Joy Buolamwini
Founder · Algorithmic Justice League
Showed that commercial face analysis failed more often on darker-skinned women, and built a public practice of auditing systems that claim to see us.
Voices
Who to read in each topic
Not a ranking. People and groups whose public work helps you go deeper. Save a piece to review later, or open it in a new tab.
How data is gathered, whose lives it represents, and whether it can be trusted.Whether models do what they claim, how they fail, and how we know.Attacks on and through AI systems: injection, jailbreaks, cyber offense.Whether people can see, contest, and hold to account what a system does.When software acts across time and tools, faster than people can intervene.The physical and operational stack models run on — clouds, chips, outages.Energy, water, minerals, and the living world that compute depends on.
Timnit Gebru
Founder · Distributed AI Research Institute
Named the labour, environmental, and stereotyping costs of large language models, and built an independent lab after leaving a major lab.
Stuart Russell
Professor of computer science · UC Berkeley · CHAI
Argues that the standard model of AI, optimizing a fixed objective, is the wrong shape, and proposes machines that remain uncertain about human preference.
Kate Crawford
Research professor · USC Annenberg
Maps AI as an extractive industry: minerals, energy, labour, and classification. Atlas of AI is the book most groups can start with.
Sasha Luccioni
Climate and AI researcher · Hugging Face
Measures the energy and carbon cost of training and serving models, so environmental claims can be compared rather than asserted.
Arvind Narayanan
Professor of computer science · Princeton
Cuts through AI marketing. AI Snake Oil is a calm test of which claims are real, which are premature, and which cannot work as sold.
Bruce Schneier
Security technologist · Harvard Kennedy School
Writes about security as a social problem. A reliable first read when a group is mixing fear of hackers with fear of the model itself.
Yoshua Bengio
Scientific director · Mila · LawZero
A founder of modern deep learning who now spends his public voice on catastrophic risk, agentic systems, and international scientific advice.
Cathy O'Neil
Mathematician and writer
Weapons of Math Destruction is still the plainest account of scoring systems that punish the people least able to contest them.
Gary Marcus
Scientist and writer
A persistent public critic of overclaiming. Useful when a group needs permission to ask 'does this actually work?'
Nicholas Carlini
Research scientist · Anthropic
Shows, with working attacks, what models leak and how they fail. A technical voice that still reads in a mixed group if you pick one paper.
Inioluwa Deborah Raji
Researcher · UC Berkeley
Wrote the case for closing the AI accountability gap: documentation, audit, and the right to know what a system was tested on.
NIST AI Risk Management Framework
Voluntary framework · U.S. National Institute of Standards and Technology
The document many institutions already have to map against. Not a voice in the personal sense, but the shared language of governance conversations.
Abeba Birhane
Cognitive scientist · Trinity College Dublin · Mozilla
Writes about relational ethics and the colonial patterns that hide in datasets. A voice for culture, identity, and who is made into data.
Emily M. Bender
Professor of linguistics · University of Washington
Insists that language models are not minds. A clarifying voice when a group is mixing 'the model said' with 'a person claims.'
Helen Nissenbaum
Professor · Cornell Tech
Contextual integrity: privacy is not secrecy, it is appropriate flow. The idea that lets a mixed group talk about data without collapsing into 'share nothing.'