A former interior minister, Tudorache is one of the most important players in European AI policy. He is one of the two lead negotiators of the AI Act in the European Parliament. The bill, the first sweeping AI law of its kind in the world, will enter into force this year. We first met two years ago, when Tudorache was appointed to his position as negotiator. But Tudorache's interest in AI started much earlier, in 2015.

Large Language Models (LLMs) have demonstrated remarkable performance across diverse tasks but are constrained by their small context window sizes. Various efforts have been proposed to expand the context window to accommodate even up to 200K input tokens. Meanwhile, building high-quality benchmarks with much longer text lengths and more demanding tasks to provide comprehensive evaluations is of immense practical interest to facilitate long context understanding research of LLMs. However, prior benchmarks create datasets that ostensibly cater to long-text comprehension by expanding the input of traditional tasks, which falls short to exhibit the unique characteristics of long-text understanding, including long dependency tasks and longer text length compatible with modern LLMs' context window size. In this paper, we introduce a benchmark for extremely long context understanding with long-range dependencies, XL$^2$Bench, which includes three scenarios: Fiction Reading, Paper Reading, and Law Reading, and four tasks of increasing complexity: Memory Retrieval, Detailed Understanding, Overall Understanding, and Open-ended Generation, covering 27 subtasks in English and Chinese. It has an average length of 100K+ words (English) and 200K+ characters (Chinese). Evaluating six leading LLMs on XL$^2$Bench, we find that their performance significantly lags behind human levels. Moreover, the observed decline in performance across both the original and enhanced datasets underscores the efficacy of our approach to mitigating data contamination.

As Large Language Models (LLMs) and generative AI become more widespread, the content safety risks associated with their use also increase. We find a notable deficiency in high-quality content safety datasets and benchmarks that comprehensively cover a wide range of critical safety areas. To address this, we define a broad content safety risk taxonomy, comprising 13 critical risk and 9 sparse risk categories. Additionally, we curate AEGISSAFETYDATASET, a new dataset of approximately 26, 000 human-LLM interaction instances, complete with human annotations adhering to the taxonomy. We plan to release this dataset to the community to further research and to help benchmark LLM models for safety. To demonstrate the effectiveness of the dataset, we instruction-tune multiple LLM-based safety models. We show that our models (named AEGISSAFETYEXPERTS), not only surpass or perform competitively with the state-of-the-art LLM-based safety models and general purpose LLMs, but also exhibit robustness across multiple jail-break attack categories. We also show how using AEGISSAFETYDATASET during the LLM alignment phase does not negatively impact the performance of the aligned models on MT Bench scores. Furthermore, we propose AEGIS, a novel application of a no-regret online adaptation framework with strong theoretical guarantees, to perform content moderation with an ensemble of LLM content safety experts in deployment

This paper investigates the influence of anthropomorphized descriptions of so-called "AI" (artificial intelligence) systems on people's self-assessment of trust in the system. Building on prior work, we define four categories of anthropomorphization (1. Properties of a cognizer, 2. Agency, 3. Biological metaphors, and 4. Properties of a communicator). We use a survey-based approach (n=954) to investigate whether participants are likely to trust one of two (fictitious) "AI" systems by randomly assigning people to see either an anthropomorphized or a de-anthropomorphized description of the systems. We find that participants are no more likely to trust anthropomorphized over de-anthropmorphized product descriptions overall. The type of product or system in combination with different anthropomorphic categories appears to exert greater influence on trust than anthropomorphizing language alone, and age is the only demographic factor that significantly correlates with people's preference for anthropomorphized or de-anthropomorphized descriptions. When elaborating on their choices, participants highlight factors such as lesser of two evils, lower or higher stakes contexts, and human favoritism as driving motivations when choosing between product A and B, irrespective of whether they saw an anthropomorphized or a de-anthropomorphized description of the product. Our results suggest that "anthropomorphism" in "AI" descriptions is an aggregate concept that may influence different groups differently, and provide nuance to the discussion of whether anthropomorphization leads to higher trust and over-reliance by the general public in systems sold as "AI".

Data are the critical fuel for Artificial Intelligence (AI) models. Poor quality data produces inaccurate and ineffective AI models that may lead to incorrect or unsafe use. Checking for data readiness is a crucial step in improving data quality. Numerous R&D efforts have been spent on improving data quality. However, standardized metrics for evaluating data readiness for use in AI training are still evolving. In this study, we perform a comprehensive survey of metrics used for verifying AI's data readiness. This survey examines more than 120 papers that are published by ACM Digital Library, IEEE Xplore, other reputable journals, and articles published on the web by prominent AI experts. This survey aims to propose a taxonomy of data readiness for AI (DRAI) metrics for structured and unstructured datasets. We anticipate that this taxonomy can lead to new standards for DRAI metrics that would be used for enhancing the quality and accuracy of AI training and inference.

Numerous decision-making tasks require estimating causal effects under interventions on different parts of a system. As practitioners consider using large language models (LLMs) to automate decisions, studying their causal reasoning capabilities becomes crucial. A recent line of work evaluates LLMs ability to retrieve commonsense causal facts, but these evaluations do not sufficiently assess how LLMs reason about interventions. Motivated by the role that interventions play in causal inference, in this paper, we conduct empirical analyses to evaluate whether LLMs can accurately update their knowledge of a data-generating process in response to an intervention. We create benchmarks that span diverse causal graphs (e.g., confounding, mediation) and variable types, and enable a study of intervention-based reasoning. These benchmarks allow us to isolate the ability of LLMs to accurately predict changes resulting from their ability to memorize facts or find other shortcuts. Our analysis on four LLMs highlights that while GPT- 4 models show promising accuracy at predicting the intervention effects, they remain sensitive to distracting factors in the prompts.

A system safety case is a compelling, comprehensible, and valid argument about the satisfaction of the safety goals of a given system operating in a given environment supported by convincing evidence. Since the publication of UL 4600 in 2020, safety cases have become a best practice for measuring, managing, and communicating the safety of autonomous vehicles (AVs). Although UL 4600 provides guidance on how to build the safety case for an AV, the complexity of AVs and their operating environments, the novelty of the used technology, the need for complying with various regulations and technical standards, and for addressing cybersecurity concerns and ethical considerations make the development of safety cases for AVs challenging. To this end, safety case frameworks have been proposed that bring strategies, argument templates, and other guidance together to support the development of a safety case. This paper introduces the Open Autonomy Safety Case Framework, developed over years of work with the autonomous vehicle industry, as a roadmap for how AVs can be deployed safely and responsibly.

Entity resolution (record linkage, microclustering) systems are notoriously difficult to evaluate. Looking for a needle in a haystack, traditional evaluation methods use sophisticated, application-specific sampling schemes to find matching pairs of records among an immense number of non-matches. We propose an alternative that facilitates the creation of representative, reusable benchmark data sets without necessitating complex sampling schemes. These benchmark data sets can then be used for model training and a variety of evaluation tasks. Specifically, we propose an entity-centric data labeling methodology that integrates with a unified framework for monitoring summary statistics, estimating key performance metrics such as cluster and pairwise precision and recall, and analyzing root causes for errors. We validate the framework in an application to inventor name disambiguation and through simulation studies. Software: https://github.com/OlivierBinette/er-evaluation/

Accurate and reliable air quality forecasting is essential for protecting public health, sustainable development, pollution control, and enhanced urban planning. This letter presents a novel WaveCatBoost architecture designed to forecast the real-time concentrations of air pollutants by combining the maximal overlapping discrete wavelet transform (MODWT) with the CatBoost model. This hybrid approach efficiently transforms time series into high-frequency and low-frequency components, thereby extracting signal from noise and improving prediction accuracy and robustness. Evaluation of two distinct regional datasets, from the Central Air Pollution Control Board (CPCB) sensor network and a low-cost air quality sensor system (LAQS), underscores the superior performance of our proposed methodology in real-time forecasting compared to the state-of-the-art statistical and deep learning architectures. Moreover, we employ a conformal prediction strategy to provide probabilistic bands with our forecasts.

Forthcoming in Collaborative Intelligence: How Humans and AI are Transforming our World, Arathi Sethumadhavan and Mira Lane (eds.), Seth Lazar, Australian National University Man, a child in understanding of himself, has placed in his hands physical tools of incalculable power. He plays with them like a child, and whether they work harm or good is largely a matter of accident. The instrumentality becomes a master and works fatally as if possessed of a will of its own-- not because it has a will but because man has not. Introduction As rapid advances in Artificial Intelligence and the rise of some of history's most potent corporations meet the diminished neoliberal state, people are increasingly subject to power exercised by means of automated systems. Machine learning, big data, and related computational technologies now underpin vital government services from criminal justice to tax auditing, public health to social services, immigration to defence (Citron, 2008; Calo and Citron, 2020; Engstrom et al., 2020). Google and Amazon connect consumers and producers in new algorithmic markets (Nadler and Cicilline, 2020). Google's search algorithm--and possibly in the near future OpenAI's GPT-4 or another large language model--determines, for many, how they find out about everything from how to vote to where to get vaccinated. Meta, Twitter, TikTok, Google and others algorithmically decide whose speech is amplified, reduced, or restricted (Vaidhyanathan, 2011; Pasquale, 2015; Gillespie, 2018; Suzor, 2019). And a new wave of products based on rapid advances in Large Language Models (LLMs) have the potential to further transform our economic and political lives. Automatic Authorities are automated computational systems used to exercise power over us by substantially determining what we may know, what we may have, and what our options will be. This chapter is based on, and substantially revises, my'Power and AI: Nature and Justification', in the Oxford Handbook of AI Governance (Justin Bullock et al., eds). My thanks to the publisher for their permission to use this material. But what normative lessons should we draw from these analyses? Power is everywhere, and is not necessarily bad.