Research Papers

Proxy Tasks and Subjective Measures Can Be Misleading in Evaluating Explainable AI Systems

Zana Buçinca, Phoebe Lin, Krzysztof Z. Gajos +1

Explainable artificially intelligent (XAI) systems form part of sociotechnical systems, e.g., human+AI teams tasked with making decisions. Yet, current XAI systems are rarely evaluated by measuring the performance of human+AI teams on actual decision-making tasks. We conducted two online experiments and one in-person think-aloud study to evaluate two currently common techniques for evaluating XAI systems: (1) using proxy, artificial tasks such as how well humans predict the AI's decision from the given explanations, and (2) using subjective measures of trust and preference as predictors of actual performance. The results of our experiments demonstrate that evaluations with proxy tasks did not predict the results of the evaluations with the actual decision-making tasks. Further, the subjective measures on evaluations with actual decision-making tasks did not predict the objective performance on those same tasks. Our results suggest that by employing misleading evaluation methods, our field may be inadvertently slowing its progress toward developing human+AI teams that can reliably perform better than humans or AIs alone.

Data Cards: Purposeful and Transparent Dataset Documentation for Responsible AI

Mahima Pushkarna, Andrew Zaldivar, Oddur Kjartansson

As research and industry moves towards large-scale models capable of numerous downstream tasks, the complexity of understanding multi-modal datasets that give nuance to models rapidly increases. A clear and thorough understanding of a dataset's origins, development, intent, ethical considerations and evolution becomes a necessary step for the responsible and informed deployment of models, especially those in people-facing contexts and high-risk domains. However, the burden of this understanding often falls on the intelligibility, conciseness, and comprehensiveness of the documentation. It requires consistency and comparability across the documentation of all datasets involved, and as such documentation must be treated as a user-centric product in and of itself. In this paper, we propose Data Cards for fostering transparent, purposeful and human-centered documentation of datasets within the practical contexts of industry and research. Data Cards are structured summaries of essential facts about various aspects of ML datasets needed by stakeholders across a dataset's lifecycle for responsible AI development. These summaries provide explanations of processes and rationales that shape the data and consequently the models, such as upstream sources, data collection and annotation methods; training and evaluation methods, intended use; or decisions affecting model performance. We also present frameworks that ground Data Cards in real-world utility and human-centricity. Using two case studies, we report on desirable characteristics that support adoption across domains, organizational structures, and audience groups. Finally, we present lessons learned from deploying over 20 Data Cards.

Auditing large language models: a three-layered approach

Jakob Mökander, Jonas Schuett, Hannah Rose Kirk +1

Large language models (LLMs) represent a major advance in artificial intelligence (AI) research. However, the widespread use of LLMs is also coupled with significant ethical and social challenges. Previous research has pointed towards auditing as a promising governance mechanism to help ensure that AI systems are designed and deployed in ways that are ethical, legal, and technically robust. However, existing auditing procedures fail to address the governance challenges posed by LLMs, which display emergent capabilities and are adaptable to a wide range of downstream tasks. In this article, we address that gap by outlining a novel blueprint for how to audit LLMs. Specifically, we propose a three-layered approach, whereby governance audits (of technology providers that design and disseminate LLMs), model audits (of LLMs after pre-training but prior to their release), and application audits (of applications based on LLMs) complement and inform each other. We show how audits, when conducted in a structured and coordinated manner on all three levels, can be a feasible and effective mechanism for identifying and managing some of the ethical and social risks posed by LLMs. However, it is important to remain realistic about what auditing can reasonably be expected to achieve. Therefore, we discuss the limitations not only of our three-layered approach but also of the prospect of auditing LLMs at all. Ultimately, this article seeks to expand the methodological toolkit available to technology providers and policymakers who wish to analyse and evaluate LLMs from technical, ethical, and legal perspectives.

Evaluating Explainable AI: Which Algorithmic Explanations Help Users Predict Model Behavior?

Peter Hase, Mohit Bansal

Algorithmic approaches to interpreting machine learning models have proliferated in recent years. We carry out human subject tests that are the first of their kind to isolate the effect of algorithmic explanations on a key aspect of model interpretability, simulatability, while avoiding important confounding experimental factors. A model is simulatable when a person can predict its behavior on new inputs. Through two kinds of simulation tests involving text and tabular data, we evaluate five explanations methods: (1) LIME, (2) Anchor, (3) Decision Boundary, (4) a Prototype model, and (5) a Composite approach that combines explanations from each method. Clear evidence of method effectiveness is found in very few cases: LIME improves simulatability in tabular classification, and our Prototype method is effective in counterfactual simulation tests. We also collect subjective ratings of explanations, but we do not find that ratings are predictive of how helpful explanations are. Our results provide the first reliable and comprehensive estimates of how explanations influence simulatability across a variety of explanation methods and data domains. We show that (1) we need to be careful about the metrics we use to evaluate explanation methods, and (2) there is significant room for improvement in current methods. All our supporting code, data, and models are publicly available at: https://github.com/peterbhase/InterpretableNLP-ACL2020

WinoGrande: An Adversarial Winograd Schema Challenge at Scale

Keisuke Sakaguchi, Ronan Le Bras, Chandra Bhagavatula +1

The Winograd Schema Challenge (WSC) (Levesque, Davis, and Morgenstern 2011), a benchmark for commonsense reasoning, is a set of 273 expert-crafted pronoun resolution problems originally designed to be unsolvable for statistical models that rely on selectional preferences or word associations. However, recent advances in neural language models have already reached around 90% accuracy on variants of WSC. This raises an important question whether these models have truly acquired robust commonsense capabilities or whether they rely on spurious biases in the datasets that lead to an overestimation of the true capabilities of machine commonsense. To investigate this question, we introduce WinoGrande, a large-scale dataset of 44k problems, inspired by the original WSC design, but adjusted to improve both the scale and the hardness of the dataset. The key steps of the dataset construction consist of (1) a carefully designed crowdsourcing procedure, followed by (2) systematic bias reduction using a novel AfLite algorithm that generalizes human-detectable word associations to machine-detectable embedding associations. The best state-of-the-art methods on WinoGrande achieve 59.4-79.1%, which are 15-35% below human performance of 94.0%, depending on the amount of the training data allowed. Furthermore, we establish new state-of-the-art results on five related benchmarks - WSC (90.1%), DPR (93.1%), COPA (90.6%), KnowRef (85.6%), and Winogender (97.1%). These results have dual implications: on one hand, they demonstrate the effectiveness of WinoGrande when used as a resource for transfer learning. On the other hand, they raise a concern that we are likely to be overestimating the true capabilities of machine commonsense across all these benchmarks. We emphasize the importance of algorithmic bias reduction in existing and future benchmarks to mitigate such overestimation.

The utilization of artificial intelligence (AI) applications has experienced tremendous growth in recent years, bringing forth numerous benefits and conveniences. However, this expansion has also provoked ethical concerns, such as privacy breaches, algorithmic discrimination, security and reliability issues, transparency, and other unintended consequences. To determine whether a global consensus exists regarding the ethical principles that should govern AI applications and to contribute to the formation of future regulations, this paper conducts a meta-analysis of 200 governance policies and ethical guidelines for AI usage published by public bodies, academic institutions, private companies, and civil society organizations worldwide. We identified at least 17 resonating principles prevalent in the policies and guidelines of our dataset, released as an open-source database and tool. We present the limitations of performing a global scale analysis study paired with a critical analysis of our findings, presenting areas of consensus that should be incorporated into future regulatory efforts. All components tied to this work can be found in https://nkluge-correa.github.io/worldwide_AI-ethics/

Higher-Order Explanations of Graph Neural Networks via Relevant Walks

Thomas Schnake, Oliver Eberle, Jonas Lederer +4

Graph Neural Networks (GNNs) are a popular approach for predicting graph structured data. As GNNs tightly entangle the input graph into the neural network structure, common explainable AI approaches are not applicable. To a large extent, GNNs have remained black-boxes for the user so far. In this paper, we show that GNNs can in fact be naturally explained using higher-order expansions, i.e. by identifying groups of edges that jointly contribute to the prediction. Practically, we find that such explanations can be extracted using a nested attribution scheme, where existing techniques such as layer-wise relevance propagation (LRP) can be applied at each step. The output is a collection of walks into the input graph that are relevant for the prediction. Our novel explanation method, which we denote by GNN-LRP, is applicable to a broad range of graph neural networks and lets us extract practically relevant insights on sentiment analysis of text data, structure-property relationships in quantum chemistry, and image classification.

Human-centered Explainable AI: Towards a Reflective Sociotechnical Approach

Upol Ehsan, Mark O. Riedl

Explanations--a form of post-hoc interpretability--play an instrumental role in making systems accessible as AI continues to proliferate complex and sensitive sociotechnical systems. In this paper, we introduce Human-centered Explainable AI (HCXAI) as an approach that puts the human at the center of technology design. It develops a holistic understanding of "who" the human is by considering the interplay of values, interpersonal dynamics, and the socially situated nature of AI systems. In particular, we advocate for a reflective sociotechnical approach. We illustrate HCXAI through a case study of an explanation system for non-technical end-users that shows how technical advancements and the understanding of human factors co-evolve. Building on the case study, we lay out open research questions pertaining to further refining our understanding of "who" the human is and extending beyond 1-to-1 human-computer interactions. Finally, we propose that a reflective HCXAI paradigm-mediated through the perspective of Critical Technical Practice and supplemented with strategies from HCI, such as value-sensitive design and participatory design--not only helps us understand our intellectual blind spots, but it can also open up new design and research spaces.

Quantus: An Explainable AI Toolkit for Responsible Evaluation of Neural Network Explanations and Beyond

Anna Hedström, Leander Weber, Dilyara Bareeva +5

The evaluation of explanation methods is a research topic that has not yet been explored deeply, however, since explainability is supposed to strengthen trust in artificial intelligence, it is necessary to systematically review and compare explanation methods in order to confirm their correctness. Until now, no tool with focus on XAI evaluation exists that exhaustively and speedily allows researchers to evaluate the performance of explanations of neural network predictions. To increase transparency and reproducibility in the field, we therefore built Quantus -- a comprehensive, evaluation toolkit in Python that includes a growing, well-organised collection of evaluation metrics and tutorials for evaluating explainable methods. The toolkit has been thoroughly tested and is available under an open-source license on PyPi (or on https://github.com/understandable-machine-intelligence-lab/Quantus/).

AI Safety Gridworlds

Jan Leike, Miljan Martic, Victoria Krakovna +5

We present a suite of reinforcement learning environments illustrating various safety properties of intelligent agents. These problems include safe interruptibility, avoiding side effects, absent supervisor, reward gaming, safe exploration, as well as robustness to self-modification, distributional shift, and adversaries. To measure compliance with the intended safe behavior, we equip each environment with a performance function that is hidden from the agent. This allows us to categorize AI safety problems into robustness and specification problems, depending on whether the performance function corresponds to the observed reward function. We evaluate A2C and Rainbow, two recent deep reinforcement learning agents, on our environments and show that they are not able to solve them satisfactorily.

Detecting Emergent Intersectional Biases: Contextualized Word Embeddings Contain a Distribution of Human-like Biases

Wei Guo, Aylin Caliskan

With the starting point that implicit human biases are reflected in the statistical regularities of language, it is possible to measure biases in English static word embeddings. State-of-the-art neural language models generate dynamic word embeddings dependent on the context in which the word appears. Current methods measure pre-defined social and intersectional biases that appear in particular contexts defined by sentence templates. Dispensing with templates, we introduce the Contextualized Embedding Association Test (CEAT), that can summarize the magnitude of overall bias in neural language models by incorporating a random-effects model. Experiments on social and intersectional biases show that CEAT finds evidence of all tested biases and provides comprehensive information on the variance of effect magnitudes of the same bias in different contexts. All the models trained on English corpora that we study contain biased representations. Furthermore, we develop two methods, Intersectional Bias Detection (IBD) and Emergent Intersectional Bias Detection (EIBD), to automatically identify the intersectional biases and emergent intersectional biases from static word embeddings in addition to measuring them in contextualized word embeddings. We present the first algorithmic bias detection findings on how intersectional group members are strongly associated with unique emergent biases that do not overlap with the biases of their constituent minority identities. IBD and EIBD achieve high accuracy when detecting the intersectional and emergent biases of African American females and Mexican American females. Our results indicate that biases at the intersection of race and gender associated with members of multiple minority groups, such as African American females and Mexican American females, have the highest magnitude across all neural language models.

Large Language Models to Identify Social Determinants of Health in Electronic Health Records

Marco Guevara, Shan Chen, Spencer Thomas +11

Social determinants of health (SDoH) have an important impact on patient outcomes but are incompletely collected from the electronic health records (EHR). This study researched the ability of large language models to extract SDoH from free text in EHRs, where they are most commonly documented, and explored the role of synthetic clinical text for improving the extraction of these scarcely documented, yet extremely valuable, clinical data. 800 patient notes were annotated for SDoH categories, and several transformer-based models were evaluated. The study also experimented with synthetic data generation and assessed for algorithmic bias. Our best-performing models were fine-tuned Flan-T5 XL (macro-F1 0.71) for any SDoH, and Flan-T5 XXL (macro-F1 0.70). The benefit of augmenting fine-tuning with synthetic data varied across model architecture and size, with smaller Flan-T5 models (base and large) showing the greatest improvements in performance (delta F1 +0.12 to +0.23). Model performance was similar on the in-hospital system dataset but worse on the MIMIC-III dataset. Our best-performing fine-tuned models outperformed zero- and few-shot performance of ChatGPT-family models for both tasks. These fine-tuned models were less likely than ChatGPT to change their prediction when race/ethnicity and gender descriptors were added to the text, suggesting less algorithmic bias (p<0.05). At the patient-level, our models identified 93.8% of patients with adverse SDoH, while ICD-10 codes captured 2.0%. Our method can effectively extracted SDoH information from clinic notes, performing better compare to GPT zero- and few-shot settings. These models could enhance real-world evidence on SDoH and aid in identifying patients needing social support.