Leading tech policy group calls for urgent review of speech recognition technology

Rigorous audit and governance practices proposed

Association for Computing Machinery

FacebookXLinkedInWeChatBlueskyMessageWhatsAppEmail

 

image: 

The overarching finding of “TechBrief: Automated Speech Recognition,” is that automated speech recognition (ASR) technologies require rigorous audit and governance practices, as well as guidelines to ensure they are inclusive and accessible.

view more 

Credit: Association for Computing Machinery

The Association for Computing Machinery’s global Technology Policy Council has published “TechBrief: Automated Speech Recognition.” It is the latest in a series of short technical bulletins that present scientifically grounded perspectives on the impact and policy implications of specific technological developments in computing. 

The overarching finding of “TechBrief: Automated Speech Recognition,” is that automated speech recognition (ASR) technologies require rigorous audit and governance practices, as well as guidelines to ensure they are inclusive and accessible. The authors note that automated speech recognition systems are quickly advancing by incorporating generative AI techniques similar to large language models (LLMs). They present an eye-opening list of ASR applications in daily life, including many uses the public may not have considered.

“The struggle is real. I can’t get AI customer support to understand me, and auto-captioning never matches what I said — it’s so frustrating,” says Shaomei Wu, a co-author of this Tech Brief, a person who stutters, and the founder and CEO of AImpower.org. “I feel voiceless and invalidated when the technology keeps interrupting or misunderstanding me.” Special attention is given to how ASR technologies can significantly impact people’s lives. For example, the brief points out that more than half a million doctors have used ASR tools to transcribe patient visits, and more than 60% of Fortune 100 companies use an ASR-based software package when hiring new employees. By providing this context, the co-authors hope readers will come to understand how important it is that ASR applications work properly.

“This is a story that has fallen under the media radar,” explained co-author Allison Koenecke, Assistant Professor, Cornell Tech. “There has been a good deal of focus on facial recognition and other new AI-based technologies, but relatively little discussion of speech recognition. Once we appreciate how critical it is that these applications function as intended, we come to realize the value of human oversight in how ASR is developed and deployed. Because the advances have been so rapid, and there has not been enough attention paid to speech recognition, we thought this was an opportune time to issue this report.”

The TechBrief authors contend that the first step to building safe and responsible oversight frameworks is to recognize that a “one size fits all” approach won’t work. Their key message is that the performance of an ASR system is highly variable depending on the setting. For example, the same medical patient's speech might reasonably be transcribed differently in varying circumstances (e.g., transcriptions used for diagnostic purposes may optimally include speech fragments and filler words, whereas transcriptions used for generating patient note summaries may not.)

In addition to the performance of speech recognition in high-stakes applications such as healthcare and employment, the co-authors raise concerns about the fairness, transparency, and accountability of ASR systems.

For example, in highlighting the Word Error Rate (WER) metric that is used to evaluate ASR programs, the co-authors relay several research findings to highlight how speech recognition can be inequitable. They cite statistics that: 1) The WER in ASR systems was 1.1 to 3.4 worse for Black American English speakers relative to White American English speakers, and 2) the WER in ASR systems was 2.8 to 4.2 times worse for Chicano English speakers relative to Standard American English speakers. The report contends that these statistics raise concerns about technological bias.

“As this technology becomes more pervasive, evaluating its impact on people of all backgrounds is a responsibility for its developers,” said co-author Jingjin Li, research fellow, AImpower.org.

"We shared emerging research on these questions because of its timely relevance," added Niranjan Sivakumar, a co-author of the ACM TechBrief, as well as Director and Head of Policy at AImpower.org. "Speech recognition is a field that has grown rapidly and is increasingly ubiquitous. We crafted this concise and accessible report to familiarize readers with how these tools are being used today and to draw attention to important issues to consider for the future. We hope our new report reaches a wide audience."

The brief’s co-authors include Allison Koenecke, Niranjan Sivakumar, Jingjin Li, and Shaomei Wu.

ACM’s TechBriefs are designed to complement ACM’s activities in the policy arena and to inform policymakers, the public, and others about the nature and implications of information technologies. Earlier ACM TechBriefs have covered topics such as tech abuse, generative artificial intelligence, climate change, facial recognition, smart cities, quantum simulation, and election auditing. Topics under consideration for future issues include AI development and considerations around open-source initiatives.

 

About the ACM Technology Policy Council
ACM’s global Technology Policy Council sets the agenda for global initiatives to address evolving technology policy issues and coordinates the activities of ACM's regional technology policy committees in the US and Europe. It serves as the central convening point for ACM's interactions with government organizations, the computing community, and the public in all matters of public policy related to computing and information technology. The Council’s members are drawn from ACM's global membership.

About ACM
ACM, the Association for Computing Machinery, is the world’s largest educational and scientific computing society, uniting computing educators, researchers, and professionals to inspire dialogue, share resources, and address the field’s challenges. ACM strengthens the computing profession’s collective voice through strong leadership, promotion of the highest standards, and recognition of technical excellence. ACM supports the professional growth of its members by providing opportunities for life-long learning, career development, and professional networking.

LLM use is reshaping scientific enterprise by increasing output, reducing quality and more

Summary author: Walter Beckwith

American Association for the Advancement of Science (AAAS)

FacebookXLinkedInWeChatBlueskyMessageWhatsAppEmail

LLM-assisted manuscripts exhibit more complexity of the written word but are lower in research quality, according to a Policy Article by Keigo Kusumegi, Paul Ginsparg, and colleagues that sought to evaluate the impacts of widespread use of generative artificial intelligence (AI) technologies on scientific production. “As AI systems advance, they will challenge our fundamental assumptions about research quality, scholarly communication, and the nature of intellectual labor,” write the authors. “Science policymakers must consider how to evolve our scientific institutions to accommodate the rapidly changing scientific production process.” Despite enormous enthusiasm and growing concern surrounding the use of generative AI and large language models (LLMS) across research and academia, there has been little systemic evidence about how these technologies are reshaping scientific production. To address this gap, Kusumegi et al. assembled five large datasets spanning 2.1 million preprints, 28,000 peer-reviewed reports, and 246 million online views and downloads of scientific documents. Then, using text-based detectors to identify first-time LLM use, they conducted difference-in-differences analyses to compare researchers’ work before and after LLM adoption. They found that LLM adoption increases a researcher’s scientific output by 23.7 – 89.3%, with especially large boosts for authors facing higher writing and language barriers. Kusumegi et al. also discovered that LLM-assisted manuscripts show a reversal of the traditional positive relationship between writing complexity and research quality. After LLM adoption, more sophisticated language was used, but in substantively weak manuscripts. Lastly, the authors show that LLM adopters read and cite more diverse literature, referencing more books, younger works, and less-cited documents. “Our findings show that LLMs have begun to reshape scientific production. These changes portend an evolving research landscape in which the value of English fluency will recede, but the importance of robust quality-assessment frameworks and deep methodological scrutiny is paramount,” write the authors. “For peer reviewers and journal editors, and the community, more broadly, who create, consume, and apply this work, this represents a major issue."

---

Journal

Science

DOI

10.1126/science.adw3000 

Article Title

Scientific production in the era of large language models

Article Publication Date

18-Dec-2025

 

 

University of Cincinnati medical students explore ChatGPT’s ability to support qualitative research

Researchers asked AI assistant to analyze survey responses from their classmates

University of Cincinnati

FacebookXLinkedInWeChatBlueskyMessageWhatsAppEmail

Newly published research from the University of Cincinnati College of Medicine highlights student-led work in medical education and examines how artificial intelligence (AI) can assist with qualitative research.

The study, published in the journal Medical Science Educator, explored whether ChatGPT can support thematic analysis. Corresponding author and third-year medical student Jonathan Bowden, of Pickerington, Ohio, collaborated with fellow third-year medical student Megha Mohanakrishnan, of San Jose, California, to lead the project. Both Bowden and Mohanakrishnan earned their undergraduate degrees in medical sciences at UC.

Both students also served as learning assistants during their first year of medical school for Andrew Thompson, PhD, professor-educator in the Department of Medical Education. Through their work with Thompson, they joined a project analyzing survey responses from fellow first-year medical students about their thoughts and feelings surrounding cadaveric dissection, which is part of their coursework.

The team identified several common themes: “We noted feelings of gratitude toward donors and their families and appreciation and excitement for a valuable learning opportunity,” said Bowden. “We also noted some nervousness and apprehension.”

With their manual analysis serving as the gold standard, the students then evaluated whether AI could perform similar thematic coding.

“So much of this project came from genuine curiosity about whether we, as medical students, could use AI to work more efficiently,” said Bowden. “We chose ChatGPT because it’s free and widely accessible.”

Three methods

The team tested three methods for prompting the AI, running each method three times:

• Method one instructed ChatGPT to code responses using only a list of themes and their definitions.

• Method two added 25 example responses with assigned themes and brief explanations. The AI was told to reference these examples in its coding.

• Method three asked ChatGPT to code each of the 25 example responses individually. After each attempt, the students provided feedback on incorrect or missing themes, and the AI revised its theme definitions accordingly. Once the 25 examples were complete, ChatGPT coded the remaining responses using the updated definitions.

• “We tried to engage more and more with the AI to improve its accuracy,” said Bowden. “Method three had the highest accuracy.”

Reflections

Bowden described the research, which earned the students a national award, as a long process that was helpful in establishing possibilities for future research projects. “I learned how to take on a project from the ground up,” he said. “We gained insight into the planning and execution of a research project.” 

Bowden said he may wish to both practice medicine and teach medical students in the future.

“I appreciate how much work current physicians have done for us and how they are helping us find our passions,” he said. Bowden is currently considering internal medicine for residency.

 

---

Journal

Medical Science Educator

Method of Research

Content analysis

Subject of Research

People

Article Title

Incorporating Artificial Intelligence in Qualitative Research: Exploring the Role of ChatGPT in Thematic Analysis

Article Publication Date

4-Dec-2025

 

For Black women with breast cancer, ultra-processed foods may worsen health outcomes

Rutgers University

FacebookXLinkedInWeChatBlueskyMessageWhatsAppEmail

A study from Rutgers Cancer Institute researchers in eClinicalMedicine is the first to link ultra-processed foods to reduced survival in Black women with breast cancer.

Black patients with breast cancer who ate the most ultra-processed foods before diagnosis were 36-40% more likely to die from their cancers or other causes of death than those who ate the fewest ultra-processed foods.

"Black women have the highest mortality rate from breast cancer compared with other racial or ethnic groups in the U.S.," said Tengteng Wang, lead author of the study and a member of the Cancer Prevention and Control Program at Rutgers Cancer Institute, New Jersey’s only NCI-designated Comprehensive Cancer Center, together with RWJBarnabas Health. "That's why we wanted to see what factors might contribute to these differences."

The researchers followed 1,733 Black women diagnosed with breast cancer in New Jersey between 2005 and 2019 in the Women’s Circle of Health Follow-Up Study (WCHFS), which was led by Elisa Bandera, professor and chief of the Department of Medicine at Rutgers Robert Wood Johnson Medical School. During home interviews conducted about 10 months after diagnosis, participants completed detailed food questionnaires covering the year before their cancer was detected. Researchers then followed up with the women for a median of 9.3 years.

Women who ate the most ultra-processed food averaged more than eight servings per day. Those who ate the least averaged fewer than three servings daily. In addition to 40% higher breast cancer mortality linked with the highest UPF intake, women in the top UPF consumption tier were 36% more likely to die of any cause than those in the lowest tier. 

Ultra-processed foods are industrial formulations that typically contain additives, preservatives and ingredients not used in home cooking. The category includes most sodas, deli meats, sweets/desserts, salty snacks, pre-prepared fast foods/mixed dishes – basically everything in the center aisles of a supermarket – and it now accounts for roughly 40-60% of calories in the American diet.

The findings echo the only other study to examine whether ultra-processed foods are associated with cancer death. Analysis of the UK Biobank, which tracks a predominantly white United Kingdom population, found a 22% increased risk of cancer-related death among cancer survivors who were high consumers of ultra-processed foods across all cancer types. The consistency suggests the potential underlying biological mechanism isn’t race-specific.

"The average consumption is very similar to the U.K. population and also other U.S.-based studies like the Nurses' Health Study," said Wang, who is also an assistant professor at Rutgers Robert Wood Johnson Medical School. "I’d like to see more studies conducted among cancer survivors to confirm this, but the current evidence looks biologically reasonable now, particularly considering we have some ideas about the underlying biological mechanisms.”

One link that connects them to cancer seems to be that they encourage people to overeat, which spurs weight gain and metabolic dysfunction. When Wang’s team statistically adjusted for total calorie intake, the association between ultra-processed foods and mortality weakened considerably.

"The total energy intake may be one of the mechanisms, but it’s not the only one because a positive association existed even after adjusting for caloric intake," Wang said.

Her team is conducting parallel analyses examining other potential pathways, including inflammation and insulin resistance. One forthcoming study will look at breast cancer tissues taken from the same women to see whether ultra-processed food consumption correlates with activation of the mammalian target of rapamycin signaling pathway, a cellular system involved in tumor growth. Another will examine blood-based inflammation markers such as C-reactive protein and interleukin-6.

In the recently published study, when Wang’s team examined whether particular foods were more associated with cancer deaths than others, processed meats emerged as the leading culprit. That finding aligns with previous research linking processed meat to cancer risk and poorer prognosis. 

"Maybe it's too complicated for breast cancer patients to think about how to reduce consumption of ultra-processed foods in general," Wang said. "But we find processed meat is the top worst contributor among all UPF subgroups. So maybe a more detailed takeaway is to avoid this one thing.”

The findings arrive as ultra-processed foods face mounting scrutiny from public health researchers. A major umbrella review published last year in The BMJ found consistent associations between high ultra-processed food consumption and dozens of adverse health outcomes, including heart disease, diabetes, depression and premature death.

Wang's research adds breast cancer survivorship to that list and, for anyone navigating life after a cancer diagnosis, points toward a modifiable risk factor.  The study examined an understudied population, but researchers suspect the findings apply broadly.

"If you can't do everything," Wang said, "at least limit consumption of processed meat." 

“Going back to cooking traditional meals can potentially save money and is generally better for your health,” said Bandera, who is also co-leader of the Cancer Prevention and Control Research Program at Rutgers Cancer Institute.

---

Journal

EClinicalMedicine

DOI

10.1016/j.eclinm.2025.103700 

Method of Research

Data/statistical analysis

Subject of Research

People

Article Title

Ultra-processed foods consumption and subsequent mortality in a cohort of Black breast cancer survivors

Article Publication Date

17-Dec-2025

REGULATION PROTECTS

Q&A: Environmental protection benefits the American economy and public health

Penn State

FacebookXLinkedInWeChatBlueskyMessageWhatsAppEmail

UNIVERSITY PARK, Pa. — The benefits of environmental protection, measured in terms of advancing public health and the economy, outweigh the costs associated with implementing and enforcing environmental regulations, according to researchers at Penn State who published a new perspective piece in the journal Nature Water.

The researchers used data from the U.S. Environmental Protection Agency (EPA) to see if the benefits of federal environmental regulations, such as the estimated number of asthma attacks prevented by air pollution regulations, outweigh the costs to comply with and enforce environmental protection rules. The team found that the benefits — calculated by looking at medical costs and financial losses from missed work, decreasing premature deaths and reducing the prevalence of diseases like lung and bladder cancer and cardiovascular issues — are up to five and six times higher than the costs associated with implementing the rules.

In the following Q&A, study co-author Onur Apul, associate professor of civil and environmental engineering at Penn State, discussed the societal impacts of environmental protection

Q: What do you mean when you say “environmental protection”?

Apul: The U.S. EPA protects human health and the environment by developing and enforcing regulations based on environmental laws passed by Congress. In addition, EPA conducts scientific research, sets national standards for air, water and soil quality, and works with states to address local pollution issues and minimize environmental risks.

Q: How have environmental regulations impacted public health and the economy over the past 50 years?

Apul: The message of our article is very simple: Environmental protection is good for public health and the American economy. Protecting our environment is not an expense, it's cost-saving. It’s like conserving energy in your house: It may look expensive at first — you have to install nice windows, maybe insulated heat exchangers — but in the long term you're going to be saving energy and saving money. 

In our article, we went to the EPA’s databases and extracted how much it costs, for example, to remediate arsenic from drinking water versus implementing and enforcing a federal-level rule regulating arsenic in drinking water. Then we looked at EPA benefit computations, like how much does it cost to lose a young person from the workforce because of a sickness? We did a summary of all major EPA environmental rules and concluded that environmental protection is a cost-savings. In some cases, it's five times; in some cases, 10 times more beneficial than the costs associated with implementing and enforcing the regulations.

For example, the Clean Air Act is predicted to prevent 100 million asthma attacks by 2050. What's the average cost or benefit of not having an asthma attack? This corresponds to about 200,000 fatalities prevented. Or look at the Lead and Copper Rule, which regulates lead and copper in drinking water and prevents almost 1 million infants per year from having low birth weight.

Q: You noted that for the past 20 years, about 24 new chemicals have been registered every minute. How many chemicals does the EPA regulate?

Apul: Think about how many chemicals you use daily. Think about the paints, pesticides, herbicides, antibiotics, pharmaceuticals and over-the-counter painkillers people use each day. The American Chemical Society Chemical Abstract Service lists about 275 million chemicals. That means 24 new chemicals have been registered every minute over the past 20 years, but the EPA regulates fewer than 100 contaminants out of 275 million. Advancing science, advancing engineering, increases life expectancy and environmental protection is one of those advancements, improving our quality of life by preventing child deaths, unexpected disease-related deaths and pollution-related deaths.

Q: How much does it cost to run the EPA?

Apul: The EPA’s operating budget is tiny compared to the total federal budget and the country’s GDP. The EPA’s budget for fiscal year 2024 was about $10 billion, or less than 0.15% of total government spending. That’s 0.03% of U.S. GDP. And in 2024 the EPA employed only about 15,000 people who regulate all these compounds, contribute to research, contribute to policy making, engage with communities and direct messaging. This small number of people, working on a small budget compared to total government spending and the nation’s GDP, help to protect the environment, preserve natural resources and save lives.

Other co-authors are Penn State graduate students Macy Hannan and Dilara Hatinoglu; Lee Blaney, University of Maryland Baltimore County; Peter Vikesland, Virginia Tech; Detlef Knappe, North Carolina State University; and Reed Miller, University of Maine.

---

Journal

Nature Water

DOI

10.1038/s44221-025-00554-0 

Article Title

Protecting the environment pays off for the USA economy and public health

Article Publication Date

11-Dec-2025