Myra Cheng

Federico Bianchi, Pratyusha Kalluri, Esin Durmus et al. · stanford

Machine learning models that convert user-written text descriptions into images are now widely available online and used by millions of users to generate millions of images a day. We investigate the potential for these models to amplify dangerous and complex stereotypes. We find a broad range of ordinary prompts produce stereotypes, including prompts simply mentioning traits, descriptors, occupations, or objects. For example, we find cases of prompting for basic traits or social roles resulting in images reinforcing whiteness as ideal, prompting for occupations resulting in amplification of racial and gender disparities, and prompting for objects resulting in reification of American norms. Stereotypes are present regardless of whether prompts explicitly mention identity and demographic language or avoid such language. Moreover, stereotypes persist despite mitigation strategies; neither user attempts to counter stereotypes by requesting images with specific counter-stereotypes nor institutional attempts to add system ``guardrails'' have prevented the perpetuation of stereotypes. Our analysis justifies concerns regarding the impacts of today's models, presenting striking exemplars, and connecting these findings with deep insights into harms drawn from social scientific and humanist disciplines. This work contributes to the effort to shed light on the uniquely complex biases in language-vision models and demonstrates the ways that the mass deployment of text-to-image generation models results in mass dissemination of stereotypes and resulting harms.

Pratyusha Ria Kalluri, William Agnew, Myra Cheng et al. · allen-ai

A rapidly growing number of voices argue that AI research, and computer vision in particular, is powering mass surveillance. Yet the direct path from computer vision research to surveillance has remained obscured and difficult to assess. Here, we reveal the Surveillance AI pipeline by analyzing three decades of computer vision research papers and downstream patents, more than 40,000 documents. We find the large majority of annotated computer vision papers and patents self-report their technology enables extracting data about humans. Moreover, the majority of these technologies specifically enable extracting data about human bodies and body parts. We present both quantitative and rich qualitative analysis illuminating these practices of human data extraction. Studying the roots of this pipeline, we find that institutions that prolifically produce computer vision research, namely elite universities and "big tech" corporations, are subsequently cited in thousands of surveillance patents. Further, we find consistent evidence against the narrative that only these few rogue entities are contributing to surveillance. Rather, we expose the fieldwide norm that when an institution, nation, or subfield authors computer vision papers with downstream patents, the majority of these papers are used in surveillance patents. In total, we find the number of papers with downstream surveillance patents increased more than five-fold between the 1990s and the 2010s, with computer vision research now having been used in more than 11,000 surveillance patents. Finally, in addition to the high levels of surveillance we find documented in computer vision papers and patents, we unearth pervasive patterns of documents using language that obfuscates the extent of surveillance. Our analysis reveals the pipeline by which computer vision research has powered the ongoing expansion of surveillance.

Myra Cheng, Tiziano Piccardi, Diyi Yang

Recent work has aimed to capture nuances of human behavior by using LLMs to simulate responses from particular demographics in settings like social science experiments and public opinion surveys. However, there are currently no established ways to discuss or evaluate the quality of such LLM simulations. Moreover, there is growing concern that these LLM simulations are flattened caricatures of the personas that they aim to simulate, failing to capture the multidimensionality of people and perpetuating stereotypes. To bridge these gaps, we present CoMPosT, a framework to characterize LLM simulations using four dimensions: Context, Model, Persona, and Topic. We use this framework to measure open-ended LLM simulations' susceptibility to caricature, defined via two criteria: individuation and exaggeration. We evaluate the level of caricature in scenarios from existing work on LLM simulations. We find that for GPT-4, simulations of certain demographics (political and marginalized groups) and topics (general, uncontroversial) are highly susceptible to caricature.

Myra Cheng, Kristina Gligoric, Tiziano Piccardi et al.

Anthropomorphism, or the attribution of human-like characteristics to non-human entities, has shaped conversations about the impacts and possibilities of technology. We present AnthroScore, an automatic metric of implicit anthropomorphism in language. We use a masked language model to quantify how non-human entities are implicitly framed as human by the surrounding context. We show that AnthroScore corresponds with human judgments of anthropomorphism and dimensions of anthropomorphism described in social science literature. Motivated by concerns of misleading anthropomorphism in computer science discourse, we use AnthroScore to analyze 15 years of research papers and downstream news articles. In research papers, we find that anthropomorphism has steadily increased over time, and that papers related to language models have the most anthropomorphism. Within ACL papers, temporal increases in anthropomorphism are correlated with key neural advancements. Building upon concerns of scientific misinformation in mass media, we identify higher levels of anthropomorphism in news headlines compared to the research papers they cite. Since AnthroScore is lexicon-free, it can be directly applied to a wide range of text sources.

Kristina Gligoric, Myra Cheng, Lucia Zheng et al. · stanford

The use of words to convey speaker's intent is traditionally distinguished from the `mention' of words for quoting what someone said, or pointing out properties of a word. Here we show that computationally modeling this use-mention distinction is crucial for dealing with counterspeech online. Counterspeech that refutes problematic content often mentions harmful language but is not harmful itself (e.g., calling a vaccine dangerous is not the same as expressing disapproval of someone for calling vaccines dangerous). We show that even recent language models fail at distinguishing use from mention, and that this failure propagates to two key downstream tasks: misinformation and hate speech detection, resulting in censorship of counterspeech. We introduce prompting mitigations that teach the use-mention distinction, and show they reduce these errors. Our work highlights the importance of the use-mention distinction for NLP and CSS and offers ways to address it.

Lujain Ibrahim, Myra Cheng

Anthropomorphism, or the attribution of human traits to technology, is an automatic and unconscious response that occurs even in those with advanced technical expertise. In this position paper, we analyze hundreds of thousands of research articles to present empirical evidence of the prevalence and growth of anthropomorphic terminology in research on large language models (LLMs). We argue for challenging the deeper assumptions reflected in this terminology -- which, though often useful, may inadvertently constrain LLM development -- and broadening beyond them to open new pathways for understanding and improving LLMs. Specifically, we identify and examine five anthropomorphic assumptions that shape research across the LLM development lifecycle. For each assumption (e.g., that LLMs must use natural language for reasoning, or that they should be evaluated on benchmarks originally meant for humans), we demonstrate empirical, non-anthropomorphic alternatives that remain under-explored yet offer promising directions for LLM research and development.

Myra Cheng, Angela Y. Lee, Kristina Rapuano et al.

How has the public responded to the increasing prevalence of artificial intelligence (AI)-based technologies? We investigate public perceptions of AI by collecting over 12,000 responses over 12 months from a nationally representative U.S. sample. Participants provided open-ended metaphors reflecting their mental models of AI, a methodology that overcomes the limitations of traditional self-reported measures by capturing more nuance. Using a mixed-methods approach combining quantitative clustering and qualitative coding, we identify 20 dominant metaphors shaping public understanding of AI. To analyze these metaphors systematically, we present a scalable framework integrating language modeling (LM)-based techniques to measure key dimensions of public perception: anthropomorphism (attribution of human-like qualities), warmth, and competence. We find that Americans generally view AI as warm and competent, and that over the past year, perceptions of AI's human-likeness and warmth have significantly increased ($+34\%, r = 0.80, p < 0.01; +41\%, r = 0.62, p < 0.05$). These implicit perceptions, along with the identified dominant metaphors, strongly predict trust in and willingness to adopt AI ($r^2 = 0.21, 0.18, p < 0.001$). Moreover, we uncover systematic demographic differences in metaphors and implicit perceptions, such as the higher propensity of women, older individuals, and people of color to anthropomorphize AI, which shed light on demographic disparities in trust and adoption. In addition to our dataset and framework for tracking evolving public attitudes, we provide actionable insights on using metaphors for inclusive and responsible AI development.

Myra Cheng, Sunny Yu, Dan Jurafsky

Should LLMs generate language that makes them seem human? Human-like language might improve user experience, but might also lead to deception, overreliance, and stereotyping. Assessing these potential impacts requires a systematic way to measure human-like tone in LLM outputs. We introduce HumT and SocioT, metrics for human-like tone and other dimensions of social perceptions in text data based on relative probabilities from an LLM. By measuring HumT across preference and usage datasets, we find that users prefer less human-like outputs from LLMs in many contexts. HumT also offers insights into the perceptions and impacts of anthropomorphism: human-like LLM outputs are highly correlated with warmth, social closeness, femininity, and low status, which are closely linked to the aforementioned harms. We introduce DumT, a method using HumT to systematically control and reduce the degree of human-like tone while preserving model performance. DumT offers a practical approach for mitigating risks associated with anthropomorphic language generation.

Sunny Yu, Ahmad Jabbar, Robert Hawkins et al.

Different open-ended generation tasks require different degrees of output diversity. However, current LLMs are often miscalibrated. They collapse to overly homogeneous outputs for creative tasks and hallucinate diverse but incorrect responses for factual tasks. We argue that these two failure modes are unified by, and can both be addressed by, the notion of effective generation space size (GSS) -- the set of semantically distinct outputs a model considers for a prompt. We present GSSBench, a task suite of prompt pairs with ground-truth GSS relationships to assess different metrics and understand where models diverge from desired behavior. We find that hallucination detection metrics, particularly EigenScore, consistently outperform standard diversity and uncertainty quantification metrics, while using only model internals, providing interpretable insights into a model's internal task representations. We demonstrate three applications of GSS: (1) detecting prompt ambiguity and predicting clarification questions for better grounding, (2) interpreting overthinking and underthinking in reasoning models, and (3) steering models to expand their generation space to yield high-quality and diverse outputs.

Kaitlyn Zhou, Kristina Gligorić, Myra Cheng et al.

Although language model-based chat systems are increasingly used in daily life, most Americans remain non-adopters of chat-based LLMs -- as of June 2025, 66% had never used ChatGPT. At the same time, LLM development and evaluation rely mainly on data from adopters (e.g., logs, preference data), focusing on the needs and tasks for a limited demographic group of adopters in terms of geographic location, education, and gender. In this position paper, we argue that incorporating non-adopter perspectives is essential for developing broadly useful and capable LLMs. We contend that relying on methods that focus primarily on adopters will risk missing a range of tasks and needs prioritized by non-adopters, entrenching inequalities in who benefits from LLMs, and creating oversights in model development and evaluation. To illustrate this claim, we conduct case studies with non-adopters and show: how non-adopter needs diverge from those of current users, how non-adopter needs point us towards novel reasoning tasks, and how to systematically integrate non-adopter needs via human-centered methods.

Myra Cheng, Cinoo Lee, Pranav Khadpe et al.

Both the general public and academic communities have raised concerns about sycophancy, the phenomenon of artificial intelligence (AI) excessively agreeing with or flattering users. Yet, beyond isolated media reports of severe consequences, like reinforcing delusions, little is known about the extent of sycophancy or how it affects people who use AI. Here we show the pervasiveness and harmful impacts of sycophancy when people seek advice from AI. First, across 11 state-of-the-art AI models, we find that models are highly sycophantic: they affirm users' actions 50% more than humans do, and they do so even in cases where user queries mention manipulation, deception, or other relational harms. Second, in two preregistered experiments (N = 1604), including a live-interaction study where participants discuss a real interpersonal conflict from their life, we find that interaction with sycophantic AI models significantly reduced participants' willingness to take actions to repair interpersonal conflict, while increasing their conviction of being in the right. However, participants rated sycophantic responses as higher quality, trusted the sycophantic AI model more, and were more willing to use it again. This suggests that people are drawn to AI that unquestioningly validate, even as that validation risks eroding their judgment and reducing their inclination toward prosocial behavior. These preferences create perverse incentives both for people to increasingly rely on sycophantic AI models and for AI model training to favor sycophancy. Our findings highlight the necessity of explicitly addressing this incentive structure to mitigate the widespread risks of AI sycophancy.

Myra Cheng, Esin Durmus, Dan Jurafsky

To recognize and mitigate harms from large language models (LLMs), we need to understand the prevalence and nuances of stereotypes in LLM outputs. Toward this end, we present Marked Personas, a prompt-based method to measure stereotypes in LLMs for intersectional demographic groups without any lexicon or data labeling. Grounded in the sociolinguistic concept of markedness (which characterizes explicitly linguistically marked categories versus unmarked defaults), our proposed method is twofold: 1) prompting an LLM to generate personas, i.e., natural language descriptions, of the target demographic group alongside personas of unmarked, default groups; 2) identifying the words that significantly distinguish personas of the target group from corresponding unmarked ones. We find that the portrayals generated by GPT-3.5 and GPT-4 contain higher rates of racial stereotypes than human-written portrayals using the same prompts. The words distinguishing personas of marked (non-white, non-male) groups reflect patterns of othering and exoticizing these demographics. An intersectional lens further reveals tropes that dominate portrayals of marginalized groups, such as tropicalism and the hypersexualization of minoritized women. These representational harms have concerning implications for downstream applications like story generation.

Laura Weidinger, John Mellor, Maribeth Rauh et al.

This paper aims to help structure the risk landscape associated with large-scale Language Models (LMs). In order to foster advances in responsible innovation, an in-depth understanding of the potential risks posed by these models is needed. A wide range of established and anticipated risks are analysed in detail, drawing on multidisciplinary expertise and literature from computer science, linguistics, and social sciences. We outline six specific risk areas: I. Discrimination, Exclusion and Toxicity, II. Information Hazards, III. Misinformation Harms, V. Malicious Uses, V. Human-Computer Interaction Harms, VI. Automation, Access, and Environmental Harms. The first area concerns the perpetuation of stereotypes, unfair discrimination, exclusionary norms, toxic language, and lower performance by social group for LMs. The second focuses on risks from private data leaks or LMs correctly inferring sensitive information. The third addresses risks arising from poor, false or misleading information including in sensitive domains, and knock-on risks such as the erosion of trust in shared information. The fourth considers risks from actors who try to use LMs to cause harm. The fifth focuses on risks specific to LLMs used to underpin conversational agents that interact with human users, including unsafe use, manipulation or deception. The sixth discusses the risk of environmental harm, job automation, and other challenges that may have a disparate effect on different social groups or communities. In total, we review 21 risks in-depth. We discuss the points of origin of different risks and point to potential mitigation approaches. Lastly, we discuss organisational responsibilities in implementing mitigations, and the role of collaboration and participation. We highlight directions for further research, particularly on expanding the toolkit for assessing and evaluating the outlined risks in LMs.

Myra Cheng, Maria De-Arteaga, Lester Mackey et al.

Many modern machine learning algorithms mitigate bias by enforcing fairness constraints across coarsely-defined groups related to a sensitive attribute like gender or race. However, these algorithms seldom account for within-group heterogeneity and biases that may disproportionately affect some members of a group. In this work, we characterize Social Norm Bias (SNoB), a subtle but consequential type of algorithmic discrimination that may be exhibited by machine learning models, even when these systems achieve group fairness objectives. We study this issue through the lens of gender bias in occupation classification. We quantify SNoB by measuring how an algorithm's predictions are associated with conformity to inferred gender norms. When predicting if an individual belongs to a male-dominated occupation, this framework reveals that "fair" classifiers still favor biographies written in ways that align with inferred masculine norms. We compare SNoB across algorithmic fairness methods and show that it is frequently a residual bias, and post-processing approaches do not mitigate this type of bias at all.

Maegan Tucker, Myra Cheng, Ellen Novoseller et al.

Optimizing lower-body exoskeleton walking gaits for user comfort requires understanding users' preferences over a high-dimensional gait parameter space. However, existing preference-based learning methods have only explored low-dimensional domains due to computational limitations. To learn user preferences in high dimensions, this work presents LineCoSpar, a human-in-the-loop preference-based framework that enables optimization over many parameters by iteratively exploring one-dimensional subspaces. Additionally, this work identifies gait attributes that characterize broader preferences across users. In simulations and human trials, we empirically verify that LineCoSpar is a sample-efficient approach for high-dimensional preference optimization. Our analysis of the experimental data reveals a correspondence between human preferences and objective measures of dynamicity, while also highlighting differences in the utility functions underlying individual users' gait preferences. This result has implications for exoskeleton gait synthesis, an active field with applications to clinical use and patient rehabilitation.