Reference List – Ethnography

Ethnography & AI Governance

Reference List

The word 'ethnography' is derived from the Greek "ethnos", meaning a people, nation, or cultural group etc. and "graphy" meaning writing.

Margaret Mead - What people say, what people do and what they say they do are entirely different things

This reading list provides a comprehensive set of resources for those interested in AI safety, ethnography, and AI governance. While there are many specialized lists on these topics, this one aims to integrate key readings across disciplines to give a well-rounded foundation.

Structure

The list is organized into 5 main themes:

• Ethnography and AI
• AI Governance
• Ethnography
• Agography
• Evaluations

Within each section, the sources are arranged roughly by relevance, with foundational readings first. This is designed as a starting point for deeper exploration, with more complex or niche sources placed later in each section.

Contributions

We welcome additional suggestions to ensure the reading list remains current and diverse.

History

Anthropology, the study of human societies and cultures, has roots that stretch back to ancient scholars who sought to understand the diversity of human experience. Early figures such as Herodotus (Rosalind 2000) (Greece and Egypt), often regarded as the "Father of History," laid the groundwork for cultural observation through his writings on the customs and practices of different peoples. Similarly, Ibn Khaldun (Marko 2019) (North Africa and the Middle East), in his seminal work Muqaddimah, explored the dynamics of civilization, emphasizing the interplay of environment, society, and historical development.

The philosophical inquiries of Plato (Greece) and Aristotle (Greece) further enriched the discourse on human nature, while Hegel (Germany) influenced later anthropological thought with his ideas on historical progress. The 19th century marked a significant turning point with Charles Darwin (the Galápagos Islands, South America) and his theory of evolution, which reshaped perspectives on humanity's development.

In the 20th century, Edward Tylor (Mexico and the Americas) and Franz Boas (North America, especially the Pacific Northwest) emerged as key figures, establishing culture as a fundamental concept in anthropology and advocating for cultural relativism through rigorous fieldwork. Additionally, the contributions of thinkers like Carl Jung (Leon 2003) (Switzerland and New Zealand), particularly his psychoanalytic explorations in New Zealand, highlight the interdisciplinary nature of understanding human behavior and the influence of diverse cultures on individual psychology. We seek to apply ethnographic methods in the study of AI systems.

Methods

In conducting effective research, a variety of methodologies can be employed to gather data and insights. Each method serves a specific purpose and can significantly impact the quality and relevance of the findings. Below are some key research methods commonly used in qualitative and action-oriented studies:

• Participation & Observation: Engaging with the community while observing behaviors.
• Action Research: Collaborating with participants to identify and solve problems.
• Direct Observation: Systematic observation in natural settings.
• Interviews: In-depth conversations for qualitative insights.
• Focus Groups: Group discussions to explore perceptions and attitudes.
• Case Studies: Detailed analysis of specific instances.
• Surveys & Questionnaires: Quantitative data collection through standardized questions.
• Ethnographic Mapping: Visual representation of cultural interactions.
• Document Analysis: Review of existing documents for contextual insights.
• Participatory Design: Involving stakeholders in the design process.
• Work Practice Studies(WPS)

Ethnography

• Studying Those Who Study Us (Diane .F)
• Ethnographic artificial intelligence (A. Blackwell 2021)
  • Ethnographic study of the culture and assumptions within artificial intelligence research. This paper describes a research agenda that sets out to question those assumptions, through a programme of ethnographic field work, collaborating with computer scientists and educators in several countries of sub-Saharan Africa.
• Anthropology and the AI-Turn in Global Governance (Maria 2021)
  • Ethnography can contribute to understanding the effects of AI on global governance, institutions, and international law.
• Anthropology, AI and the Future of Human Society 2022
  • This conference announcement about the relationship between anthropology, AI, and the future of human society.(Elie 2022) conference
  • Pervasive AI as ethnographer — Coventry University position
• Navigating Incommensurability Between Ethnomethodology, Conversation Analysis, and Artificial Intelligence (S. Reeves 2022)
  • Inspired by discussions amongst a growing network of researchers in ethnomethodology(EM) and conversation analysis(CA) traditions, this piece is about the disciplinary and conceptual questions that might be encountered, and may need addressing for engagements with AI research and its fields.
• Anthropology and the rise of artificial intelligence (Ahmed 2024)
• "If it is easy to understand then it will have value": Examining Perceptions of Explainable AI with Community Health Workers in Rural India (Chinasa 2024)
• The emergence of multispecies ethnography (S. Kirksey 2010)
• ​​Studying Those Who Study Us: An Anthropologist in the World of Artificial Intelligence. (Diana E. Forsythe).
• Hammersley, Martyn, and Paul Atkinson. "Ethnography: Principles in Practice." Routledge, 2019.
• Pink, Sarah. "Doing Sensory Ethnography." SAGE Publications, 2015.
• Stoller, Paul. "The Taste of Ethnographic Things: The Senses in Anthropology." University of Pennsylvania Press, 2012.
• Ingold, Tim. The Perception of the Environment: Essays on Livelihood, Dwelling, and Skill. Routledge, 2011.
• Clifford, James, and George E. Marcus (eds). Writing Culture: The Poetics and Politics of Ethnography. University of California Press, 2010.
• Geertz, Clifford. The Interpretation of Cultures: Selected Essays. Basic Books, 1973.
• Rosalind Thomas. Herodotus in context : ethnography, science, and the art of persuasion.
• Leon Petchkovsky Jung and the Dreaming: Analytical Psychology’s Encounters with Aboriginal Culture
• Johnson, J.M. Doing field research, New York: The Free Press, 1975.
• Suchman, L. A. (1987). Plans and situated actions: The problem of human-machine communication. Cambridge University Press.
• Jaton, F. (2021) The Constitution of Algorithms. Cambridge, MA, US: MIT Press.
• Vertesi, J., and Ribes, D. (eds) digitalSTS: A Field Guide for Science & Technology Studies.
• Hashizume, A., Kurosu, M. (2013). Understanding User Experience and Artifact Development through Qualitative Investigation: Ethnographic Approach for Human-Centered Design. In: Kurosu, M. (eds) Human-Computer Interaction. Human-Centred Design Approaches, Methods, Tools, and Environments. HCI 2013. Lecture Notes in Computer Science, vol 8004. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-642-39232-0_8.

AI Governance

• Hildebrandt, Mireille. "AI Governance: A Human Rights-Based Framework." Springer International Publishing, 2022.
• Jobin, Annabelle, et al. "The Global Landscape of AI Ethics Guidelines." Berkman Klein Center for Internet & Society at Harvard University, 2019.
• Lepri, Bruno, et al. "A Research Agenda for AI Governance." arXiv preprint arXiv:2006.06228, 2020.
• Binns, Reuben. "Fairness in Machine Learning: Lessons from Political Philosophy." Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency (FAT), ACM, 2020.
• Floridi, Luciano. The Ethics of Information. Oxford University Press, 2013.
• Zuboff, Shoshana. The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power. PublicAffairs, 2019.

Ethnography and AI Governance

• Decolonizing LLMs: An Ethnographic Framework for AI in African Contexts (2024)
• Towards Measuring the Representation of Subjective Global Opinions in Language Models – Anthropic
• Where Responsible AI meets Reality: Practitioner Perspectives on Enablers for shifting Organizational Practices
  • Ethnographic study that interviewed 26 responsible AI practitioners.
  • The paper explores how organizational culture and structure influence the implementation of responsible AI initiatives within large technology companies, aiming to enhance algorithmic accountability. Through qualitative interviews with industry practitioners, it identifies common challenges, ethical tensions, and enablers for responsible AI, mapping current organizational supports and obstacles while outlining aspirational processes and structures needed for effective transitions in practice.
• How to use LLMs in the social sciences with statistical guarantees (Naoki Egami et al. 2023; Egami et al. 2024)
  Frameworks(Egami 2023) and methodologies to harness ML and AI for social science questions while maintaining statistical validity.
• Algorithms as culture: Some tactics for the ethnography of algorithmic systems (Nick Seaver).
  The article discusses three ethnographic tactics for studying algorithms, these approaches enhance understanding of algorithms in diverse contexts:
  • Scavenging: Using informal sources and public data for insights.
  • Texture of Access: Recognizing varied access to algorithmic systems.
  • Interviews as Fieldwork: Viewing interviews as cultural interactions.
• Constitutional AI allows Anthropic to train better and more harmless AI assistants without any human feedback labels for harms.
• Boyd, danah, and Karen Levy. "AI Ethnography: A Method for Developing a Socio-Technical Understanding of Automated Systems." Proceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and Society, pp. 225-231, 2019.
• Crawford, Kate. "Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence." Yale University Press, 2021.
• Eubanks, Virginia. "Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor." St. Martin's Press, 2018.
• Suchman, Lucy. Human-Machine Reconfigurations: Plans and Situated Actions. Cambridge University Press, 2007.
• Irani, Lilly. Chasing Innovation: Making Entrepreneurial Citizens in Modern India. Princeton University Press, 2019.
• Anthropomorphisation
  • AI Is the Perfect Student: Anthropocene, Technocene, and the Ends of Intelligence
• Ethnographic audit trails: Using ethnographic methods to systematically keep track of the contextual details factored into the decisionmaking processes at AI development spaces (Yung-Hsuan Wu, 2024)

Human-centered AI

• Looking at human-centered artificial intelligence as a problem and prospect for sociology: An analytic review. Rezaev and Tregubova (2023)
• A Case for Design Anthropology for Creating Human-Centered AI

Ethnography and AI Evaluations

• Human-centered evaluation of explainable AI applications: a systematic review

Agent Ethnography (Agographia)

Agographia is derived from the Latin word agēns,meaning "agent" or "doer," and the suffix "-graphia," which signifies "study." Thus, Agographia can be understood as "the study of agents."

These studies look at simulation of agents, including understanding social contexts of behavior of AI agents.

• AI systems observing humans

  • Syncopathy
    Sycophancy is an undesirable behavior where models tailor their responses to follow a human user's view even when that view is not objectively correct (e.g., adapting liberal views once a user reveals that they are liberal). Human feedback is commonly utilized to finetune AI assistants. But human feedback may also encourage model responses that match user beliefs over truthful ones, a behavior known as sycophancy. Google DeepMind used simple synthetic data to reduce sycophancy in large language models

  • Machiavellianism
    Models are rapidly being deployed in the real world. How do we evaluate models, especially ones as complex as GPT-4, to ensure that they behave safely in pursuit of their objectives? Can we design models that robustly avoid any harm while achieving their goals?

  • Using the behavioral sciences to better understand Large Language Models. Given that LLMs can behave in unhuman-like ways, can we make them more human-like?

    • CogBench: a large language model walks into a psychology lab
    • [2402.01821] Human-like Category Learning by Injecting Ecological Priors from Large Language Models into Neural Networks
    • [2311.16093] Visual cognition in multimodal large language models
    • How do we use methods from developmental psychology to assess AI models?
      Can developmental psychology be used to understand LLMs?
    • LLMs can be turned into a cognitive model, predicting choices for unseen tasks.

  • “Collaborating” with AI: Taking a System View to Explore the Future of Work

    • It is argued that relational ethnographic approaches can assist organization theorists in navigating the methodological challenges of taking a counterpart perspective and several strategies for future research are proposed.
• Humans observing AI systems
  • ([Philipp](https://academic.oup.com/edited-volume/55209/chapter-abstract /430648569) Brandt 2023) Machine Learning, Abduction, and Computational Ethnography
    • The main section develops the computational ethnography approach across a series of examples, including computational analyses of historical and contemporary corpora, classic and recent network-analytic studies
    • The Erotetic Theory
      • Humans in Humans Out: On GPT Converging Toward Common Sense in both Success and Failure
      • P. Koralus & S. Moss (2023): Reasoning with Uncertainty.
    • Scalable Oversight
      Measuring Progress on Scalable Oversight for Large Language Models
      How humans could use AI systems to better oversee other AI systems, and demonstrate some proof-of-concept results where a language model improves human performance at a task.
  • Why not a Sociology of Machines? The Case of Sociology and Artificial Intelligence - Steve Woolgar, 1985
• AI systems observing AI systems
  • Behaviour Suite for Reinforcement Learning | OpenReview
  • Evaluating Psychological Safety of Large Language Models
  • [2409.03793] Safeguarding AI Agents: Developing and Analyzing Safety Architectures
  • Claude’s Character – Anthropic
    Character training is an open area of research and our approach to it is likely to evolve over time. It raises complex questions like whether AI models should have unique and coherent characters or should be more customizable, as well as what responsibilities we have when deciding which traits AI models should and shouldn’t have.

Contextual Evaluations

• Beyond Aesthetics: Cultural Competence in Text-to-Image Models
  • *The paper introduces a framework for evaluating the cultural competence of Text-to-Image models, focusing on cultural awareness and diversity, through a new benchmark called CUBE. The evaluations reveal significant gaps in the cultural awareness of existing models across countries and provide valuable insights into the cultural diversity*
• We need a Science of Evals — LessWrong
• Evaluating Syncopathy "Towards Understanding Sycophancy in Language Models" code
• Sparse Autoencoders Reveal Temporal Difference Learning in Large Language Models

Video Jacki @MRI: Ethnography for Artificial Intelligence