First machine learning model developed to calculate the volume of all glaciers on Earth

Università Ca' Foscari Venezia

Facebook

XLinkedInWeChatBlueskyMessageWhatsAppEmail

 

image: 

Alaska glaciers

view more 

Credit: Niccolò Maffezzoli https://nmaffe.github.io/iceboost_webapp/

VENICE – A team of researchers led by Niccolò Maffezzoli, “Marie Curie” fellow at Ca’ Foscari University of Venice and the University of California, Irvine, and an associate member of the Institute of Polar Sciences of the National Research Council of Italy, has developed the first global model based on artificial intelligence to calculate the ice thickness distribution of all the glaciers on Earth. The model has been published in the journal Geoscientific Model Development and is expected to become a reference tool for those studying future glacier melt scenarios.

Accurate knowledge of glacier volumes is essential for projecting future sea level rise, managing water resources, and assessing societal impacts linked to glacier retreat. However, estimating their absolute volume remains a major scientific challenge. Over the years, more than 4 million in situ measurements of glacier thickness have been collected, thanks in large part to NASA’s Operation IceBridge. Despite the extensive dataset, current modelling approaches have not yet exploited its potential.

AI applied to glacier data
Direct measurements of glacier thickness cover less than 1% of the planet’s glaciers, highlighting the need for models capable of providing global-scale estimates of ice thickness and volume. This newly published study is the first to leverage such observational data in conjunction with the power of machine learning algorithms.

“Our model combines two decision tree algorithms,” explains Maffezzoli, “trained on thickness measurements and 39 features including ice velocity, mass balance, temperature fields, and geometric and geodetic variables. The trained model shows errors that are up to 30-40% lower than current traditional global models, particularly at polar latitudes and along the peripheries of the ice sheets, where the majority of the planet’s ice is located.”

Improving maps and projections of sea level rise
In polar regions and in the margins of Greenland and Antarctica, having accurate ice thickness estimates is particularly important. These estimates serve as initial conditions for numerical models that simulate ice flow and its interactions with the ocean—interactions that are key to projecting sea level rise under future climate scenarios. The model demonstrates strong generalisation capabilities in these regions and, the researchers believe, may help to refine current maps of subglacial topography in specific areas of the ice sheets, such as the Geikie Plateau or the Antarctic Peninsula.

This work represents an initial step towards producing updated estimates of global glacier volumes that will be useful to modellers, the IPCC, and policymakers.

“We aim to release two datasets totalling half a million ice thickness maps by the end of 2025,” announces Maffezzoli. “There is still a long way to go, but this work shows that AI and machine learning approaches are opening up new and exciting possibilities for ice modelling.”

The significance of glaciers
At present, glaciers contribute approximately 25-30% of observed global sea level rise, and their melting is accelerating. This is particularly significant in arid regions such as the Andes or the major mountain ranges of the Himalaya and Karakoram, where glacial headwaters support the livelihoods of billions. It is also critical for understanding the stability of the polar ice sheets in Greenland and Antarctica, where peripheral interactions with the ocean influence ice sheet dynamics.

Svalbard Archipelago glaciers

Credit

Niccolò Maffezzoli 

https://nmaffe.github.io/iceboost_webapp/

Journal

Geoscientific Model Development

DOI

10.5194/gmd-18-2545-2025 

Article Title

A gradient-boosted tree framework to model the ice thickness of the world's glaciers (IceBoost v1.1)

Article Publication Date

14-May-2025

 

AI tools may be weakening the quality of published research, study warns

University of Surrey

Facebook
X

LinkedInWeChatBlueskyMessageWhatsAppEmail

Artificial intelligence could be affecting the scientific rigour of new research, according to a study from the University of Surrey. 

The research team has called for a range of measures to reduce the flood of "low-quality" and "science fiction" papers, including stronger peer review processes and the use of statistical reviewers for complex datasets. 

In a study published in PLOS Biology, researchers reviewed papers proposing an association between a predictor and a health condition using an American government dataset called the National Health and Nutrition Examination Survey (NHANES), published between 2014 and 2024.  

NHANES is a large, publicly available dataset used by researchers around the world to study links between health conditions, lifestyle and clinical outcomes. The team found that between 2014 and 2021, just four NHANES association-based studies were published each year – but this rose to 33 in 2022, 82 in 2023, and 190 in 2024. 

Dr Matt Spick, co-author of the study from the University of Surrey, said: 

“While AI has the clear potential to help the scientific community make breakthroughs that benefit society, our study has found that it is also part of a perfect storm that could be damaging the foundations of scientific rigour. 

“We’ve seen a surge in papers that look scientific but don’t hold up under scrutiny – this is ‘science fiction’ using national health datasets to masquerade as science fact. The use of these easily accessible datasets via APIs, combined with large language models, is overwhelming some journals and peer reviewers, reducing their ability to assess more meaningful research – and ultimately weakening the quality of science overall.” 

The study found that many post-2021 papers used a superficial and oversimplified approach to analysis – often focusing on single variables while ignoring more realistic, multi-factor explanations of the links between health conditions and potential causes. Some papers cherry-picked narrow data subsets without justification, raising concerns about poor research practice, including data dredging or changing research questions after seeing the results. 

Tulsi Suchak, post-graduate researcher at the University of Surrey and lead author of the study, added: 

“We’re not trying to block access to data or stop people using AI in their research – we’re asking for some common sense checks. This includes things like being open about how data is used, making sure reviewers with the right expertise are involved, and flagging when a study only looks at one piece of the puzzle. These changes don’t need to be complex, but they could help journals spot low-quality work earlier and protect the integrity of scientific publishing.” 

To help tackle the issue, the team has laid out a number of practical steps for journals, researchers and data providers. They recommend that researchers use the full datasets available to them unless there’s a clear and well-explained reason to do otherwise, and that they are transparent about which parts of the data were used, over what time periods, and for which groups.  

For journals, the authors suggest strengthening peer review by involving reviewers with statistical expertise and making greater use of early desk rejection to reduce the number of formulaic or low-value papers entering the system. Finally, they propose that data providers assign unique application numbers or IDs to track how open datasets are used – a system already in place for some UK health data platforms. 

Anietie E Aliu, co-author of the study and post-graduate student at the University of Surrey, said: 

“We believe that in the AI era, scientific publishing needs better guardrails. Our suggestions are simple things that could help stop weak or misleading studies from slipping through, without blocking the benefits of AI and open data. These tools are here to stay, so we need to act now to protect trust in research.” 

---

Journal

PLOS Biology

DOI

10.1371/journal.pbio.3003152 

Article Title

Explosion of formulaic research articles, including inappropriate study designs and false discoveries, based on the NHANES US national health database

Article Publication Date

12-May-2025

 

AI overconfidence mirrors human brain condition

A similarity between language models and aphasia points to diagnoses for both

University of Tokyo

Facebook

XLinkedInWeChatBlueskyMessageWhatsApp\\Email

 

image: 

The nature of the dynamics of signals in both the brains of people with aphasia and in large language models, or LLMs, proved strikingly similar when represented visually. ©2025 Watanabe et al. CC-BY-ND

view more 

Credit: ©2025 Watanabe et al. CC-BY-ND

Agents, chatbots and other tools based on artificial intelligence (AI) are increasingly used in everyday life by many. So-called large language model (LLM)-based agents, such as ChatGPT and Llama, have become impressively fluent in the responses they form, but quite often provide convincing yet incorrect information. Researchers at the University of Tokyo draw parallels between this issue and a human language disorder known as aphasia, where sufferers may speak fluently but make meaningless or hard-to-understand statements. This similarity could point toward better forms of diagnosis for aphasia, and even provide insight to AI engineers seeking to improve LLM-based agents.

This article was written by a human being, but the use of text-generating AI is on the rise in many areas. As more and more people come to use and rely on such things, there’s an ever-increasing need to make sure that these tools deliver correct and coherent responses and information to their users. Many familiar tools, including ChatGPT and others, appear very fluent in whatever they deliver. But their responses cannot always be relied upon due to the amount of essentially made-up content they produce. If the user is not sufficiently knowledgeable about the subject area in question, they can easily fall foul of assuming this information is right, especially given the high degree of confidence ChatGPT and others show.

“You can’t fail to notice how some AI systems can appear articulate while still producing often significant errors,” said Professor Takamitsu Watanabe from the International Research Center for Neurointelligence (WPI-IRCN) at the University of Tokyo. “But what struck my team and I was a similarity between this behavior and that of people with Wernicke’s aphasia, where such people speak fluently but don’t always make much sense. That prompted us to wonder if the internal mechanisms of these AI systems could be similar to those of the human brain affected by aphasia, and if so, what the implications might be.”

To explore this idea, the team used a method called energy landscape analysis, a technique originally developed by physicists seeking to visualize energy states in magnetic metal, but which was recently adapted for neuroscience. They examined patterns in resting brain activity from people with different types of aphasia and compared them to internal data from several publicly available LLMs. And in their analysis, the team did discover some striking similarities. The way digital information or signals are moved around and manipulated within these AI models closely matched the way some brain signals behaved in the brains of people with certain types of aphasia, including Wernicke’s aphasia.

“You can imagine the energy landscape as a surface with a ball on it. When there’s a curve, the ball may roll down and come to rest, but when the curves are shallow, the ball may roll around chaotically,” said Watanabe. “In aphasia, the ball represents the person’s brain state. In LLMs, it represents the continuing signal pattern in the model based on its instructions and internal dataset.”

The research has several implications. For neuroscience, it offers a possible new way to classify and monitor conditions like aphasia based on internal brain activity rather than just external symptoms. For AI, it could lead to better diagnostic tools that help engineers improve the architecture of AI systems from the inside out. Though, despite the similarities the researchers discovered, they urge caution not to make too many assumptions.

“We’re not saying chatbots have brain damage,” said Watanabe. “But they may be locked into a kind of rigid internal pattern that limits how flexibly they can draw on stored knowledge, just like in receptive aphasia. Whether future models can overcome this limitation remains to be seen, but understanding these internal parallels may be the first step toward smarter, more trustworthy AI too.”

The similarity between an AI tool and a brain disorder in humans could open the door to new ways of exploring both. ©2025 Takamitsu Watanabe. CC-BY-ND

Credit

©2025 Watanabe et al. CC-BY-ND 

Journal article: Takamitsu Watanabe, Katsuma Inoue, Yasuo Kuniyoshi, Kohei Nakajima, Kazuyuki Aihara “Comparison of large language model with aphasia”, Advanced Science, https://doi.org/10.1002/advs.202414016

Funding: This work was supported by Grant-in-aid for Research Activity from Japan Society for Promotion of Sciences (19H03535, 21H05679, 23H04217, JP20H05921), The University of Tokyo Excellent Young Researcher Project, Showa University Medical Institute of Developmental Disabilities Research, JST Moonshot R&D Program (JPMJMS2021), JST FOREST Program (24012854), Institute of AI and Beyond of UTokyo, Cross-ministerial Strategic Innovation Promotion Program (SIP) on “Integrated Health Care System” (JPJ012425).

Research Contact:

International Research Center for Neurointelligence (WPI-IRCN) - https://ircn.jp/en/

Public Relations Group, The University of Tokyo,
7-3-1 Hongo, Bunkyo-ku, Tokyo, 113-8656, Japan
press-releases.adm@gs.mail.u-tokyo.ac.jp
 

About The University of Tokyo:

The University of Tokyo is Japan's leading university and one of the world's top research universities. The vast research output of some 6,000 researchers is published in the world's top journals across the arts and sciences. Our vibrant student body of around 15,000 undergraduate and 15,000 graduate students includes over 4,000 international students. Find out more at www.u-tokyo.ac.jp/en/ or follow us on X (formerly Twitter) at @UTokyo_News_en.

---

Journal

Advanced Science

DOI

10.1002/advs.202414016 

Method of Research

Experimental study

Subject of Research

People

Article Title

Comparison of large language model with aphasia

Article Publication Date

15-May-2025

 

Prominent chatbots routinely exaggerate science findings, study shows

Utrecht University

Facebook

Bluesky
Email

 

image: 

Three groups of icons representing people have shapes travelling between them and a page in the middle of the image. The page is a simple rectangle with straight lines representing data. The shapes traveling towards the page are irregular and in squiggly bands.

view more 

Credit: Yasmine Boudiaf & LOTI / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/

When summarizing scientific studies, large language models (LLMs) like ChatGPT and DeepSeek produce inaccurate conclusions in up to 73% of cases, according to a new study by Uwe Peters (Utrecht University) and Benjamin Chin-Yee (Western University, Canada/University of Cambridge, UK). The researchers tested the most prominent LLMs and analyzed thousands of chatbot-generated science summaries, revealing that most models consistently produced broader conclusions than those in the summarized texts. Surprisingly, prompts for accuracy increased the problem and newer LLMs performed worse than older ones.

Almost 5,000 LLM-generated summaries analyzed

The study evaluated how accurately ten leading LLMs, including ChatGPT, DeepSeek, Claude, and LLaMA, summarize abstracts and full-length articles from top science and medical journals (e.g., Nature, Science, and Lancet). Testing LLMs over one year, the researchers collected 4,900 LLM-generated summaries. Six of ten models systematically exaggerated claims found in the original texts often in subtle but impactful ways, for instance, changing cautious, past-tense claims like “The treatment was effective in this study” to a more sweeping, present-tense version like “The treatment is effective.” These changes can mislead readers into believing that findings apply much more broadly than they actually do.

Accuracy prompts backfired

Strikingly, when the models where explicitly prompted to avoid inaccuracies, they were nearly twice as likely to produce overgeneralized conclusions than when given a simple summary request. “This effect is concerning,” Peters said: “Students, researchers, and policymakers may assume that if they ask ChatGPT to avoid inaccuracies, they’ll get a more reliable summary. Our findings prove the opposite.”

Do humans do better?

Peters and Chin-Yee also directly compared chatbot-generated to human-written summaries of the same articles. Unexpectedly, chatbots were nearly five times more likely to produce broad generalizations than their human counterparts. “Worryingly”, said Peters, “newer AI models, like ChatGPT-4o and DeepSeek, performed worse than older ones.”

Why are these exaggerations happening? “Previous studies found that overgeneralizations are common in science writing, so it’s not surprising that models trained on these texts reproduce that pattern”, Chin-Yee noted. Additionally, since human users likely often prefer LMM responses that sound helpful and widely applicable, through interactions, the models may learn to favor fluency and generality over precision, Peters suggested.

Reducing the risks

The researchers recommend using LLMs such as Claude, which had the highest generalization accuracy, setting chatbots to lower ‘temperature’ (the parameter fixing a chatbot’s ‘creativity’), and using prompts that enforce indirect, past-tense reporting in science summaries. Finally, “If we want AI to support science literacy rather than undermine it,” Peters said, “we need more vigilance and testing of these systems in science communication contexts.”

 

---

Journal

Royal Society Open Science

Method of Research

Experimental study

Subject of Research

Not applicable

Article Title

Generalization bias in large language model summarization of scientific research

 

Study links adverse childhood experiences to higher risk of homelessness

University of Cincinnati researcher focuses on prevention, intervention to reduce long-term risks

University of Cincinnati

Facebook

XLinkedIn
MessageWhatsAppEmail

Children who have lived through a series of adverse childhood experiences also face an increased risk of homelessness during their childhood, according to a new study from the University of Cincinnati School of Social Work.

The study, led by Edson Chipalo, PhD, assistant professor in UC’s College of Allied Health Sciences, was recently published in the journal Child Indicators Research. Drawing on data from the National Survey of Children’s Health, the research adds to a growing body of evidence that childhood trauma can have long-lasting harmful consequences, particularly for children living in settings with limited resources.

Adverse childhood experiences, or ACEs, are potentially traumatic events that occur before a child turns 18 years old. These can include exposure to violence, abuse, neglect, discrimination, household dysfunction and other serious psychosocial stressors. Previous research has shown that such events are linked to delayed development and poor long-term health outcomes, including mental health disorders, substance use and chronic illnesses.

“This study offers a unique perspective due to its emphasis on the impact of ACEs on children,” said Chipalo. “This research differs from previous studies that have linked the number of ACEs and the likelihood of experiencing homelessness in adult populations in different settings.”

Chipalo’s analysis found that the risk of homelessness increases with the number of adverse experiences a child has. His findings suggest that the cumulative effect of adversities can influence not only health and emotional well-being but also housing and economic stability later in life.

The study used a social ecological framework to assess the relationship between childhood trauma and child homelessness. This approach examined the interaction of biological, psychological and social factors and how they affect a child’s development and outcomes.

By identifying these patterns early, Chipalo said, interventions can be developed to support children and families before long-term consequences take hold.

“The solutions must focus on prevention and providing early support,” he said. “Addressing ACEs at their root could reduce not only individual suffering but also broader social and economic challenges related to homelessness.”

He is scheduled to present his research findings at several academic conferences across North America this year.

In addition to this study, Chipalo is planning several future research projects that further examine the effects of adverse childhood experiences on a wide range of outcomes using a biopsychosocial framework. His upcoming work will explore how ACEs influence children’s participation in community activities, quality of sleep, body image dissatisfaction, body mass index, physical activity, temperament, digital media use, health care utilization and mental health in immigrant households with limited resources.

This summer, Chipalo said he also plans to begin collecting data that will examine the impact of ACEs on mental health and socioeconomic outcomes among African refugees and immigrants in the Greater Cincinnati area.

---

Journal

Child Indicators Research

Method of Research

Data/statistical analysis

Subject of Research

People

Article Title

Association between Adverse Childhood Experiences and Child Homelessness in the United States