Large language models that replace human participants can harmfully misportray and flatten identity groups

Nature Machine Intelligence, Год журнала: 2025, Номер unknown

Опубликована: Фев. 17, 2025

Язык: Английский

---

Real Risks of Fake Data: Synthetic Data, Diversity-Washing and Consent Circumvention

2022 ACM Conference on Fairness, Accountability, and Transparency, Год журнала: 2024, Номер 86, С. 1733 - 1744

Опубликована: Июнь 3, 2024

Machine learning systems require representations of the real world for training and testing - they data, lots it. Collecting data at scale has logistical ethical challenges, synthetic promises a solution to these challenges. Instead needing collect photos people's faces train facial recognition system, model creator could create use photo-realistic, faces. The comparative ease generating this rather than relying on collecting made it common practice. We present two key risks using in development. First, we detail high risk false confidence when increase dataset diversity representation. base examination use-case where datasets were generated an evaluation technology. Second, examine how circumventing consent usage. illustrate by considering importance U.S. Federal Trade Commission's regulation collection affected models. Finally, discuss exemplify complicates existing governance practice; decoupling from those impacts, is prone consolidating power away most impacted algorithmically-mediated harm.

Язык: Английский

---

Matching GPT-simulated populations with real ones in psychological studies - the case of the EPQR-A personality test

ACM Transactions on Computing for Healthcare, Год журнала: 2025, Номер unknown

Опубликована: Янв. 23, 2025

This article analyzes how well OpenAI's LLM GPT-4 can emulate different personalities and simulate populations to answer psychological questionnaires similarly real population samples. For this purpose, we performed experiments with the Eysenck Personality Questionnaire-Revised Abbreviated (EPQR-A) in three languages (Spanish, English, Slovak). The EPQR-A measures personality on four scales: extraversion (E: sociability), neuroticism (N: emotional stability), psychoticism (P: tendency break social rules, not having empathy), lying (L: desirability). We perform a comparative analysis of answers synthetic those two samples Spanish students as unconditioned baseline GPT. Furthermore, impact time (what year questionnaire is answered), language, student age gender are analyzed. To our knowledge, first test has been used assess GPT´s language versions measured. Our reveals that exhibits an extroverted, emotionally stable low levels high desirability. replicates some differences observed terms but only partially results for populations.

Язык: Английский

---

HCI for AGI

interactions, Год журнала: 2025, Номер 32(2), С. 26 - 32

Опубликована: Фев. 25, 2025

---

Transforming Human-AI Collaboration using “Large Whatever Models” (LWMs)

Опубликована: Апрель 23, 2025

Язык: Английский

---

Forging an HCI Research Agenda with Artists Impacted by Generative AI

Опубликована: Апрель 23, 2025

Язык: Английский

---

A Bayesian Exploration on the Motivational and Behavioral Impacts of Chatbots in Language Learning

Опубликована: Апрель 23, 2025

Язык: Английский

---

"Don't You Dare Go Hollow": How Dark Souls Helps Players Cope with Depression, a Thematic Analysis of Reddit Discussions

Опубликована: Апрель 24, 2025

Язык: Английский

---

Large Language Models Amplify Human Biases in Moral Decision-Making

Опубликована: Июнь 9, 2024

As large language models (LLMs) become more widely used, people increasingly rely on them to make or advise moral decisions. Some researchers even propose using LLMs as participants in psychology experiments. It is therefore important understand how well decisions and they compare humans. We investigated this question realistic dilemmas prompts where GPT-4, Llama 3, Claude 3 give advice emulate a research participant. In Study 1, we compared responses from representative US sample (N = 285) for 22 dilemmas: social that pitted self-interest against the greater good, utilitarian cost-benefit reasoning deontological rules. dilemmas, were altruistic than participants. exhibited stronger omission bias participants: usually endorsed inaction over action. 2 490, preregistered), replicated document an additional bias: unlike humans, (except GPT-4o) tended answer ``no'' whereby phrasing of influences decision when physical action remains same. Our findings show LLM decision-making amplifies human biases introduces potentially problematic biases.

Язык: Английский

---

Connecting algorithmic fairness to quality dimensions in machine learning in official statistics and survey production

AStA Wirtschafts- und Sozialstatistisches Archiv, Год журнала: 2024, Номер 18(2), С. 131 - 184

Опубликована: Июнь 1, 2024

Abstract National Statistical Organizations (NSOs) increasingly draw on Machine Learning (ML) to improve the timeliness and cost-effectiveness of their products. When introducing ML solutions, NSOs must ensure that high standards with respect robustness, reproducibility, accuracy are upheld as codified, e.g., in Quality Framework for Algorithms (QF4SA; Yung et al. 2022, Journal IAOS ). At same time, a growing body research focuses fairness pre-condition safe deployment prevent disparate social impacts practice. However, has not yet been explicitly discussed quality aspect context application at NSOs. We employ QF4SA framework present mapping its dimensions algorithmic fairness. thereby extend several ways: First, we investigate interaction each these dimensions. Second, argue own, additional dimension, beyond what is contained so far. Third, emphasize address data, both own applied methodology. In parallel empirical illustrations, show how our can contribute methodology domains official statistics, fairness, trustworthy machine learning. Little no prior knowledge ML, statistics required provide introductions subjects. These also targeted discussion

Язык: Английский

---