How can large language models (LLMs) serve users with varying preferences that may conflict across cultural, political, or other dimensions? To advance this challenge, this paper establishes four key results. First, we demonstrate, through a large-scale multilingual human study with representative samples from five countries (N=15,000), that humans exhibit significantly more variation in preferences than the responses of 21 state-of-the-art LLMs. Second, we show that existing methods for preference dataset collection are insufficient for learning the diversity of human preferences even along two of the most salient dimensions of variability in global values, due to the underlying homogeneity of candidate responses. Third, we argue that this motivates the need for negatively-correlated sampling when generating candidate sets, and we show that simple prompt-based techniques for doing so significantly enhance the performance of alignment methods in learning heterogeneous preferences. Fourth, based on this novel candidate sampling approach, we collect and open-source Community Alignment, the largest and most representative multilingual and multi-turn preference dataset to date, featuring almost 200,000 comparisons from annotators spanning five countries. We hope that the Community Alignment dataset will be a valuable resource for improving the effectiveness of LLMs for a diverse global population.

Artificial intelligence (AI) is rapidly becoming ubiquitous, so much so it has been argued that "AI … is becoming an infrastructure that many services of today and tomorrow will depend upon."25 Current progress in the field of AI is spearheaded by machine learning (ML) techniques such as deep learning, which has rendered many tasks previously thought to be out of reach of AI more or less solved. The past decades have seen an exponential rise in the amount of compute used by ML systems,29 which has led to a subsequent rise in energy consumption and carbon emissions.17,23,37 Beyond carbon emissions, increased production and use of the hardware infrastructure needed for ML is potentially exacerbating broader environmental impacts.15 While on the one hand ML systems can be used for making progress toward the sustainable development goals (SDGs),27,34 on the other hand the factors mentioned here limit the sustainability of ML from an environmental perspective.

Hosted by Ben Byford, The Machine Ethics Podcast brings together interviews with academics, authors, business leaders, designers and engineers on the subject of autonomous algorithms, artificial intelligence, machine learning, and technology's impact on society. This episode we're chatting with Eleanor and Kerry on good technology and if it's even possible, that technology is political, watering down regulation, the magic of AI, the value of human creativity, how Feminism, Aboriginal, and mixed race studies can help AI development, the performative nature of tech, and more… Dr Kerry McInerney (née Mackereth) is a Research Fellow at the Leverhulme Centre for the Future of Intelligence at the University of Cambridge, where she co-leads the Global Politics of AI project on how AI is impacting international relations. She is also a Research Fellow at the AI Now Institute (a leading AI policy thinktank in New York), an AHRC/BBC New Generation Thinker (2023), one of the 100 Brilliant Women in AI Ethics (2022), and one of Computing's Rising Stars 30 (2023). Kerry is the co-editor of the collection Feminist AI: Critical Perspectives on Algorithms, Data, and Intelligent Machines (2023, Oxford University Press), the collection The Good Robot: Why Technology Needs Feminism (2024, Bloomsbury Academic), and the co-author of the forthcoming book Reprogram: Why Big Tech is Broken and How Feminism Can Fix It (2026, Princeton University Press). Dr Eleanor Drage is a Senior Research Fellow at the University of Cambridge Centre for the Future of Intelligence, and teaches AI professionals about AI ethics on a Masters course at Cambridge.

Detecting "toxic" language in internet content is a pressing social and technical challenge. In this work, we focus on PERSPECTIVE from Jigsaw, a state-of-the-art tool that promises to score the "toxicity" of text, with a recent model update that claims impressive results (Lees et al., 2022). We seek to challenge certain normative claims about toxic language by proposing a new benchmark, Selected Adversarial SemanticS, or SASS. We evaluate PERSPECTIVE on SASS, and compare to low-effort alternatives, like zero-shot and few-shot GPT-3 prompt models, in binary classification settings. We find that PERSPECTIVE exhibits troubling shortcomings across a number of our toxicity categories. SASS provides a new tool for evaluating performance on previously undetected toxic language that avoids common normative pitfalls. Our work leads us to emphasize the importance of questioning assumptions made by tools already in deployment for toxicity detection in order to anticipate and prevent disparate harms.

This article is brought to you thanks to the collaboration of The European Sting with the World Economic Forum. The current conversation around AI, ethics and the benefits for our global community is a heated one. The combination of high stakes and a complex, rapidly-adopted technology has created a very real state of urgency and intensity around this discussion. Promoters of the technology love to position AI as a welcome disruptor that could bring about a global revolution. It's all too easy to get caught up in the hype and create a situation whereby the world does not fully benefit from the development of AI technology.

In this chapter we argue that discourses on AI must transcend the language of 'ethics' and engage with power and political economy in order to constitute 'Good Data'. In particular, we must move beyond the depoliticised language of 'ethics' currently deployed (Wagner 2018) in determining whether AI is 'good' given the limitations of ethics as a frame through which AI issues can be viewed. In order to circumvent these limits, we use instead the language and conceptualisation of 'Good Data', as a more expansive term to elucidate the values, rights and interests at stake when it comes to AI's development and deployment, as well as that of other digital technologies. Good Data considerations move beyond recurring themes of data protection/privacy and the FAT (fairness, transparency and accountability) movement to include explicit political economy critiques of power. Instead of yet more ethics principles (that tend to say the same or similar things anyway), we offer four 'pillars' on which Good Data AI can be built: community, rights, usability and politics. Overall we view AI's 'goodness' as an explicly political (economy) question of power and one which is always related to the degree which AI is created and used to increase the wellbeing of society and especially to increase the power of the most marginalized and disenfranchised. We offer recommendations and remedies towards implementing 'better' approaches towards AI. Our strategies enable a different (but complementary) kind of evaluation of AI as part of the broader socio-technical systems in which AI is built and deployed.

This is the fifth blogpost in a series on Artificial Intelligence and Human Rights. Following Data & Society's AI & Human Rights Workshop in April, several participants continued to reflect on the convening and comment on the key issues that were discussed. The following is a summary of articles written by workshop attendees Bendert Zevenbergen, Elizabeth Eagen, and Aubra Anthony. In Marrying Ethics and Human Rights for AI Scrutiny, Bendert Zevenbergen (Princeton University) responds to a post by Christiaan van Veen and Corinne Cath, in which they advocate the value of applying a human rights framework in the development and deployment of AI. Both articles stemmed from workshop debates that considered the relevance of an ethical versus a human rights perspective in AI design and governance.

A variety of technologies have emerged in the last decade that make it easier and cheaper than ever before to make representations of everyday mobile embodiment. Increasing numbers of people are quantifying and self-tracking their everyday lives recording behavioural, biological and environmental data (Beer, 2016; Neff & Nafus, 2016) using a variety of technologies, for example: • lightweight wearable cameras such as the GoPro allowing users to record footage of their most banal everyday activities; • devices such as the Fitbit and Apple Watch bringing continuous physiological monitoring out of the medical realm and into mainstream culture; • apps like Strava allowing people to quantify their cycling, running and walking activities; • lightweight devices for measuring brain activity (EEG) and stimulation (EDA) becoming sufficiently robust and discreet to be used in non-lab environments. None of the underlying technologies are novel, but as they are made accessible in cheaper and more user-friendly packages, new techniques and sources of data are becoming more readily available for geographical analysis. Engagement with these technologies has created a rapidly expanding area of investigation within geography. The emergence of the quantified-self poses both opportunities and dilemmas for geographical thought. We wish to move past simplistic protests that dismiss such technology as offering another take on Haraway's (1988) 'god trick', presenting partial, and highly situated data as objective truth. Instead, this session will build on the potential identified by Delyser and Sui (2013) to take more inventive approaches toward mobile methods. The focus will be on how these technologies can be engaged with by critical geographers to bring new perspectives to their analysis of everyday embodiment.