these new technologies in a cohesive manner that will help advance public health care.”

Two projects already underway include: Integrating AI with disease transmission dynamics models for informed prevention and control of outbreaks in indoor and mass gathering settings (2025 to 2031) and Advanced mathematical technologies for respiratory infection risk assessment and pharmaceutical intervention scenario analysis (2024 to 2028), both led by AIPHA’s inaugural director Jianhong Wu.

“The AI for Public Health Research Centre is a coordinated innovation hub that will help improve health-care efficiency and outcomes, as well as ensure coordinated, ethical and equitable transformation of public health systems,” says AIPHA Director and University Distinguished Research Professor Jianhong Wu of the Faculty of Science. “This kind of central hub is much needed in the health-care sector today to ensure emerging AI tools are properly integrated and decision and policy makers are provided with robust information toward developing a more cohesive, ethical and equitable public health-care system.”

AIPHA will integrate epidemiological, clinical, environmental and socioeconomic data into the AI-enabled decision-support systems it develops and deploys to guide equitable and evidence-informed public health action. In ensuring the development of fair and equitable AI systems, the hub will combine not only advanced mathematical and computational modelling and AI and predictive analytics, but also health systems and policy research with social determinants of health and equity frameworks.

AIPHA will act as a research accelerator for large collaborative grants, train the next generation of AI-adept public health leaders and develop pilot AI-integrated protypes for infectious disease modelling, health system resource allocation and climate health risk forecasting.

It will also strengthen Canadian pandemic and emergency preparedness, enhance evidence-based policymaking, support climate-health adaptation strategies, improve health equity outcomes, and increase 91��ɫ’s national visibility in AI governance and public health innovation.

The post Artificial Intelligence for Public Health Advancement launches at 91��ɫ appeared first on News@91��ɫ.

TORONTO, March 25, 2026 – A new study by 91��ɫ researchers have found a potential striking flaw in artificial intelligence (AI) models.

Artificial neural networks (ANNs), a type of AI model built to solve vision tasks for computers, has surprisingly emerged as the . But does current AI really work like a primate brain?

“Artificial intelligence systems are often described as ‘brain-like’ because they can predict activity in parts of the brain that help us recognize objects,” says 91��ɫ Assistant Professor Kohitij Kar, senior author of a new study. “Until now, scientists mostly tested this in one direction. They asked whether AI models can predict brain activity.”

In this study, the researchers flipped the question – if AI truly mirrors the brain, shouldn’t brain activity also be able to predict what’s happening inside the AI model? – and developed a reverse predictivity test to find the answer.

“Ultimately, we need computational models to truly understand the underlying neural mechanisms of how we recognize objects. How do we see objects move? While it's a very easy task that we do every day, computationally, though, it's a very challenging problem,” says Kar, the Canada Research Chair in Visual Neuroscience and a member of 91��ɫ’s Centre for Vision Research and Centre for Integrative and Applied Neuroscience.

The researchers, including 91��ɫ Postdoctoral Fellow Sabine Muzellec, a Connected Minds trainee, used 1,320 natural or naturalistic synthetic images of a bear, an elephant, a face, an apple, a car, a dog, a chair, a plane, a bird and a zebra, placed against natural, indoor or outdoor background scenes. They also used an additional 300 images of the same objects rendered differently, such as outlines, drawing, schematized forms and artistic variations.

“The results were striking. While AI models can predict the neurons we recorded in the brain fairly well, the brain cannot equally predict many of the model’s internal features. Interestingly, this is not the case when neurons from one brain is compared against ones from another brain,” says Kar.

The problem with the ANNs solving vision differently is that this difference between primate brains and models will widen and compound over time if not corrected now. The direction of prediction was always to have the model predict like neurons, but if the reverse is not true then these models don't really serve as good hypotheses for the brain, adds Kar.

“The findings suggest that today’s AI systems solve visual tasks partly using internal strategies that the brain may not use. Importantly, the parts of AI models that align with the brain are also better at predicting real human behavior,” says Kar.

Why this matters

AI models are increasingly used to help design experiments to understand human behavior, including in clinical settings. It is assumed AI model see the world similarly to how a human brain does.

“Our findings challenge how similar current AI systems really are with the primate brain. We show that models that were previously thought to be brain-like rely on internal components that the brain does not appear to use. We provide a well vetted diagnostic metric for the field,” says Muzellec.

If AI models can become more brain-like, they could in the future help people with everything from post-traumatic stress disorder to autism, but for now, their use in experiments and to understand human behaviour is fraught. Similar models are also being used now for auditory systems, language systems and motor systems, but again, if they aren’t working as expected that’s an issue.

“Our approach helps identify which parts of an ANN truly match brain activity, allowing us to build more reliable models for understanding how people see and interpret the world,” says Kar. “This is especially important for our autism research program, which builds on models of the neurotypical brain as a baseline.”

The study’s authors have also made a available for AI developers to use to both test and improve their models going forward.

The , published today in Nature Machine Intelligence, introduces a new standard for building AI that is not just powerful – but truly brain-aligned.

The post Rethinking brain-like artificial intelligence: A new study reveals hidden mismatches appeared first on News@91��ɫ.

Professor says this could lead to better treatments for late-stage cancer patients

TORONTO, Dec. 8 2025 — While targeted radiation can be an effective treatment for brain tumours, subsequent potential necrosis of the treated areas can be hard to distinguish from the tumours on a standard MRI. led by a 91��ɫ professor in the Lassonde School of Engineering found that a novel AI-based method is better able to distinguish between the two types of lesions on advanced MRI than the human eye alone, a discovery that could help clinicians more accurately identify and treat the issues.

“The study shows, for the first time, that novel attention-guided AI methods coupled with advanced MRI can differentiate, with high accuracy, between tumour progression and radiation necrosis in patients with brain metastasis treated with stereotactic radiosurgery,” says 91��ɫ Research Chair Ali Sadeghi-Naini, senior author of the paper and associate professor of biomedical engineering and computer science. “Timely differentiation between tumour progression and radiation necrosis after radiotherapy in brain tumours is a crucial challenge in cancer centers, since these two conditions require quite different treatment approaches.”

The study, published in the International Journal of Radiation Oncology, Biology, Physics, was conducted in close collaboration with imaging scientists, neuro-oncologists and neuro-radiologists at Sunnybrook Health Sciences Centre using data acquired from more than 90 cancer patients whose original cancer had metastasized to the brain.

Sadeghi-Naini says the incidence of brain metastasis  is rising as treatments improve and survival rates increase. Stereotactic radiosurgery (SRS), where a concentrated doses of radiation are applied to the cancer lesions only, is effective at controlling the tumours.  In up to 30 per cent of cases, SRS is not able to control the tumour and it continues to grow. Where it is successful, healthy brain tissue immediately surrounding the tumour may also die off, called brain radiation necrosis, and it can come with significant side effects.

Sadeghi-Naini and his colleagues introduced a 3D deep learning AI model with two advanced attention mechanisms to differentiate between tumour progression and radiation necrosis using a specialized MRI technique, called chemical exchange saturation transfer (CEST), and found that the AI was able to differentiate between the two conditions with over 85 per accuracy. Sadeghi-Naini says with a standard MRI the two conditions are accurately diagnosed about 60 per cent of the time, and with more advanced MRI techniques alone, the rate increases to about 70 per cent.

“Differentiating tumour progression and  radiation necrosis is very important — one needs more anti-cancer therapies and may need to be aggressively treated with more radiation, sometimes surgery.  The other may require observation, anti-inflammatory drugs, so getting this right is crucial for patients.”

-30-

91��ɫ is a modern, multi-campus, urban university located in Toronto, Ontario. Backed by a diverse group of students, faculty, staff, alumni and partners, we bring a uniquely global perspective to help solve societal challenges, drive positive change, and prepare our students for success. 91��ɫ's fully bilingual Glendon Campus is home to Southern Ontario's Centre of Excellence for French Language and Bilingual Postsecondary Education. 91��ɫ’s campuses in Costa Rica and India offer students exceptional transnational learning opportunities and innovative programs. Together, we can make things right for our communities, our planet, and our future.

Media Contact:

Emina Gamulin, 91��ɫ Media Relations, 437-217-6362, egamulin@yorku.ca

The post Novel AI technique able to distinguish between progressive brain tumours and radiation necrosis, 91��ɫ U study finds appeared first on News@91��ɫ.

Each Solution Network will receive $700,000 over two years to design, develop and implement AI safety solutions to pressing challenges.

Reposted with permission from CIFAR and written by Justine Brooks. See .

CIFAR has launched its first two AI safety Solution Networks under the . The two research teams –  and  (the latter co-funded by the IDRC) – will spend the next two years developing and implementing open-source AI solutions to make AI safer and more inclusive for Canadians and the Global South. Each network is awarded $700,000 to support their groundbreaking research and development.

The Solution Networks are funded through the CAISI Research Program at CIFAR, an independent, multidisciplinary research arm led by CIFAR. The dedicated research program is a core component of the Government of Canada’s Canadian AI Safety Institute, launched in November 2024 with a $50 million investment to address the evolving risks of AI to Canadians.

“AI safety is crucial as the technology becomes more deeply embedded in how we live and work. At its core, it’s about two things — building trust and developing the tools to uphold it,” says the Honourable Evan Solomon, Minister of Artificial Intelligence and Digital Innovation and Minister responsible for the Federal Economic Development Agency for Southern Ontario. “Trust that AI will be used responsibly, and tools that make it safer, fairer, and more transparent. These new Solution Networks show how Canadian researchers are advancing the science of safety itself — turning ideas into real solutions that make AI work for people.”

“CIFAR’s Solution Networks provide a unique approach to trustworthy AI research and development, bringing together exceptional teams of interdisciplinary researchers – who might not otherwise cross paths – to address issues of global importance, but more importantly, to design, develop and implement solutions,” says Elissa Strome, Executive Director, Pan-Canadian AI Strategy at CIFAR. “Core to the work of both of these Solution Networks is exploring ways to mitigate the potential harms of AI to people in Canada and around the world.”

Solution Network Members

• , Solution Network Co-director (University of Toronto)
• , Solution Network Co-director (University of Waterloo, Osgoode Hall Law School (91��ɫ), Vector Institute)
• , Solution Network Member (University of Ottawa)
• , Solution Network Member (Western University) 
• , Solution Network Member (University of British Columbia, Canada CIFAR AI Chair, Vector Institute) 
• , Solution Network Member (University of Waterloo) 

Co-directed by Ebrahim Bagheri and Maura R. Grossman, this Solution Network aims to address the rising prevalence of synthetic AI-generated content in the justice system. This includes fake image or video evidence generated by people using AI tools, but also court documents that are created using large language models (LLMs) such as ChatGPT that may produce hallucinations.

“The issue now is that you can do this at scale and at convenience,” Bagheri says. Previously, one would have to spend large amounts of time and money to forge evidence. Now, evidence can be doctored quickly and easily, and even fabricated entirely from scratch.

The stakes are incredibly high, says Grossman. “Somebody can go to jail or not go to jail depending on whether something is a real or fake video.”

Somebody can go to jail or not go to jail depending on whether something is a real or fake video
Maura R. Grossman, 91��ɫ adjunct professor, Osgoode Hall Law School, CIFAR Solution Network Co-director

It’s not always financially feasible to bring in an expert who can evaluate the provenance of AI-generated content or evidence. The team proposes to develop a free, open-source framework that anyone within the court system can use to identify potentially problematic content. 

“We need a [transparent] tool that knows when it’s not sure about its output. One that is user friendly for this very unique group of users including both self-represented litigants and officers in the court system,” adds Grossman. 

Their solution could have a huge impact on the efficiency and trustworthiness of a justice system that is facing a great amount of change in a short period of time. “Even if our solution isn’t perfect, even if it gets 50, 60 or 70 percent of the way to be able to rule out [synthetic content], then we’ve really come a long way for the court system.”

Solution Network Members

• , Solution Network Co-director (91��ɫ, Vector Institute)
• , Solution Network Co-director (Brock University)
• , Solution Network Member (University of Waterloo, Canada CIFAR AI Chair, Vector Institute) 
• , Solution Network Member (University of Nigeria)
• , Solution Network Member (University of Nigeria)
• , Solution Network Member (University of Ibadan)

The use of LLMs like ChatGPT has exploded in recent years, but for speakers of non-standard English, these tools are not as safe or effective as they are for others. This is the problem Laleh Seyyed-Kalantari and Blessing Ogbuokiri are working to address. 

Their Solution Network focuses on Nigerian Pidgin English, a language spoken by over 140 million people, primarily in West Africa. LLMs trained on standard English often misinterpret marginalized dialects like Pidgin as toxic or offensive and penalize the user. This can lead to very real harms like censorship on social media and discrimination in service-delivery systems. 

I think what makes our solution unique is that it is locally rooted and culturally representative of citizens of African countries
Laleh Seyyed-Kalantari, 91��ɫ assistant professor, Lassonde School of Engineering, CIFAR Solution Network Co-director

The team will work to create the first ever bias and safety benchmarks for Pidgin English as part of an open-source audit and mitigation toolkit. These resources will be available for developers and policymakers to use to ensure AI systems are fair and safe for all users. “We are trying to create an AI system where marginalized voices can feel comfortable using these tools because it will accommodate them,” adds Ogbuokiri. 

The team will work with a citizen network in Nigeria, who will help to evaluate the data sets and LLMs used in the project. “I think what makes our solution unique is that it is locally rooted and culturally representative of citizens of African countries,” explains Seyyed-Kalantari. 

The team also has a policymaking objective, adds Seyyed-Kalantari. “We want to ensure that the research that we are developing […] brings actual positive changes for people who are using these LLMs in Africa.”

Ogbuokiri notes the impact this project could have beyond West Africa for immigrant and Indigenous communities in Canada who also use non-standard English varieties. “This will serve as a vital public resource for researchers, developers and policymakers,” he states. “This project will contribute to locally-grounded and culturally-relevant AI systems that reflect the realities of the Global South.”

About the CAISI Research Program at CIFAR

The CAISI Research Program at CIFAR is a component of the , launched by Innovation, Science and Economic Development Canada.  is the scientific engine of a broad national effort that aims to promote the safe and responsible development and deployment of AI. The research program is independently leading Canadian, multidisciplinary research to find solutions to complex AI safety challenges and develop practical tools for responsible AI so that AI is safe for all Canadians.

The post CIFAR launches new AI safety Networks to address synthetic evidence in the legal system and linguistic inequality appeared first on News@91��ɫ.

TORONTO, June 4, 2025 – If data used to train artificial intelligence models for medical applications, such as hospitals across the Greater Toronto Area, differs from the real-world data, it could lead to patient harm. A new study out today from 91��ɫ found proactive, continual and transfer learning strategies for AI models to be key in mitigating data shifts and subsequent harms.

To determine the effect of data shifts, the team built and evaluated an early warning system to predict the risk of in-hospital patient mortality and enhance the triaging of patients at seven large hospitals in the Greater Toronto Area.

The study used GEMINI, Canada’s largest hospital data sharing network, to assess the impact of data shifts and biases on clinical diagnoses, demographics, sex, age, hospital type, where patients were transferred from, such as an acute care institution or nursing home, and time of admittance. It included 143,049 patient encounters, such as lab results, transfusions, imaging reports and administrative features.

“As the use of AI in hospitals increases to predict anything from mortality and length of stay to sepsis and the occurrence of disease diagnoses, there is a greater need to ensure they work as predicted and don’t cause harm,” says senior author 91��ɫ Assistant Professor of 91��ɫ’s School of Health Policy and Management, Faculty of Health, a member of Connected Minds and a faculty affiliate at the Vector Institute.

“Building reliable and robust machine learning models, however, has proven difficult as data changes over time creating system unreliability.”

The data to train clinical AI models for hospitals and other health-care settings need to accurately reflect the variability of patients, diseases and medical practices, she adds. Without that, the model could develop irrelevant or harmful predictions, and even inaccurate diagnoses. Differences in patient subpopulations, staffing, resources, as well as unforeseen changes to policy or behaviour, differing health-care practices between hospitals or an unexpected pandemic, can also cause these potential data shifts.

“We found significant shifts in data between model training and real-life applications, including changes in demographics, hospital types, admission sources, and critical laboratory assays,” says first author Vallijah Subasri, AI scientist at University Health Network. “We also found harmful data shifts when models trained on community hospital patient visits were transferred to academic hospitals, but not the reverse.”

To mitigate these potentially harmful data shifts, the researchers used a transfer learning strategies, which allowed the model to store knowledge gained from learning one domain and apply it to a different but related domain and continual learning strategies where the AI model is updated using a continual stream of data in a sequential manner in response to drift-triggered alarms.

Although machine learning models usually remain locked once approved for use, the researchers found models specific to hospital type which leverage transfer learning, performed better than models that use all available hospitals.

Using drift-triggered continual learning helped prevent harmful data shifts due to the COVID-19 pandemic and improved model performance over time.

Depending on the data it was trained on, the AI model could also have a propensity for certain biases leading to unfair or discriminatory outcomes for some patient groups. 

“We demonstrate how to detect these data shifts, assess whether they negatively impact AI model performance, and propose strategies to mitigate their effects. We show there is a practical pathway from promise to practice, bridging the gap between the potential of AI in health and the realities of deploying and sustaining it in real-world clinical environments,” says Dolatabadi.

The study is a crucial step towards the deployment of clinical AI models as it provides strategies and workflows to ensure the safety and efficacy of these models in real-world settings.

“These findings indicate that a proactive, label-agnostic monitoring pipeline incorporating transfer and continual learning can detect and mitigate harmful data shifts in Toronto’s general internal medicine population, ensuring robust and equitable clinical AI deployment,” says Subasri.

The paper, , was published today in the journal JAMA Network Open.

The post New research finds specific learning strategies can enhance AI model effectiveness in hospitals appeared first on News@91��ɫ.

What does it mean to think, act and work as a Jewish professor when human freedoms are under siege and authoritarian power gains ground? And how can we draw on our Jewish identities to navigate the sweeping encroachment of new technologies like AI?

As communication scholars, colleagues and collaborators, we have spent a lot of time trying to answer these questions in our by taking cues from the of our .

Lately, Donald Trump’s administration has demonstrated a heavy investment in cataloguing and categorizing Jewish professors. In April, the Equal Employment Opportunity Commission (EEOC) to the personal cellphones of faculty and staff at Barnard College, asking them to self-identify as Jewish and/or Israeli. The text message also asked them to disclose any instances of antisemitic discrimination or harassment they had experienced.

Presumably, the text message inquiry itself was not recognized by its senders as an instance of .

We do not believe being a Jewish professor means silencing our students as they protest atrocities in Gaza, and it certainly doesn’t mean . Rather, it means drawing upon the tools of our forebears to question systems of oppression, wherever and however they may arise.

We simultaneously occupy within the university and North American society at large. This makes us acutely aware of how fragile conditional tolerance is, and how quickly can be used to justify repression or violence.

Collection and use of data

As communication and media scholars, we’re often critical of how . The EEOC questionnaire concerns us because it reduces the complexities of Jewish identity and the profound harms of antisemitism to a handful of abstract and ideologically determined data points.

Our recent (genAI) and its incompatibility with Jewish cultural expression shows that meaningful efforts to combat antisemitism — and other forms of oppression — must centre the knowledge and experiences of affected communities.

Our research found that outputs of chatbots such as ChatGPT are in a Jewish comedic style without resorting to offensive tropes. In another forthcoming study, we argue that genAI is equally incapable of representing the multifaceted “” of Jewish people except by smashing together rudimentary cultural signifiers (such as rainbows for queerness or bagels for Jewishness).

In each case, these platforms rely on datasets to determine what Jewishness is, and these datasets originate from the narratives that other people tell about Jewish people, rather than the ones we tell about ourselves.

Critical strategies

These platforms have increasingly become parts of daily life and communicative infrastructure. To investigate them, we adopted two critical strategies from our shared heritage as Ashkenazi Jews: and .

Both terms are Yiddish. Kibbitzing is a lively, informal way of thinking and talking together. It’s somewhere between joking, arguing and exchanging ideas. It is grounded in our relationships, histories and biases; kibbitzing is how we make shared meaning together through many voices.

Kibbitzing values contradiction, humour and the messiness of human conversation. Unlike AI chatbots, which follow scripted, dialogic, based on , kibbitzing is .

When we kibbitz, we build understanding by challenging one another and reflecting on what each of us brings to the table. In the age of genAI, kibbitzing offers a way to talk that is , laughter and deep, collective insight.

Futzing means messing around via hands-on experimentation, with no set agenda and no official guidance. This unstructured inquiry is an acknowledgement of Jews’ . As we write in our forthcoming article, these practices reflect what social theorist Michel de Certeau calls “,” a tactical means of collective empowerment in a hostile society.

Using futzing as a methodology, we started exploring genAI, drawing on our curiosity to see what might happen by playing, testing and responding in real time.

Futz first, then kibbitz

Each of us futzed on our own at first, with no ambition to crack the code or reverse-engineer the algorithm. Later, when we began kibbitzing together, we realized our scattered efforts were actually circling around shared concerns. Futzing helped us see patterns, surprises and contradictions — things we might have missed with a more rigid approach. Kibbitzing helped us connect those patterns and reconcile the contradictions.

Drawing on our culture this way allows us to imagine inclusive, anti-oppressive Jewish epistemologies that respond to the complexity of the current political moment. — like — is porous and resistant to fixed form. Our shared North American identity is just one of many possible perspectives that comprise a broader identity of Jewishness.

That is not a problem to be solved. Rather, it is a strength and a bond between us. Readers may well see their own cultural traditions, vernaculars and ancestral practices in this light too, as techniques of resilience and joy in the face of hardship and oppression.

There is an irony here. The deeper we dig into the intellectual roots of our own culture, the more common ground we might discover with everyone else’s. And that makes us feel a whole lot safer than getting a text from the EEOC ever could.

By Assistant Professor , Communication and Media Studies, 91��ɫ; Professor of Communication Studies , American University School of Communication; and Assistant Professor , Communications, Felician University

The post Experimenting with generative AI to kibbitz and futz towards more inclusive futures appeared first on News@91��ɫ.