aiaiai. Critical explorations of artificial intelligence and digital culture

Hangar presents aiaiai, a programme of public events exploring and debating the transformations shaping contemporary digital realities. Conceived by the Soft/Hardware and Interaction Lab and the Interfaces and Networks Lab, the programme is part of Open Thursdays, a space for sharing technological research and experimentation with the community.

Across three sessions, professionals from different fields — artists, technologists, activists and thinkers — will share knowledge, experiences and critical perspectives on key issues shaping our digital present. In this second edition, the programme will focus on urgent and complex questions, including artificial intelligence and its social impact, strategies for breaking free from corporate platforms in digital environments, and the ethical challenges posed by emerging technological paradigms.

aiaiai aims to create an open, non-hierarchical space where knowledge is not simply shared but collectively produced. Through talks, demonstrations, discussions and participatory activities, audiences are invited to imagine, question and redefine how we relate to technology.

1. Bias and diversity: languages, cultures and underrepresented communities
5 November
7–9 pm
Sala Ricson, Hangar
Free admission

What the model has never seen does not exist.

Most large AI models are trained on data from the Global North, predominantly in English and shaped by an Anglophone worldview. This bias is not an isolated flaw that can be patched; it is structural, embedded in the training corpus, the annotations and the decisions about what counts as “normal”. When these models are widely used to write, translate, summarise or answer questions, they return a version of the world filtered through that dominant perspective.

The result is a gradual erosion of everything that falls outside the average: languages such as Catalan, which models recognise but distort; communities such as Deaf communities and their sign languages, for which data is scarce and which are consequently left out; and diversity in a broader sense — gender, bodies and ways of life — which is simplified, caricatured or erased altogether. What the model has never seen does not exist.

In this session, we will look at where this data comes from, what gets lost along the way, and how to build alternatives: small, sovereign models and participatory processes that place affected communities at the centre of design.

Drawing on his experience working in Catalonia, Xavi Vinaixa will explain how bias in large models affects languages, cultures and underrepresented communities. He will begin with the origins of training data and the “data wall”: the point at which the supply of available online text, mostly in English, runs out, and what happens next.

He will present his research L’espoli silenciós (The Silent Plunder), published with Marçal Font. The research documents the large-scale, opaque purchasing of second-hand books across Europe through intermediaries and logistics networks that conceal the end customer, with AI training cited as the intended use. This practice puts out-of-print books in minoritised languages at risk, some of which survive in only a single copy.

Drawing on his research into how large models distort Catalan — including Coca is not cocaine and FLAMA, a family of models trained specifically in Catalan — he will demonstrate how bias operates in practice: translations that reproduce English structures, dominant cultural references replacing local ones, and nuances disappearing.

He will also discuss Signària, a project developed with Pompeu Fabra University (UPF) to enable sign recognition and synthetic voice for Catalan Sign Language (LSC). It is a telling example of a language and community that AI overlooks because so little data is available.

Lluís Gómez Bigorda will present InterVisions, a participatory, interdisciplinary initiative exploring how AI systems can reproduce bias and inequality in the ways they represent the world. The project invites diverse communities to reflect on these technologies and collectively imagine AI tools that are more inclusive, responsible and respectful of diversity.

Through a critical, intersectional feminist lens, InterVisions seeks to create spaces for dialogue, experimentation and creative practice around the future of artificial intelligence.

Programme
— Where does the data used to train AI systems come from?
— What happens to non-Anglophone minorities?

Participants
— Xavi Vinaixa, computer scientist, CTO at Sorensen.ai, researcher and science communicator who discusses ethical technology from a humanist perspective on television and radio programmes.
— Lluís Gómez Bigorda, Ramón y Cajal researcher at the UAB Department of Computer Science and the Computer Vision Center.

2. From the ethereal to the physical
12 November
7–9 pm
Sala Ricson, Hangar
Free admission

Everyone is talking about using an AI model to summarise something, create a design, do the shopping or write an email. But one question remains: where do these models actually run?

AI gigafactories are monopolising the planet’s resources across multiple sectors, from mineral extraction to environmental pollution. They are also affecting the consumer market by absorbing supplies of components that would otherwise be available to individual users.

The extraordinary sums of money flowing through these major companies have led manufacturers of all kinds of memory and GPUs to prioritise supplying the centralised data-centre industry over individual consumers. The result has been an exponential and staggering rise in the prices of components essential to modern computers.

This process of recentralisation stands in stark contrast to the developments of the first decade of the new millennium, when people outside the major tech industry gained the ability to experiment with a wide range of electronic products, take control of them and adapt their use to their own needs. This shift was made possible by the emergence of accessible hardware development boards, which made it possible to build digital, interactive devices capable of sensing and controlling physical objects.

The Arduino project was a pioneer in making low-cost hardware and relatively accessible programming available, following the principles of open-source hardware and software. This helped give rise to a wide range of projects and maker communities: people who have brought the world of open-source software closer to hardware, and connected open experimentation with electronics and industrial products, paving the way for new artistic formats.

Today, we are facing a new paradigm shaped by the rise of AI models and the shortage of consumer hardware needed to run them locally and in a decentralised way. What lies ahead for maker communities, artistic experimentation and human communities that still value meeting in person — making and unmaking, testing things out and understanding how they work?

We will speak both in person and online with one of the founders and advocates of the Arduino project.

Programme
— Where do these models run?
— What does the future hold for the maker movement?
— What alternatives do we have for decentralising away from the system?

Participants
— David Cuartielles, Arduino co-founder and professor and researcher at Malmö University.
— Andrea Noni, computer gardening and open-source software nursing.
— Miguel Ángel de Heras, head of open-source hardware and software at Hangar’s Interaction Lab.

3. Specialised micromodels and agents
19 November
7–9 pm
Sala Ricson, Hangar
Free admission

AI is here to stay, but does it have to run on someone else’s server? General-purpose models know a little about everything and charge you for the whole package: dependence on a handful of companies, private data leaving your machine — to the point of using these tools as a personal therapist — excessive energy consumption, and an entire supply chain of semiconductor manufacturing and rare earth extraction behind every answer.

There is another way. A small model, trained or distilled for a specific task — transcription, translation, classification or answering questions within a defined domain — can run on a home computer, work offline and, within its area of expertise, perform as well as or better than a much larger model. Give that model tools and let it take action, and you have an agent: local, auditable and under your control.

In this session, we will explain how these micromodels are built, what hardware is actually needed, what can be done with them today, and why they offer a meaningful path towards technological sovereignty, lower energy use and ecological sustainability. Live demonstrations included.

Programme
— Why bigger models are not the answer to everything: dependence, privacy and energy consumption.
— What is a specialised micromodel? Size, tasks and limitations.
— How they are built: your own datasets, fine-tuning and distillation from larger models.
— Home hardware: what can run on a laptop, a mini PC or a Raspberry Pi — and what cannot.
— Local agents: giving models tools to take action, with user control and no data leaks.
— Sovereignty, energy and ecology: the numbers behind each option.
— Live demonstration.
— Discussion: what kind of AI do we want to run at home?

Participants
— Xavi Vinaixa, computer scientist, CTO at Sorensen.ai, researcher and science communicator who discusses ethical technology from a humanist perspective on television and radio programmes.
— Lluís Gómez Bigorda, Ramón y Cajal researcher at the UAB Department of Computer Science and the Computer Vision Center.
— Iván Paz, lecturer at the Faculty of Informatics of Barcelona (FIB), in the Department of Computer Science, and co-founder of the live-coding collective TOPLAP and the Axolot.cat collective.

Supported by Arduino CC and David Cuartielles