International AI Safety Report

The first International AI Safety Report comprehensively synthesizes the current evidence on the capabilities, risks, and safety of advanced AI systems. The report was mandated by the nations attending the AI Safety Summit in Bletchley, UK. Thirty nations, the UN, the OECD, and the EU each nominated a representative to the report's Expert Advisory Panel. A total of 100 AI experts contributed, representing diverse perspectives and disciplines. Led by the report's Chair, these independent experts collectively had full discretion over the report's content.

Local vs distributed representations: What is the right basis for interpretability?

Much of the research on the interpretability of deep neural networks has focused on studying the visual features that maximally activate individual neurons. However, recent work has cast doubts on the usefulness of such local representations for understanding the behavior of deep neural networks because individual neurons tend to respond to multiple unrelated visual patterns, a phenomenon referred to as "superposition". A promising alternative to disentangle these complex patterns is learning sparsely distributed vector representations from entire network layers, as the resulting basis vectors seemingly encode single identifiable visual patterns consistently. Thus, one would expect the resulting code to align better with human perceivable visual patterns, but supporting evidence remains, at best, anecdotal. To fill this gap, we conducted three large-scale psychophysics experiments collected from a pool of 560 participants. Our findings provide (i) strong evidence that features obtained from sparse distributed representations are easier to interpret by human observers and (ii) that this effect is more pronounced in the deepest layers of a neural network. Complementary analyses also reveal that (iii) features derived from sparse distributed representations contribute more to the model's decision. Overall, our results highlight that distributed representations constitute a superior basis for interpretability, underscoring a need for the field to move beyond the interpretation of local neural codes in favor of sparsely distributed ones.

The Disparate Benefits of Deep Ensembles

Ensembles of Deep Neural Networks, Deep Ensembles, are widely used as a simple way to boost predictive performance. However, their impact on algorithmic fairness is not well understood yet. Algorithmic fairness investigates how a model's performance varies across different groups, typically defined by protected attributes such as age, gender, or race. In this work, we investigate the interplay between the performance gains from Deep Ensembles and fairness. Our analysis reveals that they unevenly favor different groups in what we refer to as a disparate benefits effect. We empirically investigate this effect with Deep Ensembles applied to popular facial analysis and medical imaging datasets, where protected group attributes are given and find that it occurs for multiple established group fairness metrics, including statistical parity and equal opportunity. Furthermore, we identify the per-group difference in predictive diversity of ensemble members as the potential cause of the disparate benefits effect. Finally, we evaluate different approaches to reduce unfairness due to the disparate benefits effect. Our findings show that post-processing is an effective method to mitigate this unfairness while preserving the improved performance of Deep Ensembles.

Lookism: The overlooked bias in computer vision

In recent years, there have been significant advancements in computer vision which have led to the widespread deployment of image recognition and generation systems in socially relevant applications, from hiring to security screening. However, the prevalence of biases within these systems has raised significant ethical and social concerns. The most extensively studied biases in this context are related to gender, race and age. Yet, other biases are equally pervasive and harmful, such as lookism, i.e., the preferential treatment of individuals based on their physical appearance. Lookism remains under-explored in computer vision but can have profound implications not only by perpetuating harmful societal stereotypes but also by undermining the fairness and inclusivity of AI technologies. Thus, this paper advocates for the systematic study of lookism as a critical bias in computer vision models. Through a comprehensive review of existing literature, we identify three areas of intersection between lookism and computer vision. We illustrate them by means of examples and a user study. We call for an interdisciplinary approach to address lookism, urging researchers, developers, and policymakers to prioritize the development of equitable computer vision systems that respect and reflect the diversity of human appearances.

Leveraging Large Language Models to Measure Gender Bias in Gendered Languages

Gender bias in text corpora used in various natural language processing (NLP) contexts, such as for training large language models (LLMs), can lead to the perpetuation and amplification of societal inequalities. This is particularly pronounced in gendered languages like Spanish or French, where grammatical structures inherently encode gender, making the bias analysis more challenging. Existing methods designed for English are inadequate for this task due to the intrinsic linguistic differences between English and gendered languages. This paper introduces a novel methodology that leverages the contextual understanding capabilities of LLMs to quantitatively analyze gender representation in Spanish corpora. By utilizing LLMs to identify and classify gendered nouns and pronouns in relation to their reference to human entities, our approach provides a nuanced analysis of gender biases. We empirically validate our method on four widely-used benchmark datasets, uncovering significant gender disparities with a male-to-female ratio ranging from 4:1 to 6:1. These findings demonstrate the value of our methodology for bias quantification in gendered languages and suggest its application in NLP, contributing to the development of more equitable language technologies.

Can ChatGPT Read Who You Are?

The interplay between artificial intelligence (AI) and psychology, particularly in personality assessment, represents an important emerging area of research. Accurate personality trait estimation is crucial not only for enhancing personalization in human-computer interaction but also for a wide variety of applications ranging from mental health to education. This paper analyzes the capability of a generic chatbot, ChatGPT, to effectively infer personality traits from short texts. We report the results of a comprehensive user study featuring texts written in Czech by a representative population sample of 155 participants. Their self-assessments based on the Big Five Inventory (BFI) questionnaire serve as the ground truth. We compare the personality trait estimations made by ChatGPT against those by human raters and report ChatGPT's competitive performance in inferring personality traits from text. We also uncover a 'positivity bias' in ChatGPT's assessments across all personality dimensions and explore the impact of prompt composition on accuracy. This work contributes to the understanding of AI capabilities in psychological assessment, highlighting both the potential and limitations of using large language models for personality inference. Our research underscores the importance of responsible AI development, considering ethical implications such as privacy, consent, autonomy, and bias in AI applications.

Addressing Membership Inference Attack in Federated Learning with Model Compression

Federated Learning (FL) has been proposed as a privacy-preserving solution for machine learning. However, recent works have shown that Federated Learning can leak private client data through membership attacks. In this paper, we show that the effectiveness of these attacks on the clients negatively correlates with the size of the client datasets and model complexity. Based on this finding, we propose model-agnostic Federated Learning as a privacy-enhancing solution because it enables the use of models of varying complexity in the clients. To this end, we present $\texttt{MaPP-FL}$, a novel privacy-aware FL approach that leverages model compression on the clients while keeping a full model on the server. We compare the performance of $\texttt{MaPP-FL}$ against state-of-the-art model-agnostic FL methods on the CIFAR-10, CIFAR-100, and FEMNIST vision datasets. Our experiments show the effectiveness of $\texttt{MaPP-FL}$ in preserving the clients' and the server's privacy while achieving competitive classification accuracies.

Algorithms for Social Justice: Affirmative Action in Social Networks

Link recommendation algorithms contribute to shaping human relations of billions of users worldwide in social networks. To maximize relevance, they typically propose connecting users that are similar to each other. This has been found to create information silos, exacerbating the isolation suffered by vulnerable salient groups and perpetuating societal stereotypes. To mitigate these limitations, a significant body of work has been devoted to the implementation of fair link recommendation methods. However, most approaches do not question the ultimate goal of link recommendation algorithms, namely the monetization of users' engagement in intricate business models of data trade. This paper advocates for a diversification of players and purposes of social network platforms, aligned with the pursue of social justice. To illustrate this conceptual goal, we present ERA-Link, a novel link recommendation algorithm based on spectral graph theory that counteracts the systemic societal discrimination suffered by vulnerable groups by explicitly implementing affirmative action. We propose four principled evaluation measures, derived from effective resistance, to quantitatively analyze the behavior of the proposed method and compare it to three alternative approaches. Experiments with synthetic and real-world networks illustrate how ERA-Link generates better outcomes according to all evaluation measures, not only for the vulnerable group but for the whole network. In other words, ERA-Link recommends connections that mitigate the structural discrimination of a vulnerable group, improves social cohesion and increases the social capital of all network users. Furthermore, by promoting the access to a diversity of users, ERA-Link facilitates innovation opportunities.

FairShap: A Data Re-weighting Approach for Algorithmic Fairness based on Shapley Values

In this paper, we propose FairShap, a novel and interpretable pre-processing (re-weighting) method for fair algorithmic decision-making through data valuation. FairShap is based on the Shapley Value, a well-known mathematical framework from game theory to achieve a fair allocation of resources. Our approach is easily interpretable, as it measures the contribution of each training data point to a predefined fairness metric. We empirically validate FairShap on several state-of-the-art datasets of different nature, with different training scenarios and models. The proposed approach outperforms other methods, yielding significantly fairer models with similar levels of accuracy. In addition, we illustrate FairShap's interpretability by means of histograms and latent space visualizations. We believe this work represents a promising direction in interpretable, model-agnostic approaches to algorithmic fairness.

BIASeD: Bringing Irrationality into Automated System Design

Human perception, memory and decision-making are impacted by tens of cognitive biases and heuristics that influence our actions and decisions. Despite the pervasiveness of such biases, they are generally not leveraged by today's Artificial Intelligence (AI) systems that model human behavior and interact with humans. In this theoretical paper, we claim that the future of human-machine collaboration will entail the development of AI systems that model, understand and possibly replicate human cognitive biases. We propose the need for a research agenda on the interplay between human cognitive biases and Artificial Intelligence. We categorize existing cognitive biases from the perspective of AI systems, identify three broad areas of interest and outline research directions for the design of AI systems that have a better understanding of our own biases.