The Critical AI Centre (CrAIC)

The project Favel IA has been co-designed with the Papo Reto Institute (IPR), a favela media activist and human rights organisation based in Rio de Janeiro. The project analyses a core question developed jointly with IPR: How does the favela use AI as opposed to how AI uses the favela in extractive ways? Drawing from critical and decolonial data and AI studies (Valente and Grohmann, 20024; Goodlad, 2023; Ricaurte, 2019), Favel IA interrogates the politics of empathy as a form of solidarity. It is also attentive to the gendered, raced, classed, and transnational geopolitical dimensions that shape who is permitted to perform empathy and who is positioned as its beneficiary. These dynamics contrast sharply with the everyday invisible forms of political empathy exercised by favela communities, which are rarely recognised or valued.  

Favel IA employs participatory action research (PAR) as its main methodological approach, inspired by Latin American scholars such as Orlando Fals Borda (1987) and Paulo Freire (2002). The project involves three main groups: postgraduate students, favela media activists, and community leaders. The research team engaged participants in five consecutive days of workshops. We asked them to document their daily encounters with AI, to carry out empathy exercises and to reflect on critical questions about AI and data from a favela standpoint. Collectively, we also analysed existing and emerging AI regulation in Brazil and elsewhere and co-devised a conceptual note for policymakers, articulating the types of AI favela residents desire and the alternative futures that they envision. 

The name Favel IA is a reference to the favelas themselves as forms of intelligence and to inteligência artificial, artificial intelligence in Portuguese. IÁ is also a common ending to many samba refrains (la iá la iá), evoking the idea that favela residents must “dance to the music” of AI systems designed by powerful private companies. While digital media platforms constrain agency, participants also identified instances of mundane resistance (Madianou, 2025), using generative AI to craft texts that signal socioeconomic mobility, employing AI tools to decode complex legislation and explain it for favela residents and/or using chatbots for emotional support. The latter, especially the use of AI as a therapist in contexts of grief caused by racist police violence, prompted critical debates about the myth of “empathetic AI.” These reflections culminated in a collective call to favelise AI, that is, to invert the centre-periphery relationship so that the lived experiences and voices off favela residents become central rather than marginal in AI debates and governance. 

The project has been funded by the INCT DSI Consortium (Brazil). 

Principal Investigator: Andrea Medrado 

Co-Investigator: Thainã de Medeiros   

Wikipedia sits at the heart of the Internet’s knowledge infrastructure, openly and collaboratively documenting the ever-evolving record of human understanding. As a key training source for large language models, it now shapes the information that AI produces—yet these AI writing tools are increasingly being used by editors back on the platform itself. This project investigates how this reciprocal relationship may reproduce or reinforce biases within collective knowledge systems, focusing on the issue of notability: what and who is included in, or excluded from, these resources. Through ethnographic and computational analysis, including a public “edit-AI-thon”, we will examine how algorithmic and human practices intertwine in defining who and what counts.

This project is funded by the UKRI AHRC BRAID DOT programme

Project lead: Patrick Gildersleve

Project co-lead (International): Francesca Tripodi

Project co-lead: Brett Zehner

The release of ChatGPT-5 disappointed many, with Sam Altman acknowledging its flaws. Neural networks’ limitations are increasingly apparent. Over fifty years after Marvin Minsky and Seymour Papert warned of their fragility, AI models generate impressive outputs but lack a proper understanding of tasks. Their logic is fragile, trained on narrow datasets that often ignore context and nuance. They produce responses that seem insightful but are superficial, mirroring biases in their training data.

This newfound necessity for more quality human data could call for forms of creativity beyond statistical aesthetics. The qualitative turn in AI research represents a value proposition for the humanities. Stagnation underscores the importance of human insight, reasoning, and creativity. As AI increasingly relies on synthetic data, tech firms are hiring humanities scholars to guide product development.

An applied humanities approach is now necessary that emphasizes social and ethical dimensions at the local level. This research advocates for qualitative critique at every stage—data collection, algorithm design, and output interpretation—to remain attentive to context and difference. Insights from the humanities expose biases and inequalities.

Brett is currently working with the bourgeoning group – The Public Intelligence research group – which explores the risks of knowledge dispossession in the AI boom. The group aims to develop new models of community intelligence that draw on richer forms of collective knowledge, examining the role of universities in local knowledge ecosystems and how diversity can drive innovation.

Principal Investigator: Brett Zehner