US growth leans on AI as rest of rich world stalls - Oxford Economics
Without the artificial intelligence boom, the US economy would have grown at a moderate annualised 1.4% in the first half of 2026, and other big advanced economies more slowly still, Oxford Economics estimated in a research briefing on October 29.
Should the AI boom slow or go into reverse, the risk of recession in the advanced economies would be substantial, according to Adam Slater, lead economist at the consultancy and author of the briefing, "The AI and non-AI economies - two different growth stories".
US: AI contribution to GDP growth, direct and indirect (ppt contributions to annualised GDP growth). Source: Oxford Economics. Note: investment is net of estimated imports
The estimate lands as investors grow nervous about how the AI build-out is being financed, after worries over AI debt triggered a chip sell-off across Asia in mid-October. It suggests that much of the strength in US output this year has come from a single sector, and that the rest of the rich world has little underlying momentum to fall back on.
A narrow US boom
Oxford Economics counts two main channels: investment in computer equipment, software, data centres and power generation, and the wealth effect of an AI-driven surge in share prices on consumer spending. Some estimates suggest AI-related stocks have accounted for about 75% of recent S&P 500 returns. Surging imports of the equipment used in the build-out partly offset the gain.
AI-related factors added about 1.1 percentage points to annualised US GDP growth in 1Q26 and about 0.8ppt in 2Q26, the consultancy estimates. In 2Q26 investment net of imports added 0.4ppt and stock market wealth effects a similar amount, so much of the headline 2.2% annualised growth rate can be put down to AI.
The US is leaning on technology far more than it did in the dotcom boom. That boom added about 0.8ppt a year to growth in 1996-2000, but underlying growth was strong too, at more than 3.0% a year. The narrow spread of today's gains, which accrue to a fairly small slice of US households, may help explain why consumer confidence has fallen since 2024 and sits below dotcom-era levels.
US: contributions to growth, 1996-2001 versus 2023-2026 (ppt contributions to GDP growth; Michigan consumer confidence index, right-hand scale). Source: Oxford Economics, Haver Analytics
Little to show elsewhere
Outside the US the effect is much smaller. Across six other advanced economies, AI-related factors added an average of 0.2ppt to growth in 2025-1H26, about a third of the US effect, ranging from about 0.1ppt in Germany to 0.3-0.4ppt in Sweden and Japan. Stripped of AI, growth in the group ran mostly at 0.5-1.0% a year.
Advanced economies: estimated shares of AI-related and non-AI factors in GDP growth (ppt contributions to y/y growth, 2025-1H26 average). Source: Oxford Economics
US growth without AI would still beat most of its peers, but by a narrower margin. Part of the weakness lies in interest-sensitive spending, which has been flat for three years outside the US as rates rose from 2022. Residential investment has shrunk in most economies, with especially big falls in Canada, Germany, France and Italy.
In the US, interest-sensitive sectors have grown by about 5% since 2022, less than GDP as a whole, but they are a smaller drag than they once would have been. Their share of US investment has fallen from 50-55% in 1980-2005 to below 40%, partly because tech has taken a growing share of equipment spending, and they respond less to rate changes than before.
The cost of the build-out
Data centres' energy needs, which grew by more than 25% last year, are pushing up US wholesale electricity prices, one of several costs the boom imposes on the rest of the economy. The Dallas Fed estimated earlier this year that existing data centres had raised prices by 3-5% nationwide, and US power prices relative to the rest of the G7 had opened a gap of about 12% by 3Q26. That could add up to 0.3ppt to consumer prices and, other things equal, take about 0.2% off GDP, Oxford Economics estimates.
US and other G7: electricity prices (indices, January 2024=100; data centre investment, $bn at 2017 prices; $/MWh, 4Q25). Source: Oxford Economics, Haver Analytics, IEA
US prices of computers and peripherals are up 10% y/y and software nearly 30%, together adding about 0.04ppt to the consumer price index, while construction materials are rising about 5% y/y. More speculatively, heavy borrowing by AI companies may have helped push up bond yields: a 100bp rise in 10-year yields cuts US GDP by 0.1-0.3%, by the consultancy's modelling.
Altogether, these spillovers could cut 0.2-0.3% from US GDP. In economies that have gained little from the boom, they could wipe out much of what modest gains there have been, according to Oxford Economics.
Without the artificial intelligence boom, the US economy would have grown at a moderate annualised 1.4% in the first half of 2026, and other big advanced economies more slowly still, Oxford Economics estimated in a research briefing on October 29.
Should the AI boom slow or go into reverse, the risk of recession in the advanced economies would be substantial, according to Adam Slater, lead economist at the consultancy and author of the briefing, "The AI and non-AI economies - two different growth stories".
US: AI contribution to GDP growth, direct and indirect (ppt contributions to annualised GDP growth). Source: Oxford Economics. Note: investment is net of estimated imports
The estimate lands as investors grow nervous about how the AI build-out is being financed, after worries over AI debt triggered a chip sell-off across Asia in mid-October. It suggests that much of the strength in US output this year has come from a single sector, and that the rest of the rich world has little underlying momentum to fall back on.
A narrow US boom
Oxford Economics counts two main channels: investment in computer equipment, software, data centres and power generation, and the wealth effect of an AI-driven surge in share prices on consumer spending. Some estimates suggest AI-related stocks have accounted for about 75% of recent S&P 500 returns. Surging imports of the equipment used in the build-out partly offset the gain.
AI-related factors added about 1.1 percentage points to annualised US GDP growth in 1Q26 and about 0.8ppt in 2Q26, the consultancy estimates. In 2Q26 investment net of imports added 0.4ppt and stock market wealth effects a similar amount, so much of the headline 2.2% annualised growth rate can be put down to AI.
The US is leaning on technology far more than it did in the dotcom boom. That boom added about 0.8ppt a year to growth in 1996-2000, but underlying growth was strong too, at more than 3.0% a year. The narrow spread of today's gains, which accrue to a fairly small slice of US households, may help explain why consumer confidence has fallen since 2024 and sits below dotcom-era levels.
US: contributions to growth, 1996-2001 versus 2023-2026 (ppt contributions to GDP growth; Michigan consumer confidence index, right-hand scale). Source: Oxford Economics, Haver Analytics
Little to show elsewhere
Outside the US the effect is much smaller. Across six other advanced economies, AI-related factors added an average of 0.2ppt to growth in 2025-1H26, about a third of the US effect, ranging from about 0.1ppt in Germany to 0.3-0.4ppt in Sweden and Japan. Stripped of AI, growth in the group ran mostly at 0.5-1.0% a year.
Advanced economies: estimated shares of AI-related and non-AI factors in GDP growth (ppt contributions to y/y growth, 2025-1H26 average). Source: Oxford Economics
US growth without AI would still beat most of its peers, but by a narrower margin. Part of the weakness lies in interest-sensitive spending, which has been flat for three years outside the US as rates rose from 2022. Residential investment has shrunk in most economies, with especially big falls in Canada, Germany, France and Italy.
In the US, interest-sensitive sectors have grown by about 5% since 2022, less than GDP as a whole, but they are a smaller drag than they once would have been. Their share of US investment has fallen from 50-55% in 1980-2005 to below 40%, partly because tech has taken a growing share of equipment spending, and they respond less to rate changes than before.
The cost of the build-out
Data centres' energy needs, which grew by more than 25% last year, are pushing up US wholesale electricity prices, one of several costs the boom imposes on the rest of the economy. The Dallas Fed estimated earlier this year that existing data centres had raised prices by 3-5% nationwide, and US power prices relative to the rest of the G7 had opened a gap of about 12% by 3Q26. That could add up to 0.3ppt to consumer prices and, other things equal, take about 0.2% off GDP, Oxford Economics estimates.
US and other G7: electricity prices (indices, January 2024=100; data centre investment, $bn at 2017 prices; $/MWh, 4Q25). Source: Oxford Economics, Haver Analytics, IEA
US prices of computers and peripherals are up 10% y/y and software nearly 30%, together adding about 0.04ppt to the consumer price index, while construction materials are rising about 5% y/y. More speculatively, heavy borrowing by AI companies may have helped push up bond yields: a 100bp rise in 10-year yields cuts US GDP by 0.1-0.3%, by the consultancy's modelling.
Altogether, these spillovers could cut 0.2-0.3% from US GDP. In economies that have gained little from the boom, they could wipe out much of what modest gains there have been, according to Oxford Economics.
South Korea's chipmakers bank record AI profits as markets wobble
Samsung Electronics booked the biggest quarterly operating profit in its history on October 29, yet the AI spending behind the result looks shakier than it did a month ago.
The South Korean chipmaker (KRX: 005930) made an operating profit of KRW107.4 trillion ($80.2bn) in July-September, 782.8% more than a year earlier and the first time it has cleared KRW100 trillion in a single quarter. Heavy buying of AI hardware by the hyperscalers carried the figure past the KRW106 trillion consensus compiled by FnGuide.
Samsung now earns more in a quarter than Microsoft, Alphabet or Meta, the US groups whose data-centre budgets fill its order book. Memory makers have been the surest winners of the AI build-out so far, but their profits depend on those budgets holding, and investors spent October asking whether they will.
The rest of the income statement was less convincing. Revenue rose 126.9% to KRW195.3 trillion, short of the KRW200.3 trillion analysts had pencilled in, and net profit of KRW75.1 trillion fell well below the KRW87.4 trillion they expected.
SK hynix (KRX: 000660), the main supplier of high-bandwidth memory (HBM) to Nvidia, set the tone a day earlier. Its operating profit climbed 557.2% year on year to KRW83.33 trillion, at a margin above 80%, and it announced a fresh shareholder return plan worth KRW60 trillion on top of a KRW40 trillion programme it had just completed.
"Virtually all of our DRAM and HBM production capacity for the next three years through 2029 has already been sold out," the company told analysts on its results call.
Seoul took that as a licence to buy. The Kospi rose 6.6% to 6,012.44 on October 28, its steepest one-day gain in three months, as SK hynix shares added 10%, Saxo Bank said in a market note.
The rally came after a punishing month. The index lost 15.7% over the first three weeks of October, and on October 15 an 8% slide tripped a circuit breaker in Seoul as investors questioned the debt piling up behind US data-centre construction. SK hynix shelved a US listing of its Solidigm subsidiary on October 23, citing market conditions.
US results that evening clouded the picture again. Alphabet shares dropped about 10% in after-hours trading on disappointing revenue and negative free cash flow of $9.07bn for the third quarter, according to Saxo. Microsoft gained 3% as its Azure cloud sales grew 48%, but its capital spending of $39.4bn came in well below estimates, which is cold comfort to anyone selling it chips. Nasdaq 100 futures were down 0.9% in early Asian trade on October 29, and the Kospi opened flat at 6,017.22.
Swings of this size in both directions within a fortnight say more about investors' nerves than about the chipmakers' order books. Samsung's preliminary guidance put its operating margin at a record 55.1%, KB Securities expects its HBM prices to more than double in 2027, and SK hynix sees memory shortages lasting until 2030. The doubt lies with the buyers. While Alphabet, Microsoft and their peers keep their capex plans intact, Korea's memory makers will keep posting records. The first clear sign of a cutback would land hardest in Seoul, where the market's mood now rises and falls with two companies.
Samsung Electronics booked the biggest quarterly operating profit in its history on October 29, yet the AI spending behind the result looks shakier than it did a month ago.
The South Korean chipmaker (KRX: 005930) made an operating profit of KRW107.4 trillion ($80.2bn) in July-September, 782.8% more than a year earlier and the first time it has cleared KRW100 trillion in a single quarter. Heavy buying of AI hardware by the hyperscalers carried the figure past the KRW106 trillion consensus compiled by FnGuide.
Samsung now earns more in a quarter than Microsoft, Alphabet or Meta, the US groups whose data-centre budgets fill its order book. Memory makers have been the surest winners of the AI build-out so far, but their profits depend on those budgets holding, and investors spent October asking whether they will.
The rest of the income statement was less convincing. Revenue rose 126.9% to KRW195.3 trillion, short of the KRW200.3 trillion analysts had pencilled in, and net profit of KRW75.1 trillion fell well below the KRW87.4 trillion they expected.
SK hynix (KRX: 000660), the main supplier of high-bandwidth memory (HBM) to Nvidia, set the tone a day earlier. Its operating profit climbed 557.2% year on year to KRW83.33 trillion, at a margin above 80%, and it announced a fresh shareholder return plan worth KRW60 trillion on top of a KRW40 trillion programme it had just completed.
"Virtually all of our DRAM and HBM production capacity for the next three years through 2029 has already been sold out," the company told analysts on its results call.
Seoul took that as a licence to buy. The Kospi rose 6.6% to 6,012.44 on October 28, its steepest one-day gain in three months, as SK hynix shares added 10%, Saxo Bank said in a market note.
The rally came after a punishing month. The index lost 15.7% over the first three weeks of October, and on October 15 an 8% slide tripped a circuit breaker in Seoul as investors questioned the debt piling up behind US data-centre construction. SK hynix shelved a US listing of its Solidigm subsidiary on October 23, citing market conditions.
US results that evening clouded the picture again. Alphabet shares dropped about 10% in after-hours trading on disappointing revenue and negative free cash flow of $9.07bn for the third quarter, according to Saxo. Microsoft gained 3% as its Azure cloud sales grew 48%, but its capital spending of $39.4bn came in well below estimates, which is cold comfort to anyone selling it chips. Nasdaq 100 futures were down 0.9% in early Asian trade on October 29, and the Kospi opened flat at 6,017.22.
Swings of this size in both directions within a fortnight say more about investors' nerves than about the chipmakers' order books. Samsung's preliminary guidance put its operating margin at a record 55.1%, KB Securities expects its HBM prices to more than double in 2027, and SK hynix sees memory shortages lasting until 2030. The doubt lies with the buyers. While Alphabet, Microsoft and their peers keep their capex plans intact, Korea's memory makers will keep posting records. The first clear sign of a cutback would land hardest in Seoul, where the market's mood now rises and falls with two companies.
Fluent language makes a hallucinatory chatbot just as credible as a curt, flawless one
An eloquent AI system that shares incorrect information is rated just as highly by users as a curt AI system that provides correct answers
A chatbot that speaks naturally is more likely to be perceived by users as intelligent and useful, even if it spouts nonsense, research by communication scientists at Radboud University shows. An eloquent AI system that shares incorrect information is rated just as highly by users as a curt AI system that provides correct answers.
Most people are by now quite familiar with ChatGPT, Gemini, Claude and other chatbots: every day, people ask millions of questions and receive neatly formulated answers in return. Yet these language models still tend to run wild with so-called ‘hallucinations’: unreliable, dubious answers that are presented with great certainty as the truth. Communication scientist Marianna de Sá Siqueira and her colleagues investigated the effects of elegant language use on the credibility of chatbots.
An eloquent AI system that shares incorrect information is rated just as highly by users as a curt AI system that provides correct answers
A chatbot that speaks naturally is more likely to be perceived by users as intelligent and useful, even if it spouts nonsense, research by communication scientists at Radboud University shows. An eloquent AI system that shares incorrect information is rated just as highly by users as a curt AI system that provides correct answers.
Most people are by now quite familiar with ChatGPT, Gemini, Claude and other chatbots: every day, people ask millions of questions and receive neatly formulated answers in return. Yet these language models still tend to run wild with so-called ‘hallucinations’: unreliable, dubious answers that are presented with great certainty as the truth. Communication scientist Marianna de Sá Siqueira and her colleagues investigated the effects of elegant language use on the credibility of chatbots.
Nonsense in fluent language
For the experiment, participants were shown one of four variants of a conversation with a chatbot about The Lord of the Rings. Two variants featured fluent sentences and human-like language, one of which gave correct answers and the other did not. In addition, there were two variants with short, businesslike answers, one of which provided entirely correct information whilst the other presented (subtly) fabricated facts.
Unsurprisingly, users were most positive about the chatbot with elegant language and correct answers. However, a chatbot that writes fluently but spouts factual nonsense was rated by participants as just as intelligent and pleasant as a chatbot that communicates in a stilted manner but does provide factually correct information. It also emerged that as many as 80 per cent of participants did not notice the errors at all, even though they had been selected in advance for their basic knowledge of The Lord of the Rings.
"Above all, it shows how easily people can be led astray and how quickly we come to trust something that is actually spouting nonsense," says Barbara Müller, associate professor and co-author of the study. "If a text sounds elegant and natural, people automatically attribute intelligence to it. You then subconsciously tend to think: this is fine, I don’t need to check it any further."
For the experiment, participants were shown one of four variants of a conversation with a chatbot about The Lord of the Rings. Two variants featured fluent sentences and human-like language, one of which gave correct answers and the other did not. In addition, there were two variants with short, businesslike answers, one of which provided entirely correct information whilst the other presented (subtly) fabricated facts.
Unsurprisingly, users were most positive about the chatbot with elegant language and correct answers. However, a chatbot that writes fluently but spouts factual nonsense was rated by participants as just as intelligent and pleasant as a chatbot that communicates in a stilted manner but does provide factually correct information. It also emerged that as many as 80 per cent of participants did not notice the errors at all, even though they had been selected in advance for their basic knowledge of The Lord of the Rings.
"Above all, it shows how easily people can be led astray and how quickly we come to trust something that is actually spouting nonsense," says Barbara Müller, associate professor and co-author of the study. "If a text sounds elegant and natural, people automatically attribute intelligence to it. You then subconsciously tend to think: this is fine, I don’t need to check it any further."
Blind trust
The research highlights the danger of blind trust in artificial intelligence. Professor and co-author Tibor Bosse says: "So-called language models such as GPT and Claude are primarily trained to produce statistically plausible and easy-to-read sentences. These systems do not always take the truth very seriously. They are designed to sound human and convincing, but whether the information is factually correct is of secondary importance."
"People used to complain that chatbots were annoying and knew nothing. Now they sound so clever that we’re too quick to believe them," says Bosse. According to him, this cannot be solved with a simple disclaimer at the outset. "Most chatbots show a single line of text at the start stating that the output may not be entirely accurate, but those companies are certainly not going to indicate of their own accord with every answer that the system is uncertain. If we really want to protect users from chatbots spouting nonsense, we need to, for example, oblige tech companies to make that uncertainty visible. If a chatbot itself expresses doubt, you prevent people from taking nonsense as the truth."
The research highlights the danger of blind trust in artificial intelligence. Professor and co-author Tibor Bosse says: "So-called language models such as GPT and Claude are primarily trained to produce statistically plausible and easy-to-read sentences. These systems do not always take the truth very seriously. They are designed to sound human and convincing, but whether the information is factually correct is of secondary importance."
"People used to complain that chatbots were annoying and knew nothing. Now they sound so clever that we’re too quick to believe them," says Bosse. According to him, this cannot be solved with a simple disclaimer at the outset. "Most chatbots show a single line of text at the start stating that the output may not be entirely accurate, but those companies are certainly not going to indicate of their own accord with every answer that the system is uncertain. If we really want to protect users from chatbots spouting nonsense, we need to, for example, oblige tech companies to make that uncertainty visible. If a chatbot itself expresses doubt, you prevent people from taking nonsense as the truth."
Journal
Computers in Human Behavior Artificial Humans
Computers in Human Behavior Artificial Humans
DOI
Article Title
Users’ perception of chatbot language fluency and hallucination
Users’ perception of chatbot language fluency and hallucination
Article Publication Date
28-Oct-2026
28-Oct-2026
Hertz Fellow Owen Dugan helps launch AI company built for American factories
A new artificial intelligence company is working to help American factories bring products from design to production faster.
Arda, which launched Oct. 2026, is developing AI-powered software that turns product designs into assembly plans and updates those plans as designs change.
The technology aims to help manufacturers adapt production processes more quickly, reducing the time between designing a product and making it. Led by former OpenAI chief research officer Bob McGrew, Arda has raised $232 million, is valued at $1.5 billion, and plans to expand into robotic assembly, developing AI that could allow automated production lines to adapt to new designs in days rather than months.
Among the company’s founding researchers is 2024 Hertz Fellow Owen Dugan, who joined Arda in January as the first AI hire. During his intial weeks at Arda, he built a demo that became Arda’s first product.
“I joined Arda because I wanted to help American factories build faster. The Hertz Fellowship gave me the freedom to take a bet on a young company and the confidence that bold ideas are worth pursuing,” said Dugan. “Strengthening American manufacturing is one way I hope to honor the Hertz Fellows’ moral commitment to help the country in times of need.”
Dugan will present his recent AI model research at NeurIPS 2026 in Sydney on a framework for jointly training incrementally refinable Transformer widths with restart-free adaptive inference.
Dugan holds patents for algorithms that remove satellite trails from astronomical images, addressing a growing challenge for astronomers as more satellites enter orbit. He also co-founded a startup developing technology to verify the authenticity of digital images. These projects applied computational research to practical problems, experience he now brings to Arda, where he works alongside researchers and engineers from MIT, Palantir and Tesla to develop AI-powered manufacturing technology.
Dugan was one of 18 Hertz Fellows selected in 2024. The Hertz Fellowship is the nation’s most competitive doctoral fellowship in the applied sciences, engineering and mathematics. Awarded through a rigorous selection process honed over eight decades, Hertz Fellows receive up to five years of funding, providing the rare freedom to pursue bold ideas and a community of influential peers dedicated to their success.
Dugan’s involvement at Arda reflects a longstanding tradition of entrepreneurship within the Hertz Fellow community, where researchers apply their scientific expertise to build companies, develop new technologies and bring innovative ideas into practical use.
Throughout its history, the Hertz Foundation has supported generations of scientific entrepreneurs. Hertz Fellows have founded hundreds of companies, earned thousands of patents and made significant contributions to the nation’s economy. It continues to foster entrepreneurship through career support, mentorship, and up to $25,000 in funding through the Harold Newman and David Galas Entrepreneurial Initiative.
About the Hertz Foundation
The Hertz Foundation is the nation’s preeminent nonprofit organization committed to advancing American scientific and technological leadership. For more than 60 years, it has stood as an unwavering pillar of independent support through the renowned Hertz Fellowship, cultivating a multidisciplinary network of innovators whose work has positively impacted millions of lives. Learn more at hertzfoundation.org
Big data and AI drive the future of precision poultry nutrition
image:
Graphical abstract showing big data and AI-powered precision poultry nutrition, illustrating the cycle of data collection, AI modeling, precision farming, and biotechnologies.
Credit: HIGHER EDUCATION PRESS
LEAD
Integrating big data and artificial intelligence offers a sustainable solution to optimize feed efficiency and reduce the environmental footprint of global poultry farming.
BODY
[Paragraph 1 — The problem]
Global poultry production faces mounting challenges, including feed source shortages, environmental pollution, and the need to maintain animal welfare amid rising consumer demand. Traditional nutrition models often fail to accurately predict dynamic nutrient requirements, leading to wasted resources and excess greenhouse gas emissions. As the industry seeks sustainable expansion, there is an urgent need for more precise, data-driven nutritional strategies.
[Paragraph 2 — What was done]
Researchers from Jilin University, China Agricultural University and Jilin Academy of Agricultural Sciences conducted a comprehensive review of current precision nutrition technologies. They explored how the Internet of Things (IoT), computer vision, and sensor-based monitoring can be combined with advanced AI algorithms and machine learning. The team also summarized how biotechnologies, including omics techniques and biotechnology-derived proteins, can be integrated into modern poultry farming.
[Paragraph 3 — Key findings]
The review highlights that traditional linear regression models are insufficient for capturing the dynamic physiological states and environmental variations in poultry. In contrast, machine learning algorithms demonstrate superior performance in predicting feed energy values and nutrient requirements. Furthermore, the integration of big data allows for precise diet formulation and real-time adjustments, significantly enhancing the accuracy of nutritional supply. The authors emphasize that combining AI with omics technologies provides a deeper biological understanding of nutrient metabolism.
[Paragraph 4 — Significance and implications]
This integration promises to transform current poultry practices by improving feed efficiency and reducing production costs. By precisely matching nutrient supply to animal requirements, producers can significantly lower nitrogen and carbon emissions, mitigating environmental impacts. Ultimately, this data-driven approach supports sustainable agricultural development, ensuring food security while promoting better animal welfare.
Schematic diagram detailing the workflow of precision poultry farming.
Sankey diagram illustrating the application of AI and big data in poultry production based on a review of 32 articles.
Credit
HIGHER EDUCATION PRESS
HIGHER EDUCATION PRESS
Method of Research
Experimental study
Subject of Research
Not applicable
Article Title
Integration of Big Data and AI Empowered Precision Poultry Nutrition: A Review
New review illuminates the future of AI in pathology
Researchers bridge algorithms with clinical practice
image:
AI: Artificial intelligence; CNN: Convolutional neural networks; ResNet: Residual neural network.
view moreCredit: Xing-hua Gao from The First Hospital of China Medical University, China Image Source Link: https://www.ovid.com/jnls/cmj/fulltext/10.1097/cm9.0000000000004154~deep-learning-based-computational-pathology-technologies
Pathology, the gold standard for disease diagnosis, is undergoing a profound transformation. The shift from traditional microscopy to digital pathology, powered by whole-slide imaging (WSI), has opened the door for artificial intelligence (AI) to revolutionize diagnostic and prognostic workflows. However, a critical gap persists between the rapid advancements in AI algorithms and the practical demands of clinical practice. Pathologists need tools that are not only accurate but also interpretable, reliable, and seamlessly integrated into their work.
To bridge this divide, a team of researchers led by Prof. Xing-hua Gao from The First Hospital of China Medical University and Prof. Xiao-yu Cui from Northeastern University has published a comprehensive review. Their work introduces a novel framework for systematically connecting cutting-edge deep learning technologies with core clinical tasks in pathology. The review, which analyzes the field from 2020 to 2025, was made available online on July 22, 2026 and published in Chinese Medical Journal.
“The challenge isn’t just about creating powerful algorithms; it’s about making them clinically relevant and actionable,” explains Prof. Gao. “We introduced a dual-perspective framework that maps specific deep learning architectures—like CNNs, Vision Transformers, and Graph Neural Networks—to essential clinical needs such as diagnostic classification, tissue segmentation, and prognostic prediction. This provides a much-needed roadmap for developing AI solutions that truly address the challenges pathologists face every day.”
The review highlights how these advanced AI models are already making an impact. For instance, Vision Transformers can achieve over 94% accuracy in classifying oral epithelial dysplasia, while other models can predict cancer patient survival directly from standard H&E-stained slides, reducing the reliance on more expensive and time-consuming genetic tests. The paper also points to the landmark FDA approval of the Paige Prostate AI system as a sign that the digital era of pathology has truly arrived.
A significant hurdle for developing pathology AI is the need for vast amounts of expertly annotated data, which is both time-consuming and costly to produce. The review details emerging trends that tackle this bottleneck, including weakly supervised and self-supervised learning methods that minimize the reliance on these costly annotations. These approaches allow AI models to learn from large, unlabeled datasets, accelerating development and deployment.
“Our goal is to foster the development of AI systems that enhance, rather than replace, the invaluable expertise of pathologists,” notes Prof. Cui. “This means focusing on critical challenges like model interpretability, ensuring regulatory approval, and validating performance across multiple medical centers. The future lies in creating integrated, trustworthy, and equitable AI that becomes an indispensable partner in clinical decision-making.”
By providing a systematic overview and a clear algorithm-to-task mapping, the review serves as a vital resource for both researchers developing the next generation of AI tools and clinicians looking to integrate them into practice. The authors conclude with a forward-looking vision for a future where computational pathology tools are seamlessly woven into the fabric of clinical care, ultimately improving diagnostic accuracy and patient outcomes.
***
Reference
Title of original paper: Deep learning-based computational pathology: Technologies, clinical applications, and future directions
Journal: Chinese Medical Journal
DOI: 10.1097/CM9.0000000000004154
Funding
This work was supported by the National Key R&D Program of China (Grant No. 2023YFC2508200) and National Natural Science Foundation of China (Grant Nos. 82573977 and 82373483).
Journal
Chinese Medical Journal
Method of Research
Literature review
Subject of Research
Not applicable
Article Title
Deep learning-based computational pathology: Technologies, clinical applications, and future directions
No comments:
Post a Comment