As machine learning technology gets applied to actual products and solutions, new challenges have emerged. Models unexpectedly fail to generalize to small changes in the distribution, tend to be confident on novel data they have never seen, or cannot communicate the rationale behind their decisions effectively with the end users. Collectively, we face a trustworthiness issue with the current machine learning technology. This textbook on Trustworthy Machine Learning (TML) covers a theoretical and technical background of four key topics in TML: Out-of-Distribution Generalization, Explainability, Uncertainty Quantification, and Evaluation of Trustworthiness. We discuss important classical and contemporary research papers of the aforementioned fields and uncover and connect their underlying intuitions. The book evolved from the homonymous course at the University of T\"ubingen, first offered in the Winter Semester of 2022/23. It is meant to be a stand-alone product accompanied by code snippets and various pointers to further sources on topics of TML. The dedicated website of the book is https://trustworthyml.io/.

Large language models (LLMs) have garnered significant attention for their remarkable performance in a continuously expanding set of natural language processing tasks. However, these models have been shown to harbor inherent societal biases, or stereotypes, which can adversely affect their performance in their many downstream applications. In this paper, we introduce a novel, purely prompt-based approach to uncover hidden stereotypes within any arbitrary LLM. Our approach dynamically generates a knowledge representation of internal stereotypes, enabling the identification of biases encoded within the LLM's internal knowledge. By illuminating the biases present in LLMs and offering a systematic methodology for their analysis, our work contributes to advancing transparency and promoting fairness in natural language processing systems.

Research in XAI aims to improve fairness in human-AI decision-making by providing insights into model predictions, and thereby allowing humans to understand and correct for model biases. On the other hand, in the context of human-AI decision-making, previous work has noted that humans often over-rely on AI predictions, and explanations can exacerbate this concern [9]. This is especially troubling if the underlying model contains systematic biases, which may go unnoticed even when teamed with a human. In order for the human-AI team to be successful, the human needs to be able to determine when to rely on or override potentially biased AI predictions. Previous work has shown that explanations can help human-AI teams alleviate model biases when those biases depend directly on protected attributes [18, 54], but little is known in the very common case that protected attributes are not explicitly included, and rather the features used for prediction contain proxies thereof (e.g., zip code for race, length of credit for age, and university attended for gender). In particular, it may be difficult for humans to identify and resolve biased model predictions based on the proxy features present in real-world data, even when explanations are provided. In this work, we study whether explanations can help people to identify model biases and to calibrate their reliance on an AI model based on these biases. We extend this line of investigation beyond direct biases that are revealed through the use of protected (i.e., sensitive) features by considering the effect of explanations when indirect bias is revealed Both co-first authors contributed equally to this manuscript, and each has the right to list their name first on their CV.

We propose a belief-formation model where agents attempt to discriminate between two theories, and where the asymmetry in strength between confirming and disconfirming evidence tilts beliefs in favor of theories that generate strong (and possibly rare) confirming evidence and weak (and frequent) disconfirming evidence. In our model, limitations on information processing provide incentives to censor weak evidence, with the consequence that for some discrimination problems, evidence may become mostly one-sided, independently of the true underlying theory. Sophisticated agents who know the characteristics of the censored data-generating process are not lured by this accumulation of ``evidence'', but less sophisticated ones end up with biased beliefs.

Probabilities of Causation (PoC) play a fundamental role in decision-making in law, health care and public policy. Nevertheless, their point identification is challenging, requiring strong assumptions, in the absence of which only bounds can be derived. Existing work to further tighten these bounds by leveraging extra information either provides numerical bounds, symbolic bounds for fixed dimensionality, or requires access to multiple datasets that contain the same treatment and outcome variables. However, in many clinical, epidemiological and public policy applications, there exist external datasets that examine the effect of different treatments on the same outcome variable, or study the association between covariates and the outcome variable. These external datasets cannot be used in conjunction with the aforementioned bounds, since the former may entail different treatment assignment mechanisms, or even obey different causal structures. Here, we provide symbolic bounds on the PoC for this challenging scenario. We focus on combining either two randomized experiments studying different treatments, or a randomized experiment and an observational study, assuming causal sufficiency. Our symbolic bounds work for arbitrary dimensionality of covariates and treatment, and we discuss the conditions under which these bounds are tighter than existing bounds in literature. Finally, our bounds parameterize the difference in treatment assignment mechanism across datasets, allowing the mechanisms to vary across datasets while still allowing causal information to be transferred from the external dataset to the target dataset.

As Artificial Intelligence (AI) becomes pervasive in most fields, from healthcare to autonomous driving, it is essential that we find successful ways of building morality into our machines, especially for decision-making. However, the question of what it means to be moral is still debated, particularly in the context of AI. In this paper, we highlight the different aspects that should be considered when building moral agents, including the most relevant moral paradigms and challenges. We also discuss the top-down and bottom-up approaches to design and the role of emotion and sentience in morality. We then propose solutions including a hybrid approach to design and a hierarchical approach to combining moral paradigms. We emphasize how governance and policy are becoming ever more critical in AI Ethics and in ensuring that the tasks we set for moral agents are attainable, that ethical behavior is achieved, and that we obtain good AI.

Machine unlearning, the study of efficiently removing the impact of specific training points on the trained model, has garnered increased attention of late, driven by the need to comply with privacy regulations like the Right to be Forgotten. Although unlearning is particularly relevant for LLMs in light of the copyright issues they raise, achieving precise unlearning is computationally infeasible for very large models. To this end, recent work has proposed several algorithms which approximate the removal of training data without retraining the model. These algorithms crucially rely on access to the model parameters in order to update them, an assumption that may not hold in practice due to computational constraints or when the LLM is accessed via API. In this work, we propose a new class of unlearning methods for LLMs we call ''In-Context Unlearning'', providing inputs in context and without having to update model parameters. To unlearn a particular training instance, we provide the instance alongside a flipped label and additional correctly labelled instances which are prepended as inputs to the LLM at inference time. Our experimental results demonstrate that these contexts effectively remove specific information from the training set while maintaining performance levels that are competitive with (or in some cases exceed) state-of-the-art unlearning methods that require access to the LLM parameters.

The count of pending cases has shown an exponential rise across nations (e.g., with more than 10 million pending cases in India alone). The main issue lies in the fact that the number of cases submitted to the law system is far greater than the available number of legal professionals present in a country. Given this worldwide context, the utilization of AI technology has gained paramount importance to enhance the efficiency and speed of legal procedures. In this study we partcularly focus on helping legal professionals in the process of analyzing a legal case. Our specific investigation delves into harnessing the generative capabilities of open-sourced large language models to create arguments derived from the facts present in legal cases. Experimental results show that the generated arguments from the best performing method have on average 63% overlap with the benchmark set gold standard annotations.

While large language models (LLMs) like GPT-4 have recently demonstrated astonishing zero-shot capabilities in general domain tasks, they often generate content with hallucinations in specific domains such as Chinese law, hindering their application in these areas. This is typically due to the absence of training data that encompasses such a specific domain, preventing GPT-4 from acquiring in-domain knowledge. A pressing challenge is that it's not plausible to continue training LLMs of such scale on in-domain data. This paper introduces a simple and effective domain adaptation framework for GPT-4 by reformulating generation as an \textbf{adapt-retrieve-revise} process. The initial step is to \textbf{adapt} an affordable 7B LLM to the target domain by continuing learning on in-domain data. When solving a task, we leverage the adapted LLM to generate a draft answer given a task query. Then, the draft answer will be used to \textbf{retrieve} supporting evidence candidates from an external in-domain knowledge base. Finally, the draft answer and retrieved evidence are concatenated into a whole prompt to let GPT-4 assess the evidence and \textbf{revise} the draft answer to generate the final answer. Our proposal combines the advantages of the efficiency of adapting a smaller 7B model with the evidence-assessing capability of GPT-4 and effectively prevents GPT-4 from generating hallucinatory content. In the zero-shot setting of four Chinese legal tasks, our method improves accuracy by 33.3\% compared to the direct generation by GPT-4. When compared to two stronger retrieval-based baselines, our method outperforms them by 15.4\% and 23.9\%. Our code will be released

Generative artificial intelligence (AI) systems, particularly large language models (LLMs), have improved at a rapid pace. For example, ChatGPT recently showcased its advanced capacity to perform complex tasks and human-like behaviors (OpenAI 2023b), reaching 100 million users within two months of its 2022 launch (Hu 2023). This progress is not limited to text generation, as demonstrated by other recent generative AI systems such as Midjourney (Midjourney 2023) (a text-to-image generative AI) and GitHub Copilot (Github 2023) (an AI pair programmer that can autocomplete code). Eloundou et al. (2023) estimated that about 80% of the U.S. workforce could be affected by the introduction of LLMs, and 19% of the workers may have at least 50% of their tasks impacted. In particular, AI can make users more productive by generating complex content in seconds, while users can simply communicate their preferences. For example, Noy and Zhang (2023) highlighted that ChatGPT can substantially improve productivity in writing tasks, and GitHub claims that Copilot increases developer productivity by up to 55% (Kalliamvakou 2023). However, content generated with the help of AI is not exactly the same as content generated without AI. The boost in productivity may come at the expense of users' idiosyncrasies, such as personal style and tastes, preferences we would naturally express without AI. To let users express their preferences, many AI systems let users edit their prompt (e.g., Midjourney) or allow more