Catherine Stinson

Aarohi Srivastava, Abhinav Rastogi, Abhishek Rao et al. · allen-ai, amazon-science

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet poorly characterized. In order to inform future research, prepare for disruptive new model capabilities, and ameliorate socially harmful effects, it is vital that we understand the present and near-future capabilities and limitations of language models. To address this challenge, we introduce the Beyond the Imitation Game benchmark (BIG-bench). BIG-bench currently consists of 204 tasks, contributed by 450 authors across 132 institutions. Task topics are diverse, drawing problems from linguistics, childhood development, math, common-sense reasoning, biology, physics, social bias, software development, and beyond. BIG-bench focuses on tasks that are believed to be beyond the capabilities of current language models. We evaluate the behavior of OpenAI's GPT models, Google-internal dense transformer architectures, and Switch-style sparse transformers on BIG-bench, across model sizes spanning millions to hundreds of billions of parameters. In addition, a team of human expert raters performed all tasks in order to provide a strong baseline. Findings include: model performance and calibration both improve with scale, but are poor in absolute terms (and when compared with rater performance); performance is remarkably similar across model classes, though with benefits from sparsity; tasks that improve gradually and predictably commonly involve a large knowledge or memorization component, whereas tasks that exhibit "breakthrough" behavior at a critical scale often involve multiple steps or components, or brittle metrics; social bias typically increases with scale in settings with ambiguous context, but this can be improved with prompting.

Mohamed Abdalla, Sahar Abdalla, Alicia Cappello et al.

Workplace norms in computer science have received growing attention due to a series of recent ethical scandals. One type of response has been a push to improve the ethics education provided to computer science students. Evidence for the effectiveness of ethics education remains mixed; some evidence suggests that norms are changing, while gaps between stated values and practice remain. Our focus here is on whether students, who have received some contemporary CS ethics education, are able to effectively apply ethical reasoning to their own decision-making in what is typically the first significant ethical decision of their careers: their job search. Our study examines the ethical decision making of 129 computer science students and recent graduates during their job searches. We find that most students prioritize factors like compensation, location, and workplace culture over ethical and social issues. Even when expressing ethical concerns, respondents often justify taking actions contradicting their moral views through commonly-shared explanations such as desire to make money or the perceived inability to avoid unethical workplaces. This work sheds light on the disconnect between ethics education and real-world CS graduate decision making. We offer insights for evolving curricula to better address practical ethical dilemmas, with implications for educators and industry.

Rina Khan, Annabelle Sauve, Imaan Bayoumi et al.

Artificial intelligence (AI) in healthcare has led to many promising developments; however, increasingly, AI research is funded by the private sector leading to potential trade-offs between benefits to patients and benefits to industry. Health AI practitioners should prioritize successful adaptation into clinical practice in order to provide meaningful benefits to patients, but translation usually requires collaboration with industry. We discuss three features of AI studies that hamper the integration of AI into clinical practice from the perspective of researchers and clinicians. These include lack of clinically relevant metrics, lack of clinical trials and longitudinal studies to validate results, and lack of patient and physician involvement in the development process. For partnerships between industry and health research to be sustainable, a balance must be established between patient and industry benefit. We propose three approaches for addressing this gap: improved transparency and explainability of AI models, fostering relationships with industry partners that have a reputation for centering patient benefit in their practices, and prioritization of overall healthcare benefits. With these priorities, we can sooner realize meaningful AI technologies used by clinicians where mutua

Ernesto Lang Oreamuno, Rohan Faiyaz Khan, Abdul Ali Bangash et al.

Model stores offer third-party ML models and datasets for easy project integration, minimizing coding efforts. One might hope to find detailed specifications of these models and datasets in the documentation, leveraging documentation standards such as model and dataset cards. In this study, we use statistical analysis and hybrid card sorting to assess the state of the practice of documenting model cards and dataset cards in one of the largest model stores in use today--Hugging Face (HF). Our findings show that only 21,902 models (39.62\%) and 1,925 datasets (28.48\%) have documentation. Furthermore, we observe inconsistency in ethics and transparency-related documentation for ML models and datasets.

Will Aitken, Mohamed Abdalla, Karen Rudie et al.

Impressive performance of pre-trained models has garnered public attention and made news headlines in recent years. Almost always, these models are produced by or in collaboration with industry. Using them is critical for competing on natural language processing (NLP) benchmarks and correspondingly to stay relevant in NLP research. We surveyed 100 papers published at EMNLP 2022 to determine the degree to which researchers rely on industry models, other artifacts, and contributions to publish in prestigious NLP venues and found that the ratio of their citation is at least three times greater than what would be expected. Our work serves as a scaffold to enable future researchers to more accurately address whether: 1) Collaboration with industry is still collaboration in the absence of an alternative or 2) if NLP inquiry has been captured by the motivations and research direction of private corporations.

Rina Khan, Catherine Stinson

Facial expression recognition (FER) algorithms classify facial expressions into emotions such as happy, sad, or angry. An evaluative challenge facing FER algorithms is the fall in performance when detecting spontaneous expressions compared to posed expressions. An ethical (and evaluative) challenge facing FER algorithms is that they tend to perform poorly for people of some races and skin colors. These challenges are linked to the data collection practices employed in the creation of FER datasets. In this study, we audit two state-of-the-art FER datasets. We take random samples from each dataset and examine whether images are spontaneous or posed. In doing so, we propose a methodology for identifying spontaneous or posed images. We discover a significant number of images that were posed in the datasets purporting to consist of in-the-wild images. Since performance of FER models vary between spontaneous and posed images, the performance of models trained on these datasets will not represent the true performance if such models were to be deployed in in-the-wild applications. We also observe the skin color of individuals in the samples, and test three models trained on each of the datasets to predict facial expressions of people from various races and skin tones. We find that the FER models audited were more likely to predict people labeled as not white or determined to have dark skin as showing a negative emotion such as anger or sadness even when they were smiling. This bias makes such models prone to perpetuate harm in real life applications.

Nicola Neophytou, Bhaskar Mitra, Catherine Stinson

Recommendation algorithms are susceptible to popularity bias: a tendency to recommend popular items even when they fail to meet user needs. A related issue is that the recommendation quality can vary by demographic groups. Marginalized groups or groups that are under-represented in the training data may receive less relevant recommendations from these algorithms compared to others. In a recent study, Ekstrand et al. investigate how recommender performance varies according to popularity and demographics, and find statistically significant differences in recommendation utility between binary genders on two datasets, and significant effects based on age on one dataset. Here we reproduce those results and extend them with additional analyses. We find statistically significant differences in recommender performance by both age and gender. We observe that recommendation utility steadily degrades for older users, and is lower for women than men. We also find that the utility is higher for users from countries with more representation in the dataset. In addition, we find that total usage and the popularity of consumed content are strong predictors of recommender performance and also vary significantly across demographic groups.

Catherine Stinson

Discussions of algorithmic bias tend to focus on examples where either the data or the people building the algorithms are biased. This gives the impression that clean data and good intentions could eliminate bias. The neutrality of the algorithms themselves is defended by prominent Artificial Intelligence researchers. However, algorithms are not neutral. In addition to biased data and biased algorithm makers, AI algorithms themselves can be biased. This is illustrated with the example of collaborative filtering, which is known to suffer from popularity, and homogenizing biases. Iterative information filtering algorithms in general create a selection bias in the course of learning from user responses to documents that the algorithm recommended. These are not merely biases in the statistical sense; these statistical biases can cause discriminatory outcomes. Data points on the margins of distributions of human data tend to correspond to marginalized people. Popularity and homogenizing biases have the effect of further marginalizing the already marginal. This source of bias warrants serious attention given the ubiquity of algorithmic decision-making.