Building for the Long Tail

Correctness is all in the eye of the beholder.

In this CRAFT workshop, we invite machine learning researchers and practitioners building small, task specific models addressing the needs of specific communities they know and are parts of, to discuss how knowledge of their communities impacts the decisions they make while building ML products.

FAccT 2024, Brazil

Background

Big Tech’s “one-model-for-everything” approach aids centralization of power by multinational corporations, while resulting in subpar performance, even by metrics used by these companies, for non-dominant populations. Simultaneously, hype from Big Tech companies promoting the one-model-for-everything approach, has harmed small, community-rooted organizations which focus on building task-specific models suited for their contexts. These small organizations have developed models that outperform those from Big Tech because they understand the complexities of the cultures they operate in. This contextual understanding results in datasets, model architectures, and evaluation metrics that are different from those of Big Tech. In this session, we ask: how do the design choices of a researcher deeply familiar with the social context of a particular community, differ from those of a researcher with limited understanding of this context? We will use case studies and participatory methods to map how contextual knowledge impacts what types of machine learning models are built.

Speakers

Caleb Moses

Caleb Moses is of Māori descent from the Te Mahurehure sub-tribe in Hokianga, New Zealand. He is currently a PhD student of computer science at McGill University and Mila Quebec where he is supervised by Prof Jackie Cheung (CompSci) and Prof Noelani Arista (Indigenous Studies). His research focus is on AI applications to support Language Revitalisation, and encompasses digital archives, Indigenous traditional knowledge and language education.

Kathleen Siminyu

Kathleen Siminyu is an Artificial Intelligence Researcher working in Natural Language Processing. She is a researcher at DAIR. Kathleen focuses her research on African languages as a form of activism and from a desire to see these languages better represented on digital platforms. In her research, Kathleen has worked on resources for Kiswahili speech recognition as part of the Mozilla Common Voice project, on machine translation for four Kenyan languages as part of Masakhane and on phoneme transcription for Luhya languages in collaboration with Neulab at CMU.


Kathleen also works with African AI communities to enable ecosystem capacity building and research relevant to Africa. For some of these efforts, Kathleen was listed as one of the MIT Technology Review 35 Innovators under 35 for 2022. She continues to organise with communities as part of the Deep Learning Indaba, where she is a trustee and the Masakhane Research Foundation, where she is a director. Among her other contributions to the African AI ecosystem, Kathleen served as Regional Coordinator of AI4D Africa in the first 2 years of the programme’s inception and was part of an African Union expert consultative committee working to develop a continental strategy for Artificial Intelligence in Africa.

Ivana Feldfeber Kisilevsky

Ivana Feldfeber is a feminist activist from Bariloche, Argentina. She is the co-founder and Executive Directress of the first Gender Data Observatory in Latin America, "DataGénero." She is part of the Feminist AI network for Latin America and she was a team leader at the Center of Artificial Intelligence and Digital Policy (CAIDP), where she worked to analyze different AI policies in her region. Ivana has been working as a programming and robotics instructor since 2017 and holds a postgraduate diploma in Data Science, Machine Learning, and its Applications from the Universidad de Córdoba, Argentina. In 2022 Ivana and her team were selected by A+ Alliance to research AI tools for Criminal Courts working with gender-based violence data in Argentina. With DataGénero, Ivana worked with several governments and companies to build inclusive data processes, train teams, write recommendations and help decision-makers to create better data policies.

Photo of Nina, a Black woman with short, curly black hair. She is wearing glasses, long earrings, and a bright teal shirt. She has a big smile. The background of the photo is out of focus and white.

Nina da Hora

Nina da Hora is an Afro-Brazilian activist and computer scientist who is renowned for her research on algorithmic racism and her efforts to democratize access to technology by making its operations transparent using accessible language. She is a fellow at the Center for Technology and Society at Fundação Getúlio Vargas (FGV). Her research focuses on mitigating algorithmic racism through the intersection of ethics and artificial intelligence. She is also a senior researcher in Data and AI at Thoughtworks Brazil.Nina has been honored with the Ford Foundation Global Fellowship 2024, the Forbes Under 30 award, the Sabia Award from the University of Cambridge, and she is recognized among the 100 most influential people in Ethics in AI. In 2020, she established the Instituto da Hora, dedicated to advancing digital rights in Brazil and promoting a more collective and critical approach to the field of artificial intelligence in society.

A digital portrait of Joana, who has light skin and long, curly brown hair. One eye is closed and she has a hand held in front of her mouth, covering it. Her other hand frames her open eye, like she's focusing a camera. She wears rings on both hands and a long, thin gold piece of jewelry spans across four fingers on her left hand.

Joana Varon

Joana Varon is the Executive Directress and Creative Chaos Catalyst at Coding Rights, a women-run organization working to expose and redress the power imbalances built into technology and its application, particularly those that reinforce gender and North/South inequalities. She just completed a Technology and Human Rights Fellowship at the Carr Center for Human Rights Policy from Harvard Kennedy School and affiliated to the Berkman Klein Center for Internet and Society at Harvard University. Former Mozilla Media Fellow, she is co-creator of several creative projects operating in the interplay between activism, arts and technologies, such as transfeministech.org, chupadados.com, #safersisters, Safer Nudes, protestos.org, Net of Rights and freenetfilm.org. She is also part of the groups of researchers who kick-started the working group on Human Rights Considerations for Standards and Protocols at the Internet Engineering Task Force (IETF). Brazilian, with Colombian ancestry, she is engaged in several international civil society networks, such as Privacy International Network, the feminist hackers collective DeepLab, the Open Observatory of Network Interference (Ooni), Al Sur, among others, always focused on bringing Latin American perspectives in the search of feminist techno-political frameworks for human rights based development, deployment and usages of technologies.

Shaimaa Lazem

Shaimaa Lazem is an Associate Research Professor at the City of Scientific Research and Technological Applications (SRTA-City), Egypt. She earned her PhD in Computer Science from Virginia Tech, USA with a focus on Human-Computer Interaction (HCI). Her interests include participatory design, community-based design, and Human-centered AI. She is a Leaders-in-Innovation Fellow with the Royal Academy of Engineering in London, and the co-founder of the ArabHCI community. She works on localising Responsible Human-Centered AI in the African tech ecosystem in partnership with Prof. Anicia Peters from Namibia with support from Google 2020 Award for Inclusion Research and Google AI 2021 Awards.

Agenda

[Date & Time]

5 min - Intro
25 min - Keynote talk
50 min - Participatory working session with a moderator
35 min - fireside chat and Q&A/Discussion
5 min - closing

We look forward to seeing you all there!

Organizers
Nyalleng Moorosi

Senior Researcher, DAIR

Raesetje Sefala

Research Fellow, DAIR

Asmelash Teka Hadgu

Founder, Lesan AI and Fellow, DAIR