Different approaches have been adopted in addressing the challenges of Artificial Intelligence (AI), some centred on personal data and others on ethics, respectively narrowing and broadening the scope of AI regulation. This contribution aims to demonstrate that a third way is possible, starting from the acknowledgement of the role that human rights can play in regulating the impact of data-intensive systems. The focus on human rights is neither a paradigm shift nor a mere theoretical exercise. Through the analysis of more than 700 decisions and documents of the data protection authorities of six countries, we show that human rights already underpin the decisions in the field of data use. Based on empirical analysis of this evidence, this work presents a methodology and a model for a Human Rights Impact Assessment (HRIA). The methodology and related assessment model are focused on AI applications, whose nature and scale require a proper contextualisation of HRIA methodology. Moreover, the proposed models provide a more measurable approach to risk assessment which is consistent with the regulatory proposals centred on risk thresholds. The proposed methodology is tested in concrete case-studies to prove its feasibility and effectiveness. The overall goal is to respond to the growing interest in HRIA, moving from a mere theoretical debate to a concrete and context-specific implementation in the field of data-intensive applications based on AI.

Pre-trained LMs have had a significant impact on Natural Language Processing (NLP), with the "pre-train and fine-tune" paradigm rapidly becoming the predominant approach to apply effective models on a wide variety of downstream tasks [1-3, inter alia]. However, one of the main concerns when working with LMs is the paucity of annotated data, especially for specific domains or low-resource languages, required to fine-tune the additional classification layer on top of these models for downstream tasks, such as classification. Recently, prompt-based tuning has started to affirm as a promising way to perform similar tasks, significantly reducing the need for annotated data. This approach has been proven to be very effective with Large Language Models (LLMs) [4]. However, it is often the case that LLMs are not available for low-resource languages, and that their performance drastically decreases when they are challenged on specific domains. Moreover, in the Digital Transformation era, businesses frequently need to integrate artificial intelligence systems into their application ecosystems. This requires them to utilize specialized, publicly available models while also employing effective methods to leverage these models in scenarios where annotated language resources are unavailable, thereby operating in a zero-shot mode. Hence, we decided to evaluate two smaller domain-specific encoder models: BureauBERTo [5], a LM further pre-trained on Italian bureaucratic texts (i.e., administrative acts, banking and insurance documents), and Italian Legal BERT [6] (henceforth referred to as Ita-Legal-BERT), a LM adapted to the Italian legal domain, on various classification tasks on domain-specific data exploiting a prompt-based technique in a zero-shot scenario. Additionally, we compared the performance of both models with that of a generic Italian model, UmBERTo.

Generative AI technologies have been deployed in many places, such as (multimodal) large language models and vision generative models. Their remarkable performance should be attributed to massive training data and emergent reasoning abilities. However, the models would memorize and generate sensitive, biased, or dangerous information originated from the training data especially those from web crawl. New machine unlearning (MU) techniques are being developed to reduce or eliminate undesirable knowledge and its effects from the models, because those that were designed for traditional classification tasks could not be applied for Generative AI. We offer a comprehensive survey on many things about MU in Generative AI, such as a new problem formulation, evaluation methods, and a structured discussion on the advantages and limitations of different kinds of MU techniques. It also presents several critical challenges and promising directions in MU research. A curated list of readings can be found: https://github.com/franciscoliu/GenAI-MU-Reading.

The risk of language models unintentionally reproducing copyrighted material from their training data has led to the development of various protective measures. In this paper, we propose model fusion as an effective solution to safeguard against copyright infringement. In particular, we introduce Copyright-Protecting Fusion (CP-Fuse), an algorithm that adaptively combines language models to minimize the reproduction of protected materials. CP-Fuse is inspired by the recently proposed Near-Access Free (NAF) framework and additionally incorporates a desirable balancing property that we demonstrate prevents the reproduction of memorized training data. Our results show that CP-Fuse significantly reduces the memorization of copyrighted content while maintaining high-quality text and code generation. Furthermore, we demonstrate how CP-Fuse can be integrated with other techniques for enhanced protection.

The sharing economy is sprawling across almost every sector and activity around the world. About a decade ago, there were only a handful of platform driven companies operating on the market. Zipcar, BlaBlaCar and Couchsurfing among them. Then Airbnb and Uber revolutionized the transportation and hospitality industries with a presence in virtually every major city. Access over ownership is the paradigm shift from the traditional business model that grants individuals the use of products or services without the necessity of buying them. Digital platforms, data and algorithm-driven companies as well as decentralized blockchain technologies have tremendous potential. But they are also changing the rules of the game. One of such technologies challenging the legal system are AI systems that will also reshape the current legal framework concerning the liability of operators, users and manufacturers. Therefore, this introductory chapter deals with explaining and describing the legal issues of some of these disruptive technologies. The chapter argues for a more forward-thinking and flexible regulatory structure.

This chapter examines the challenges of the revised opt out system and the secondary use of health data in England. The analysis of this data could be very valuable for science and medical treatment as well as for the discovery of new drugs. For this reason, the UK government established the care.data program in 2013. The aim of the project was to build a central nationwide database for research and policy planning. However, the processing of personal data was planned without proper public engagement. Research has suggested that IT companies, such as in the Google DeepMind deal case, had access to other kinds of sensitive data and failed to comply with data protection law. Since May 2018, the government has launched the national data opt out system with the hope of regaining public trust. Nevertheless, there are no evidence of significant changes in the ND opt out, compared to the previous opt out system. Neither in the use of secondary data, nor in the choices that patients can make. The only notorious difference seems to be in the way that these options are communicated and framed to the patients. Most importantly, according to the new ND opt out, the type 1 opt out option, which is the only choice that truly stops data from being shared outside direct care, will be removed in 2020. According to the Behavioral Law and Economics literature (Nudge Theory), default rules, such as the revised opt out system in England, are very powerful, because people tend to stick to the default choices made readily available to them. The crucial question analyzed in this chapter is whether it is desirable for the UK government to stop promoting the type 1 opt outs, and whether this could be seen as a kind of hard paternalism.

Identifying and then implementing an effective response to disruptive new AI technologies is enormously challenging for any business looking to integrate AI into their operations, as well as regulators looking to leverage AI-related innovation as a mechanism for achieving regional economic growth. These business and regulatory challenges are particularly significant given the broad reach of AI, as well as the multiple uncertainties surrounding such technologies and their future development and effects. This article identifies two promising strategies for meeting the AI challenge, focusing on the example of Fintech. First, dynamic regulation, in the form of regulatory sandboxes and other regulatory approaches that aim to provide a space for responsible AI-related innovation. An empirical study provides preliminary evidence to suggest that jurisdictions that adopt a more proactive approach to Fintech regulation can attract greater investment. The second strategy relates to so-called innovation ecosystems. It is argued that such ecosystems are most effective when they afford opportunities for creative partnerships between well-established corporations and AI-focused startups and that this aspect of a successful innovation ecosystem is often overlooked in the existing discussion. The article suggests that these two strategies are interconnected, in that greater investment is an important element in both fostering and signaling a well-functioning innovation ecosystem and that a well-functioning ecosystem will, in turn, attract more funding. The resulting synergies between these strategies can, therefore, provide a jurisdiction with a competitive edge in becoming a regional hub for AI-related activity.

In this paper, we introduce SaulLM-54B and SaulLM-141B, two large language models (LLMs) tailored for the legal sector. These models, which feature architectures of 54 billion and 141 billion parameters, respectively, are based on the Mixtral architecture. The development of SaulLM-54B and SaulLM-141B is guided by large-scale domain adaptation, divided into three strategies: (1) the exploitation of continued pretraining involving a base corpus that includes over 540 billion of legal tokens, (2) the implementation of a specialized legal instruction-following protocol, and (3) the alignment of model outputs with human preferences in legal interpretations. The integration of synthetically generated data in the second and third steps enhances the models' capabilities in interpreting and processing legal texts, effectively reaching state-of-the-art performance and outperforming previous open-source models on LegalBench-Instruct. This work explores the trade-offs involved in domain-specific adaptation at this scale, offering insights that may inform future studies on domain adaptation using strong decoder models. Building upon SaulLM-7B, this study refines the approach to produce an LLM better equipped for legal tasks. We are releasing base, instruct, and aligned versions on top of SaulLM-54B and SaulLM-141B under the MIT License to facilitate reuse and collaborative research.

Persona agents, which are LLM agents that act according to an assigned persona, have demonstrated impressive contextual response capabilities across various applications. These persona agents offer significant enhancements across diverse sectors, such as education, healthcare, and entertainment, where model developers can align agent responses to different user requirements thereby broadening the scope of agent applications. However, evaluating persona agent performance is incredibly challenging due to the complexity of assessing persona adherence in free-form interactions across various environments that are relevant to each persona agent. We introduce PersonaGym, the first dynamic evaluation framework for assessing persona agents, and PersonaScore, the first automated human-aligned metric grounded in decision theory for comprehensive large-scale evaluation of persona agents. Our evaluation of 6 open and closed-source LLMs, using a benchmark encompassing 200 personas and 10,000 questions, reveals significant opportunities for advancement in persona agent capabilities across state-of-the-art models. For example, Claude 3.5 Sonnet only has a 2.97% relative improvement in PersonaScore than GPT 3.5 despite being a much more advanced model. Importantly, we find that increased model size and complexity do not necessarily imply enhanced persona agent capabilities thereby highlighting the pressing need for algorithmic and architectural invention towards faithful and performant persona agents.

Deep anomaly detection (AD) is perhaps the most controversial of data analytic tasks as it identifies entities that are then specifically targeted for further investigation or exclusion. Also controversial is the application of AI to facial imaging data. This work explores the intersection of these two areas to understand two core questions: "Who" these algorithms are being unfair to and equally important "Why". Recent work has shown that deep AD can be unfair to different groups despite being unsupervised with a recent study showing that for portraits of people: men of color are far more likely to be chosen to be outliers. We study the two main categories of AD algorithms: autoencoder-based and single-class-based which effectively try to compress all the instances with those that can not be easily compressed being deemed to be outliers. We experimentally verify sources of unfairness such as the under-representation of a group (e.g. people of color are relatively rare), spurious group features (e.g. men are often photographed with hats), and group labeling noise (e.g. race is subjective). We conjecture that lack of compressibility is the main foundation and the others cause it but experimental results show otherwise and we present a natural hierarchy amongst them.