Dr. Kasahara and his collaborators have investigated the dynamics of face recognition using motor-visual synchrony — the coordination between a person’s physical movements and the visual feedback they receive from those movements. They found that whether we influence the movement of our self-image or not, levels of identification with our face remain consistent. Therefore, our sense of agency, or subjective feelings of control, do not impact our level of identification with our self-image. Their results have been published in Scientific Reports.

The effect of agency on perceptions of identity

With psychological experiments using displays and cameras, the scientists investigated where the “self-identification boundary” is and what impacts this boundary. Participants were seated and asked to look at screens showing their faces gradually changing. At some point, the participants could notice a change in their facial identity and were asked to press a button when they felt that the image on the screen was no longer them. The experiment was done in both directions: the image changing from self to other and other to self.

“It’s like watching your face in a mirror as you move it and you identify yourself, but your face slowly changes up to a point and you realize this is no longer you,” Dr. Kasahara explained.

The researchers examined how three movement conditions affect the facial boundary: synchronous, asynchronous, and static. They hypothesized that if the motions are synchronized, participants would identify with the images to a greater extent. Surprisingly, they found that whether movements were synchronized or not, their facial identity boundaries were similar. Additionally, participants were more likely to identify with static images of themselves than images with their faces moving.

Interestingly, the direction of morphing — whether from self to other or other to self — influenced how participants perceived their own facial boundaries: participants were more likely to identify with their facial images when these images morphed from self to other rather than from other to self. Overall, the results suggest that a sense of agency of facial movements does not significantly impact our ability to judge our facial identity.

“Consider the example of deepfakes, which are essentially a form of asynchronous movement. When I remain still but the visual representation moves, it creates an asynchronous situation. Even in these deepfake scenarios, we can still experience a feeling of identity connection with ourselves,” Dr. Kasahara explained. “This suggests that even when we see a fake or manipulated version of our image, for example, someone else using our face, we might still identify with that face. Our findings raise important questions about our perception of self and identity in the digital age.”

How does identity impact perceptions of control?

What about the other way around? How does our sense of identity impact our sense of agency? Dr. Kasahara recently published a paper in collaboration with Professor of Psychology at Rikkyo University, Dr. Wen Wen, who specializes in research on our sense of agency. They investigated how recognizing oneself through facial features might affect how people perceive control over their own movements.

During experiments, participants observed either their own face or another person’s face on a screen and could interact and control the facial and head movements. They were asked to observe the screen for about 20 seconds while moving their faces and changing their facial expressions. The motion of the face was controlled either only by their own facial and head motion or by an average of the participant’s and the experimenter’s motion (full control vs. partial control). Thereafter, they were asked “how much did you feel that this face looks like you?” and “how much control did you feel over this presented face?”

Again, the main findings were intriguing: participants reported a higher sense of agency over the “other face” rather than the “self-face.” Additionally, controlling someone else’s face resulted in more variety of facial movements than controlling one’s own face.

“We gave the participants a different face, but they could control the facial movements of this face — similar to deepfake technology, where AI can transfer movement to other objects. This AI technology allows us to go beyond the conventional experience of simply looking into a mirror, enabling us to disentangle and investigate the relationship between facial movements and visual identity,” Dr. Kasahara stated.

“Based on previous research, one might expect that if I see my own face, I will feel more control over it. Conversely, if it’s not my face, I might expect to feel less control because it’s someone else’s face. That’s the intuitive expectation. However, the results are the opposite — when people see their own face, they report a lower sense of agency. Conversely, when they see another person’s face, they’re more likely to feel a sense of agency.” These surprising results challenge what we thought we knew about how we see ourselves in images.

Dr. Kasahara emphasized that the acceptance of technology in society plays a crucial role in technological advancements and human evolution: “The relationship between technology and human evolution is cyclical; we evolve together. But concerns about certain computer technology may lead to restrictions. My goal is to help foster acceptance within society and update our understanding of “the self” in relation to human-computer integration technology.”

The study of topological states of matter and their HOT counterparts has attracted considerable attention among physicists and engineers. This fervent interest stems from the discovery of topological insulators — materials that conduct electricity only on the surface or edges — while their interiors remain insulating. Due to the unique mathematical properties of topology, the electrons flowing along the edges are not hampered by any defects or deformations present in the material. Hence, devices made from such topological materials hold great potential for more robust transport or signal transmission technology.

Using many-body quantum interactions, a team of researchers led by Assistant Professor Lee Ching Hua from the Department of Physics under the NUS Faculty of Science has developed a scalable approach to encode large, high-dimensional HOT lattices representative of actual topological materials into the simple spin chains that exist in current-day digital quantum computers. Their approach leverages the exponential amounts of information that can be stored using quantum computer qubits while minimising quantum computing resource requirements in a noise-resistant manner. This breakthrough opens up a new direction in the simulation of advanced quantum materials using digital quantum computers, thereby unlocking new potential in topological material engineering.

The findings from this research have been published in the journal Nature Communications.

Asst Prof Lee said, “Existing breakthrough studies in quantum advantage are limited to highly-specific tailored problems. Finding new applications for which quantum computers provide unique advantages is the central motivation of our work.”

“Our approach allows us to explore the intricate signatures of topological materials on quantum computers with a level of precision that was previously unattainable, even for hypothetical materials existing in four dimensions” added Asst Prof Lee.

Despite the limitations of current noisy intermediate-scale quantum (NISQ) devices, the team is able to measure topological state dynamics and protected mid-gap spectra of higher-order topological lattices with unprecedented accuracy thanks to advanced in-house developed error mitigation techniques. This breakthrough demonstrates the potential of current quantum technology to explore new frontiers in material engineering. The ability to simulate high-dimensional HOT lattices opens new research directions in quantum materials and topological states, suggesting a potential route to achieving true quantum advantage in the future.

Reconstructing early human colonization on Mediterranean islands is challenging due to limited archaeological evidence. By studying a 25-foot submerged bridge, an interdisciplinary research team — led by USF geology Professor Bogdan Onac — was able to provide compelling evidence of earlier human activity inside Genovesa Cave, located in the Spanish island of Mallorca.

“The presence of this submerged bridge and other artifacts indicates a sophisticated level of activity, implying that early settlers recognized the cave’s water resources and strategically built infrastructure to navigate it,” Onac said.

The cave, located near Mallorca’s coast, has passages now flooded due to rising sea levels, with distinct calcite encrustations forming during periods of high sea level. These formations, along with a light-colored band on the submerged bridge, serve as proxies for precisely tracking historical sea-level changes and dating the bridge’s construction.

Mallorca, despite being the sixth largest island in the Mediterranean, was among the last to be colonized. Previous research suggested human presence as far back as 9,000 years, but inconsistencies and poor preservation of the radiocarbon dated material, such as nearby bones and pottery, led to doubts about these findings. Newer studies have used charcoal, ash and bones found on the island to create a timeline of human settlement about 4,400 years ago. This aligns the timeline of human presence with significant environmental events, such as the extinction of the goat-antelope genus Myotragus balearicus.

By analyzing overgrowths of minerals on the bridge and the elevation of a coloration band on the bridge, Onac and the team discovered the bridge was constructed nearly 6,000 years ago, more than two-thousand years older than the previous estimation — narrowing the timeline gap between eastern and western Mediterranean settlements.

“This research underscores the importance of interdisciplinary collaboration in uncovering historical truths and advancing our understanding of human history,” Onac said.

This study was supported by several National Science Foundation grants and involved extensive fieldwork, including underwater exploration and precise dating techniques. Onac will continue exploring cave systems, some of which have deposits that formed millions of years ago, so he can identify preindustrial sea levels and examine the impact of modern greenhouse warming on sea-level rise.

This research was done in collaboration with Harvard University, the University of New Mexico and the University of Balearic Islands.

But as these datasets are combined and recombined into multiple collections, important information about their origins and restrictions on how they can be used are often lost or confounded in the shuffle.

Not only does this raise legal and ethical concerns, it can also damage a model’s performance. For instance, if a dataset is miscategorized, someone training a machine-learning model for a certain task may end up unwittingly using data that are not designed for that task.

In addition, data from unknown sources could contain biases that cause a model to make unfair predictions when deployed.

To improve data transparency, a team of multidisciplinary researchers from MIT and elsewhere launched a systematic audit of more than 1,800 text datasets on popular hosting sites. They found that more than 70 percent of these datasets omitted some licensing information, while about 50 percent had information that contained errors.

Building off these insights, they developed a user-friendly tool called the Data Provenance Explorer that automatically generates easy-to-read summaries of a dataset’s creators, sources, licenses, and allowable uses.

“These types of tools can help regulators and practitioners make informed decisions about AI deployment, and further the responsible development of AI,” says Alex “Sandy” Pentland, an MIT professor, leader of the Human Dynamics Group in the MIT Media Lab, and co-author of a new open-access paper about the project.

The Data Provenance Explorer could help AI practitioners build more effective models by enabling them to select training datasets that fit their model’s intended purpose. In the long run, this could improve the accuracy of AI models in real-world situations, such as those used to evaluate loan applications or respond to customer queries.

“One of the best ways to understand the capabilities and limitations of an AI model is understanding what data it was trained on. When you have misattribution and confusion about where data came from, you have a serious transparency issue,” says Robert Mahari, a graduate student in the MIT Human Dynamics Group, a JD candidate at Harvard Law School, and co-lead author on the paper.

Mahari and Pentland are joined on the paper by co-lead author Shayne Longpre, a graduate student in the Media Lab; Sara Hooker, who leads the research lab Cohere for AI; as well as others at MIT, the University of California at Irvine, the University of Lille in France, the University of Colorado at Boulder, Olin College, Carnegie Mellon University, Contextual AI, ML Commons, and Tidelift. The research is published today in Nature Machine Intelligence.

Focus on finetuning

Researchers often use a technique called fine-tuning to improve the capabilities of a large language model that will be deployed for a specific task, like question-answering. For finetuning, they carefully build curated datasets designed to boost a model’s performance for this one task.

The MIT researchers focused on these fine-tuning datasets, which are often developed by researchers, academic organizations, or companies and licensed for specific uses.

When crowdsourced platforms aggregate such datasets into larger collections for practitioners to use for fine-tuning, some of that original license information is often left behind.

“These licenses ought to matter, and they should be enforceable,” Mahari says.

For instance, if the licensing terms of a dataset are wrong or missing, someone could spend a great deal of money and time developing a model they might be forced to take down later because some training data contained private information.

“People can end up training models where they don’t even understand the capabilities, concerns, or risk of those models, which ultimately stem from the data,” Longpre adds.

To begin this study, the researchers formally defined data provenance as the combination of a dataset’s sourcing, creating, and licensing heritage, as well as its characteristics. From there, they developed a structured auditing procedure to trace the data provenance of more than 1,800 text dataset collections from popular online repositories.

After finding that more than 70 percent of these datasets contained “unspecified” licenses that omitted much information, the researchers worked backward to fill in the blanks. Through their efforts, they reduced the number of datasets with “unspecified” licenses to around 30 percent.

Their work also revealed that the correct licenses were often more restrictive than those assigned by the repositories.

In addition, they found that nearly all dataset creators were concentrated in the global north, which could limit a model’s capabilities if it is trained for deployment in a different region. For instance, a Turkish language dataset created predominantly by people in the U.S. and China might not contain any culturally significant aspects, Mahari explains.

“We almost delude ourselves into thinking the datasets are more diverse than they actually are,” he says.

Interestingly, the researchers also saw a dramatic spike in restrictions placed on datasets created in 2023 and 2024, which might be driven by concerns from academics that their datasets could be used for unintended commercial purposes.

A user-friendly tool

To help others obtain this information without the need for a manual audit, the researchers built the Data Provenance Explorer. In addition to sorting and filtering datasets based on certain criteria, the tool allows users to download a data provenance card that provides a succinct, structured overview of dataset characteristics.

“We are hoping this is a step, not just to understand the landscape, but also help people going forward to make more informed choices about what data they are training on,” Mahari says.

In the future, the researchers want to expand their analysis to investigate data provenance for multimodal data, including video and speech. They also want to study how terms of service on websites that serve as data sources are echoed in datasets.

As they expand their research, they are also reaching out to regulators to discuss their findings and the unique copyright implications of fine-tuning data.

“We need data provenance and transparency from the outset, when people are creating and releasing these datasets, to make it easier for others to derive these insights,” Longpre says.