New IZA DPs -- Artificial Intelligence
0 views
Skip to first unread message
IZA Publications
Aug 30, 2025, 1:41:35 PMAug 30
to it...@umich.edu
Dear Italo Gutierrez,
These new IZA Discussion Papers are now available online.
DP 18055 - Contractor/Reyes:
Generative AI in Higher Education: Evidence from an Elite College
DP 18062 - Askitas:
The Behavioral Signature of GenAI in Scientific Communication
DP 18070 - Askitas:
Notes on a World with Generative AI
DP 18074 - Si/Meng/Chen/An/Mao/Li/Bateman/Zhang/Fan/Zu/Gong/Zhou/Miao:
Quality, Safety, and Disparities of AI Chatbots in Managing Chronic Diseases: Experimental Evidence
Please find the abstracts and download links below.
You might also be interested in this IZA World of Labor content:
Innovation and employment in the era of artificial intelligence
IZA DP No. 18055
Zara Contractor, Germán Reyes:
Generative AI in Higher Education: Evidence from an Elite College
Abstract:
Generative AI is transforming higher education, yet systematic evidence on student adoption remains limited. Using novel survey data from a selective U.S. college, we document over 80 percent of students using AI academically within two years of ChatGPT's release. Adoption varies across disciplines, demographics, and achievement levels, highlighting AI's potential to reshape educational inequalities. Students predominantly use AI for augmenting learning (e.g., explanations, feedback), but also to automate tasks (e.g., essay generation). Positive perceptions of AI's educational benefits strongly predict adoption. Institutional policies can influence usage patterns but risk creating unintended disparate impacts across student groups due to uneven compliance.
https://docs.iza.org/dp18055.pdf
IZA DP No. 18062
Nikos Askitas:
The Behavioral Signature of GenAI in Scientific Communication
Abstract:
We examine the uptake of GPT-assisted writing in economics working paper abstracts. Using data from the IZA DP series, we detect a clear stylistic shift after the release of ChatGPT-3.5 in March 2023. This shift is evident in core textual metrics—mean word length, type-token ratio, and readability—and reflects growing convergence with machine-generated writing. While the ChatGPT launch was an exogenous shock, adoption is endogenous: authors choose whether to use AI. To capture this behavioral response, we combine stylometric analysis, machine learning classification, and prompt-based similarity testing. Event-study regressions with fixed effects and placebo checks confirm that the change is abrupt, persistent, and not explained by pre-existing trends. A similarity experiment using OpenAI’s API shows that post-ChatGPT abstracts resemble their GPT-optimized versions more closely than pre-ChatGPT resemble theirs. A classifier, trained on these variants, flags a growing share of post-March 2023 texts as GPT-like. Rather than suggesting full automation, our findings indicate selective human–AI augmentation. Our framework generalizes to other contexts such as e.g. resumes, job ads, legal briefs, research proposals, or programming code.
https://docs.iza.org/dp18062.pdf
IZA DP No. 18070
Nikos Askitas:
Notes on a World with Generative AI
Abstract:
Generative AI (GenAI) and Large Language Models (LLMs) are moving into domains once seen as uniquely human: reasoning, synthesis, abstraction, and rhetoric. Addressed to labor economists and informed readers, this paper clarifies what is truly new about LLMs, what is not, and why it matters. Using an analogy to auto-regressive models from economics, we explain their stochastic nature, whose fluency is often mistaken for agency. We place LLMs in the longer history of human–machine outsourcing, from digestion to cognition, and examine disruptive effects on white-collar labor, institutions, and epistemic norms. Risks emerge when synthetic content becomes both product and input, creating feedback loops that erode originality and reliability. Grounding the discussion in conceptual clarity over hype, we argue that while GenAI may substitute for some labor, statistical limits will, probably but not without major disruption, preserve a key role for human judgment. The question is not only how these tools are used, but which tasks we relinquish and how we reallocate expertise in a new division of cognitive labor.
https://docs.iza.org/dp18070.pdf
IZA DP No. 18074
Yafei Si, Yurun Meng, Xi Chen, Ruopeng An, Limin Mao, Bingqin Li, Hazel Bateman, Han Zhang, Hongbin Fan, Jiaqi Zu, Shaoqing Gong, Zhongliang Zhou, Yudong Miao:
Quality, Safety, and Disparities of AI Chatbots in Managing Chronic Diseases: Experimental Evidence
Abstract:
The rapid development of AI solutions reveals opportunities to address the underdiagnosis and poor management of chronic conditions in developing settings. Using the method of simulated patients and experimental designs, we evaluate the quality, safety, and disparity of medical consultation with ERNIE Bot in China among 384 patient-AI trials. ERNIE Bot reached a diagnostic accuracy of 77.3%, correct drug prescriptions of 94.3%, but prescribed high rates of unnecessary medical tests (91.9%) and unnecessary medications (57.8%). Disparities were observed based on patient age and household economic status, with older and wealthier patients receiving more intensive care. Under standardized conditions, ERNIE Bot, ChatGPT, and DeepSeek demonstrated higher diagnostic accuracy but a greater tendency toward overprescription than human physicians. The results suggest the great potential of ERNIE Bot in empowering quality, accessibility, and affordability of healthcare provision in devel oping contexts but also highlight critical risks related to safety and amplification of sociodemographic disparities.
https://docs.iza.org/dp18074.pdf
Please click here to change your subscription status.
If you have trouble downloading the papers, or for any other questions regarding the IZA Discussion Paper Series, contact public...@iza.org.
These new IZA Discussion Papers are now available online.
DP 18055 - Contractor/Reyes:
Generative AI in Higher Education: Evidence from an Elite College
DP 18062 - Askitas:
The Behavioral Signature of GenAI in Scientific Communication
DP 18070 - Askitas:
Notes on a World with Generative AI
DP 18074 - Si/Meng/Chen/An/Mao/Li/Bateman/Zhang/Fan/Zu/Gong/Zhou/Miao:
Quality, Safety, and Disparities of AI Chatbots in Managing Chronic Diseases: Experimental Evidence
Please find the abstracts and download links below.
You might also be interested in this IZA World of Labor content:
Innovation and employment in the era of artificial intelligence
IZA DP No. 18055
Zara Contractor, Germán Reyes:
Generative AI in Higher Education: Evidence from an Elite College
Abstract:
Generative AI is transforming higher education, yet systematic evidence on student adoption remains limited. Using novel survey data from a selective U.S. college, we document over 80 percent of students using AI academically within two years of ChatGPT's release. Adoption varies across disciplines, demographics, and achievement levels, highlighting AI's potential to reshape educational inequalities. Students predominantly use AI for augmenting learning (e.g., explanations, feedback), but also to automate tasks (e.g., essay generation). Positive perceptions of AI's educational benefits strongly predict adoption. Institutional policies can influence usage patterns but risk creating unintended disparate impacts across student groups due to uneven compliance.
https://docs.iza.org/dp18055.pdf
IZA DP No. 18062
Nikos Askitas:
The Behavioral Signature of GenAI in Scientific Communication
Abstract:
We examine the uptake of GPT-assisted writing in economics working paper abstracts. Using data from the IZA DP series, we detect a clear stylistic shift after the release of ChatGPT-3.5 in March 2023. This shift is evident in core textual metrics—mean word length, type-token ratio, and readability—and reflects growing convergence with machine-generated writing. While the ChatGPT launch was an exogenous shock, adoption is endogenous: authors choose whether to use AI. To capture this behavioral response, we combine stylometric analysis, machine learning classification, and prompt-based similarity testing. Event-study regressions with fixed effects and placebo checks confirm that the change is abrupt, persistent, and not explained by pre-existing trends. A similarity experiment using OpenAI’s API shows that post-ChatGPT abstracts resemble their GPT-optimized versions more closely than pre-ChatGPT resemble theirs. A classifier, trained on these variants, flags a growing share of post-March 2023 texts as GPT-like. Rather than suggesting full automation, our findings indicate selective human–AI augmentation. Our framework generalizes to other contexts such as e.g. resumes, job ads, legal briefs, research proposals, or programming code.
https://docs.iza.org/dp18062.pdf
IZA DP No. 18070
Nikos Askitas:
Notes on a World with Generative AI
Abstract:
Generative AI (GenAI) and Large Language Models (LLMs) are moving into domains once seen as uniquely human: reasoning, synthesis, abstraction, and rhetoric. Addressed to labor economists and informed readers, this paper clarifies what is truly new about LLMs, what is not, and why it matters. Using an analogy to auto-regressive models from economics, we explain their stochastic nature, whose fluency is often mistaken for agency. We place LLMs in the longer history of human–machine outsourcing, from digestion to cognition, and examine disruptive effects on white-collar labor, institutions, and epistemic norms. Risks emerge when synthetic content becomes both product and input, creating feedback loops that erode originality and reliability. Grounding the discussion in conceptual clarity over hype, we argue that while GenAI may substitute for some labor, statistical limits will, probably but not without major disruption, preserve a key role for human judgment. The question is not only how these tools are used, but which tasks we relinquish and how we reallocate expertise in a new division of cognitive labor.
https://docs.iza.org/dp18070.pdf
IZA DP No. 18074
Yafei Si, Yurun Meng, Xi Chen, Ruopeng An, Limin Mao, Bingqin Li, Hazel Bateman, Han Zhang, Hongbin Fan, Jiaqi Zu, Shaoqing Gong, Zhongliang Zhou, Yudong Miao:
Quality, Safety, and Disparities of AI Chatbots in Managing Chronic Diseases: Experimental Evidence
Abstract:
The rapid development of AI solutions reveals opportunities to address the underdiagnosis and poor management of chronic conditions in developing settings. Using the method of simulated patients and experimental designs, we evaluate the quality, safety, and disparity of medical consultation with ERNIE Bot in China among 384 patient-AI trials. ERNIE Bot reached a diagnostic accuracy of 77.3%, correct drug prescriptions of 94.3%, but prescribed high rates of unnecessary medical tests (91.9%) and unnecessary medications (57.8%). Disparities were observed based on patient age and household economic status, with older and wealthier patients receiving more intensive care. Under standardized conditions, ERNIE Bot, ChatGPT, and DeepSeek demonstrated higher diagnostic accuracy but a greater tendency toward overprescription than human physicians. The results suggest the great potential of ERNIE Bot in empowering quality, accessibility, and affordability of healthcare provision in devel oping contexts but also highlight critical risks related to safety and amplification of sociodemographic disparities.
https://docs.iza.org/dp18074.pdf
Please click here to change your subscription status.
If you have trouble downloading the papers, or for any other questions regarding the IZA Discussion Paper Series, contact public...@iza.org.
IZA Publications
Aug 30, 2025, 1:43:47 PMAug 30
to italo...@gmail.com
Reply all
Reply to author
Forward
0 new messages