Beneficial societal outcomes cannot be guaranteed by aligning individual AI systems with the intentions of their operators or users. Even an AI system that is perfectly aligned to the intentions of its operating organization can lead to bad outcomes if the goals of that organization are misaligned with those of other institutions and individuals. For this reason, we need full-stack alignment, the concurrent alignment of AI systems and the institutions that shape them with what people value. This can be done without imposing a particular vision of individual or collective flourishing. We argue that current approaches for representing values, such as utility functions, preference orderings, or unstructured text, struggle to address these and other issues effectively. They struggle to distinguish values from other signals, to support principled normative reasoning, and to model collective goods. We propose thick models of value will be needed. These structure the way values and norms are represented, enabling systems to distinguish enduring values from fleeting preferences, to model the social embedding of individual choices, and to reason normatively, applying values in new domains. We demonstrate this approach in five areas: AI value stewardship, normatively competent agents, win-win negotiation systems, meaning-preserving economic mechanisms, and democratic regulatory institutions.

Normative reasoning is a type of reasoning that involves normative or deontic modality, such as obligation and permission. While large language models (LLMs) have demonstrated remarkable performance across various reasoning tasks, their ability to handle normative reasoning remains underexplored. In this paper, we systematically evaluate LLMs' reasoning capabilities in the normative domain from both logical and modal perspectives. Specifically, to assess how well LLMs reason with normative modals, we make a comparison between their reasoning with normative modals and their reasoning with epistemic modals, which share a common formal structure. To this end, we introduce a new dataset covering a wide range of formal patterns of reasoning in both normative and epistemic domains, while also incorporating non-formal cognitive factors that influence human reasoning. Our results indicate that, although LLMs generally adhere to valid reasoning patterns, they exhibit notable inconsistencies in specific types of normative reasoning and display cognitive biases similar to those observed in psychological studies of human reasoning. These findings highlight challenges in achieving logical consistency in LLMs' normative reasoning and provide insights for enhancing their reliability. All data and code are released publicly at https://github.com/kmineshima/NeuBAROCO.

Reinforcement Learning is a machine learning methodology that has demonstrated strong performance across a variety of tasks. In particular, it plays a central role in the development of artificial autonomous agents. As these agents become increasingly capable, market readiness is rapidly approaching, which means those agents, for example taking the form of humanoid robots or autonomous cars, are poised to transition from laboratory prototypes to autonomous operation in real-world environments. This transition raises concerns leading to specific requirements for these systems - among them, the requirement that they are designed to behave ethically. Crucially, research directed toward building agents that fulfill the requirement to behave ethically - referred to as artificial moral agents(AMAs) - has to address a range of challenges at the intersection of computer science and philosophy. This study explores the development of reason-based artificial moral agents (RBAMAs). RBAMAs are build on an extension of the reinforcement learning architecture to enable moral decision-making based on sound normative reasoning, which is achieved by equipping the agent with the capacity to learn a reason-theory - a theory which enables it to process morally relevant propositions to derive moral obligations - through case-based feedback. They are designed such that they adapt their behavior to ensure conformance to these obligations while they pursue their designated tasks. These features contribute to the moral justifiability of the their actions, their moral robustness, and their moral trustworthiness, which proposes the extended architecture as a concrete and deployable framework for the development of AMAs that fulfills key ethical desiderata. This study presents a first implementation of an RBAMA and demonstrates the potential of RBAMAs in initial experiments.

Compliance with the GDPR privacy regulation places a significant burden on organisations regarding the handling of personal data. The perceived efforts and risks of complying with the GDPR further increase when data processing activities span across organisational boundaries, as is the case in both small-scale data sharing settings and in large-scale international data spaces. This paper addresses these concerns by proposing a case-generic method for automated normative reasoning that establishes legal arguments for the lawfulness of data processing activities. The arguments are established on the basis of case-specific legal qualifications made by privacy experts, bringing the human in the loop. The obtained expert system promotes transparency and accountability, remains adaptable to extended or altered interpretations of the GDPR, and integrates into novel or existing distributed data processing systems. This result is achieved by defining a formal ontology and semantics for automated normative reasoning based on an analysis of the purpose-limitation principle of the GDPR. The ontology and semantics are implemented in eFLINT, a domain-specific language for specifying and reasoning with norms. The XACML architecture standard, applicable to both access and usage control, is extended, demonstrating how GDPR-based normative reasoning can integrate into (existing, distributed) systems for data processing. The resulting system is designed and critically assessed in reference to requirements extracted from the GPDR.

Human activity is moderated by norms. However, machines are often trained without explicit supervision on norm understanding and reasoning, especially when the norms are grounded in a physical and social context. To improve and evaluate the normative reasoning capability of vision-language models (VLMs), we present EgoNormia $\|\epsilon\|$, consisting of 1,853 ego-centric videos of human interactions, each of which has two related questions evaluating both the prediction and justification of normative actions. The normative actions encompass seven categories: safety, privacy, proxemics, politeness, cooperation, coordination/proactivity, and communication/legibility. To compile this dataset at scale, we propose a novel pipeline leveraging video sampling, automatic answer generation, filtering, and human validation. Our work demonstrates that current state-of-the-art vision-language models lack robust norm understanding, scoring a maximum of 45% on EgoNormia (versus a human bench of 92%). Our analysis of performance in each dimension highlights the significant risks of safety, privacy, and the lack of collaboration and communication capability when applied to real-world agents. We additionally show that through a retrieval-based generation method, it is possible to use EgoNormia to enhance normative reasoning in VLMs.

Software agents, both human and computational, do not exist in isolation and often need to collaborate or coordinate with others to achieve their goals. In human society, social mechanisms such as norms ensure efficient functioning, and these techniques have been adopted by researchers in multi-agent systems (MAS) to create socially aware agents. However, traditional techniques have limitations, such as operating in limited environments often using brittle symbolic reasoning. The advent of Large Language Models (LLMs) offers a promising solution, providing a rich and expressive vocabulary for norms and enabling norm-capable agents that can perform a range of tasks such as norm discovery, normative reasoning and decision-making. This paper examines the potential of LLM-based agents to acquire normative capabilities, drawing on recent Natural Language Processing (NLP) and LLM research. We present our vision for creating normative LLM agents. In particular, we discuss how the recently proposed "LLM agent" approaches can be extended to implement such normative LLM agents. We also highlight challenges in this emerging field. This paper thus aims to foster collaboration between MAS, NLP and LLM researchers in order to advance the field of normative agents.

Norms are an important component of the social fabric of society by prescribing expected behaviour. In Multi-Agent Systems (MAS), agents interacting within a society are equipped to possess social capabilities such as reasoning about norms and trust. Norms have long been of interest within the Normative Multi-Agent Systems community with researchers studying topics such as norm emergence, norm violation detection and sanctioning. However, these studies have some limitations: they are often limited to simple domains, norms have been represented using a variety of representations with no standard approach emerging, and the symbolic reasoning mechanisms generally used may suffer from a lack of extensibility and robustness. In contrast, Large Language Models (LLMs) offer opportunities to discover and reason about norms across a large range of social situations. This paper evaluates the capability of LLMs to detecting norm violations. Based on simulated data from 80 stories in a household context, with varying complexities, we investigated whether 10 norms are violated. For our evaluations we first obtained the ground truth from three human evaluators for each story. Then, the majority result was compared against the results from three well-known LLM models (Llama 2 7B, Mixtral 7B and ChatGPT-4). Our results show the promise of ChatGPT-4 for detecting norm violations, with Mixtral some distance behind. Also, we identify areas where these models perform poorly and discuss implications for future work.

LegalRuleML is a comprehensive XML-based representation framework for modeling and exchanging normative rules. The TPTP input and output formats, on the other hand, are general-purpose standards for the interaction with automated reasoning systems. In this paper we provide a bridge between the two communities by (i) defining a logic-pluralistic normative reasoning language based on the TPTP format, (ii) providing a translation scheme between relevant fragments of LegalRuleML and this language, and (iii) proposing a flexible architecture for automated normative reasoning based on this translation. We exemplarily instantiate and demonstrate the approach with three different normative logics.

In this paper we introduce an abstract theory of normative reasoning, whose central notion is the generation of obligations, permissions and institutional facts from conditional norms.

Several solutions for specifying normative artefacts (norms, contracts, policies) in a computational processable way have been presented in the literature. Legal core ontologies have been proposed to systematize concepts and relationships relevant to normative reasoning. However, no solution amongst those has achieved general acceptance, and no common ground (representational, computational) has been identified enabling us to easily compare them. Yet, all these efforts share the same motivation of representing normative directives, therefore it is plausible that there may be a representational model encompassing all of them. This presentation will introduce DPCL, a domain-specific language (DSL) for specifying higher-level policies (including norms, contracts, etc.), centred on Hohfeld's framework of fundamental legal concepts. DPCL has to be seen primarily as a "template", i.e. as an informational model for architectural reference, rather than a fully-fledged formal language; it aims to make explicit the general requirements that should be expected in a language for norm specification. In this respect, it goes rather in the direction of legal core ontologies, but differently from those, our proposal aims to keep the character of a DSL, rather than a set of axioms in a logical framework: it is meant to be cross-compiled to underlying languages/tools adequate to the type of target application. We provide here an overview of some of the language features.