This paper proposes a Multinational Artificial General Intelligence Consortium (MAGIC) to mitigate existential risks from advanced artificial intelligence (AI). MAGIC would be the only institution in the world permitted to develop advanced AI, enforced through a global moratorium by its signatory members on all other advanced AI development. MAGIC would be exclusive, safety-focused, highly secure, and collectively supported by member states, with benefits distributed equitably among signatories. MAGIC would allow narrow AI models to flourish while significantly reducing the possibility of misaligned, rogue, breakout, or runaway outcomes of general-purpose systems. We do not address the political feasibility of implementing a moratorium or address the specific legislative strategies and rules needed to enforce a ban on high-capacity AGI training runs. Instead, we propose one positive vision of the future, where MAGIC, as a global governance regime, can lay the groundwork for long-term, safe regulation of advanced AI.

Quantification, i.e., the task of training predictors of the class prevalence values in sets of unlabeled data items, has received increased attention in recent years. However, most quantification research has concentrated on developing algorithms for binary and multiclass problems in which the classes are not ordered. Here, we study the ordinal case, i.e., the case in which a total order is defined on the set of n>2 classes. We give three main contributions to this field. First, we create and make available two datasets for ordinal quantification (OQ) research that overcome the inadequacies of the previously available ones. Second, we experimentally compare the most important OQ algorithms proposed in the literature so far. To this end, we bring together algorithms proposed by authors from very different research fields, such as data mining and astrophysics, who were unaware of each others' developments. Third, we propose a novel class of regularized OQ algorithms, which outperforms existing algorithms in our experiments. The key to this gain in performance is that our regularization prevents ordinally implausible estimates, assuming that ordinal distributions tend to be smooth in practice. We informally verify this assumption for several real-world applications.

Information in industry, research, and the public sector is widely stored as rendered documents (e.g., PDF files, scans). Hence, to enable downstream tasks, systems are needed that map rendered documents onto a structured hierarchical format. However, existing systems for this task are limited by heuristics and are not end-to-end trainable. In this work, we introduce the Document Structure Generator (DSG), a novel system for document parsing that is fully end-to-end trainable. DSG combines a deep neural network for parsing (i) entities in documents (e.g., figures, text blocks, headers, etc.) and (ii) relations that capture the sequence and nested structure between entities. Unlike existing systems that rely on heuristics, our DSG is trained end-to-end, making it effective and flexible for real-world applications. We further contribute a new, large-scale dataset called E-Periodica comprising real-world magazines with complex document structures for evaluation. Our results demonstrate that our DSG outperforms commercial OCR tools and, on top of that, achieves state-of-the-art performance. To the best of our knowledge, our DSG system is the first end-to-end trainable system for hierarchical document parsing.

The cost of delays was estimated as 33 billion US dollars only in 2019 for the US National Airspace System, a peak value following a growth trend in past years. Aiming to address this huge inefficiency, we designed and developed a novel Data Analytics and Machine Learning system, which aims at reducing delays by proactively supporting re-routing decisions. Given a time interval up to a few days in the future, the system predicts if a reroute advisory for a certain Air Route Traffic Control Center or for a certain advisory identifier will be issued, which may impact the pertinent routes. To deliver such predictions, the system uses historical reroute data, collected from the System Wide Information Management (SWIM) data services provided by the FAA, and weather data, provided by the US National Centers for Environmental Prediction (NCEP). The data is huge in volume, and has many items streamed at high velocity, uncorrelated and noisy. The system continuously processes the incoming raw data and makes it available for the next step where an interim data store is created and adaptively maintained for efficient query processing. The resulting data is fed into an array of ML algorithms, which compete for higher accuracy. The best performing algorithm is used in the final prediction, generating the final results. Mean accuracy values higher than 90% were obtained in our experiments with this system. Our algorithm divides the area of interest in units of aggregation and uses temporal series of the aggregate measures of weather forecast parameters in each geographical unit, in order to detect correlations with reroutes and where they will most likely occur. Aiming at practical application, the system is formed by a number of microservices, which are deployed in the cloud, making the system distributed, scalable and highly available.

The growing capabilities and increasingly widespread deployment of AI systems necessitate robust benchmarks for measuring their cooperative capabilities. Unfortunately, most multi-agent benchmarks are either zero-sum or purely cooperative, providing limited opportunities for such measurements. We introduce a general-sum variant of the zero-sum board game Diplomacy -- called Welfare Diplomacy -- in which players must balance investing in military conquest and domestic welfare. We argue that Welfare Diplomacy facilitates both a clearer assessment of and stronger training incentives for cooperative capabilities. Our contributions are: (1) proposing the Welfare Diplomacy rules and implementing them via an open-source Diplomacy engine; (2) constructing baseline agents using zero-shot prompted language models; and (3) conducting experiments where we find that baselines using state-of-the-art models attain high social welfare but are exploitable. Our work aims to promote societal safety by aiding researchers in developing and assessing multi-agent AI systems. Code to evaluate Welfare Diplomacy and reproduce our experiments is available at https://github.com/mukobi/welfare-diplomacy.

Artificial Intelligence (AI) is rapidly integrating into various aspects of our daily lives, influencing decision-making processes in areas such as targeted advertising and matchmaking algorithms. As AI systems become increasingly sophisticated, ensuring their transparency and explainability becomes crucial. Functional transparency is a fundamental aspect of algorithmic decision-making systems, allowing stakeholders to comprehend the inner workings of these systems and enabling them to evaluate their fairness and accuracy. However, achieving functional transparency poses significant challenges that need to be addressed. In this paper, we propose a design for user-centered compliant-by-design transparency in transparent systems. We emphasize that the development of transparent and explainable AI systems is a complex and multidisciplinary endeavor, necessitating collaboration among researchers from diverse fields such as computer science, artificial intelligence, ethics, law, and social science. By providing a comprehensive understanding of the challenges associated with transparency in AI systems and proposing a user-centered design framework, we aim to facilitate the development of AI systems that are accountable, trustworthy, and aligned with societal values.

Question answering (QA) systems have attracted much attention from the artificial intelligence community as they can learn to answer questions based on the given knowledge source (e.g., images in visual question answering). However, the research into question answering systems with human mobility data remains unexplored. Mining human mobility data is crucial for various applications such as smart city planning, pandemic management, and personalised recommendation system. In this paper, we aim to tackle this gap and introduce a novel task, that is, human mobility question answering (MobQA). The aim of the task is to let the intelligent system learn from mobility data and answer related questions. This task presents a new paradigm change in mobility prediction research and further facilitates the research of human mobility recommendation systems. To better support this novel research topic, this vision paper also proposes an initial design of the dataset and a potential deep learning model framework for the introduced MobQA task. We hope that this paper will provide novel insights and open new directions in human mobility research and question answering research.

Fact-checkers are often hampered by the sheer amount of online content that needs to be fact-checked. NLP can help them by retrieving already existing fact-checks relevant to the content being investigated. This paper introduces a new multilingual dataset -- MultiClaim -- for previously fact-checked claim retrieval. We collected 28k posts in 27 languages from social media, 206k fact-checks in 39 languages written by professional fact-checkers, as well as 31k connections between these two groups. This is the most extensive and the most linguistically diverse dataset of this kind to date. We evaluated how different unsupervised methods fare on this dataset and its various dimensions. We show that evaluating such a diverse dataset has its complexities and proper care needs to be taken before interpreting the results. We also evaluated a supervised fine-tuning approach, improving upon the unsupervised method significantly.

Transformer-based language models (LMs) are known to capture factual knowledge in their parameters. While previous work looked into where factual associations are stored, only little is known about how they are retrieved internally during inference. We investigate this question through the lens of information flow. Given a subject-relation query, we study how the model aggregates information about the subject and relation to predict the correct attribute. With interventions on attention edges, we first identify two critical points where information propagates to the prediction: one from the relation positions followed by another from the subject positions. Next, by analyzing the information at these points, we unveil a three-step internal mechanism for attribute extraction. First, the representation at the last-subject position goes through an enrichment process, driven by the early MLP sublayers, to encode many subject-related attributes. Second, information from the relation propagates to the prediction. Third, the prediction representation "queries" the enriched subject to extract the attribute. Perhaps surprisingly, this extraction is typically done via attention heads, which often encode subject-attribute mappings in their parameters. Overall, our findings introduce a comprehensive view of how factual associations are stored and extracted internally in LMs, facilitating future research on knowledge localization and editing.

Adversarial examples can be useful for identifying vulnerabilities in AI systems before they are deployed. In reinforcement learning (RL), adversarial policies can be developed by training an adversarial agent to minimize a target agent's rewards. Prior work has studied black-box versions of these attacks where the adversary only observes the world state and treats the target agent as any other part of the environment. However, this does not take into account additional structure in the problem. In this work, we study white-box adversarial policies and show that having access to a target agent's internal state can be useful for identifying its vulnerabilities. We make two contributions. (1) We introduce white-box adversarial policies where an attacker observes both a target's internal state and the world state at each timestep. We formulate ways of using these policies to attack agents in 2-player games and text-generating language models. (2) We demonstrate that these policies can achieve higher initial and asymptotic performance against a target agent than black-box controls. Code is available at https://github.com/thestephencasper/lm_white_box_attacks