We focus on explainability as a desideratum for automated decision-making systems, rather than only models. Although the explainable artificial intelligence (XAI) paradigm offers an impressive variety of solutions to increase the transparency of automated decisions, XAI contributions rarely account for the complete systems—social and institutional environments—where models operate. Our work focuses on one such system in the domain of social welfare, which increasingly turns to automated decision-making to carry out targeted digital surveillance. Specifically, we present a case study of a black-box machine learning model previously used in a major Dutch city to support its officials in the task of detecting fraud. Employing analyses established in the field of system safety, we identify five types of hazards that could have occurred after the introduction of the model. For each of them, we reason about the potential value of XAI interventions as hazard mitigation strategies. The case study illustrates how the deployment of models may impact processes that exist far upstream and downstream from their decision logic, making explainability and/or interpretability insufficient to guarantee the systems’ safe operation. In many cases, XAI techniques may only be able to reasonably address a small fraction of hazards related to the use of algorithms; several major hazards that we identify would have still posed risks if the system had relied on an interpretable model. Thus, we empirically demonstrate that the values, which lie at the heart of XAI research, such as responsibility, safety, or transparency, ultimately necessitate a broader outlook on automated decision-making systems.

Developments in the field of Artificial Intelligence (AI), and particularly large language models (LLMs), have created a 'perfect storm' for observing 'sparks' of Artificial General Intelligence (AGI) that are spurious. Like simpler models, LLMs distill meaningful representations in their latent embeddings that have been shown to correlate with external variables. Nonetheless, the correlation of such representations has often been linked to human-like intelligence in the latter but not the former. We probe models of varying complexity including random projections, matrix decompositions, deep autoencoders and transformers: all of them successfully distill information that can be used to predict latent or external variables and yet none of them have previously been linked to AGI. We argue and empirically demonstrate that the finding of meaningful patterns in latent spaces of models cannot be seen as evidence in favor of AGI. Additionally, we review literature from the social sciences that shows that humans are prone to seek such patterns and anthropomorphize. We conclude that both the methodological setup and common public image of AI are ideal for the misinterpretation that correlations between model representations and some variables of interest are 'caused' by the model's understanding of underlying 'ground truth' relationships. We, therefore, call for the academic community to exercise extra caution, and to be keenly aware of principles of academic integrity, in interpreting and communicating about AI research outcomes.

Counterfactual explanations offer an intuitive and straightforward way to explain black-box models and offer algorithmic recourse to individuals. To address the need for plausible explanations, existing work has primarily relied on surrogate models to learn how the input data is distributed. This effectively reallocates the task of learning realistic explanations for the data from the model itself to the surrogate. Consequently, the generated explanations may seem plausible to humans but need not necessarily describe the behaviour of the black-box model faithfully. We formalize this notion of faithfulness through the introduction of a tailored evaluation metric and propose a novel algorithmic framework for generating Energy-Constrained Conformal Counterfactuals that are only as plausible as the model permits. Through extensive empirical studies, we demonstrate that ECCCo reconciles the need for faithfulness and plausibility. In particular, we show that for models with gradient access, it is possible to achieve state-of-the-art performance without the need for surrogate models. To do so, our framework relies solely on properties defining the black-box model itself by leveraging recent advances in energy-based modelling and conformal prediction. To our knowledge, this is the first venture in this direction for generating faithful counterfactual explanations. Thus, we anticipate that ECCCo can serve as a baseline for future research. We believe that our work opens avenues for researchers and practitioners seeking tools to better distinguish trustworthy from unreliable models.