Latent Intimacies

Latent intimacies logo

Latent Intimacies propone exploraciones experimentales y abiertas de posibles formas de intimidad entre humanos y máquinas, a través de modelos lingüísticos de código abierto. Se trata de una conversación performativa entre un humano y un objeto agente de voz, en la que ambas entidades se exponen mutuamente. El proyecto busca momentos deseables de vulnerabilidad en nuestras interacciones cotidianas con la IA, demostrando lo difícil que es para las máquinas hablar con los humanos, y cómo deberíamos apreciar estos momentos.

Colaboradores seleccionados: Valeria Castillo, Moisés Ramírez, Janire Goikoetxea, Paulina Casas Lias y Justin Pickard.


  • Visuals?
  • Installation / Usage?
  • License?

Latent intimacies

[Title in the correct typeface as PNG]

What is a “latent intimacy”? Where do we catch glimpses of intimacy in today’s human-machine interactions? And what else could intimacy be? A latent intimacy is, for us, a gesture to AI’s potential to propose alternative forms of intimacy, echoing semantic accidents captured in earlier, more primitive human-machine interactions. 

In Latent Intimacies, we used interviews, observation, and prototyping to produce three Études Brut, “studies in rawness”, refracting what we thought we knew about intimacy through three specific speech synthesis and customised generative language models. The industrial form of these studies reflects neither purpose nor clear functional application, but they are ‘not mere accidents, waypoints on the way to something else.’ (Bogost, 2012: 93) They are here, they work, and therefore they are possible. 

What else could AI be? We approached the language model as a minor technology, detached from scalar norms of efficiency and optimisation. (Anderson and Cox, 2023; Fartan, 2023) These raw studies are small-scale, contrarian technologies, standing apart from the dominant narratives of technology and innovation. Our three prototypes have little commercial value, but they provide alternative routes into the digital world, promising slower, more considered connections than those currently offered by cloud-based AI platforms.


Intimacy through vulnerability, the capacity to be damaged or harmed.

[image of demolition]

Our first study, Demolition, is site-specific, a block of concrete encapsulating half a dozen WhatsApp chats, our online interactions with those friends and family members who agreed to take part. We can ask anything we want, and the voice agent will try to give the best possible answer, by sampling text from this tiny, intimate dataset. It will then ask us to judge if its answer is relevant.

Emotions become embedded in latent space. Residue and resonance of past performances of generosity, care, and intimacy echo in the chambers of everyday life.

“I invite you to take a part of me.”

Using automation to erase unfavourable memories of past experiences liberates us from the fangs of yesterday. Confronting textual snippets of the past, stripped from their original context, can supply catharsis. At the same time, the interaction allows users to gauge and adjust their level of engagement, during the difficult process of mourning and moving on. 

Exit. Exist. Enter. Erase. Repeat.

Red Book

Intimacy through protocol, a codified series of guidelines that facilitate orderly interaction, through a shared understanding of the rules of engagement. 

[imagE of red book]

The Red Book is inspired by a mystical social game played by Mexican children. Blending chance, prophecy, and divination, playing the game requires a set of protocols for activation and engagement, while stressing the importance of disengagement. Leaving without permission may have undesirable consequences.


The future is uncertain. People experience uncertainty as a source of discomfort and unease. It can leave them feeling vulnerable and precarious, threatening what they know of their place in the world. Given this, who do people turn to for answers, advice, guidance?


The Red Book is a curated language model, trained on texts (often books) contributed by its users. A group looking for answers designates a spokesperson to address the Book, much as one would a voice-activated digital assistant. Once activated, the speaker delivers prompts and questions, interpreting the needs and interests of others in the group. Some questions linger, unanswered. As the interaction draws to a close, the speaker seeks permission to leave. 

"Your presence is no longer required."


Intimacy through latency, the delay between a cause and its effect.

[image of cats]

This study intertwines the daily lives of cats, their caretakers, and the community. A tiny AI narrator, Cats offers a unique perspective on the world. Translating snapshots from a cat-worn camera into basic spoken texts, the AI emulates a cat. It sticks around, but sets its own pace, providing pleasure, ungoverned, on its own terms.

Cats takes an hour to load, and roughly 40 minutes to compute a text describing a randomly taken image. The AI outputs around ten images a day; ten short texts spoken aloud, captioning everyday moments seen through a cat’s eyes. We seek connection, closeness, and this device forces us to wait, tracking the low-maintenance intimacies of pet ownership. Cats have latency. Durations of absence are experienced when you go about your ordinary everyday life outside the home, leaving your furry friend alone during business hours. 

“I saw a man sleeping on a bench in a grassy park.”

Other technologies promise to monitor our pets, but plug-and-play surveillance casts them as captives, not peers. Giving your cat a voice introduces an alternative context. Cats casts the language model as a wearable minor technology. Its technical limitations demand patience. Rather than inundating users with a barrage of texts and images, Cats yawns and stretches, relaxing into the interval between outputs.



[Link to moises’ video of the process?]

Vytas Jankauskas


Paulina Casas


Valeria Castillo


Janire Goikoetxea


Justin Pickard


Moises Ramirez



‘Latent Intimacies’ was produced as part of Medialab Matadero’s LAB#03 Synthetic Minds Collaborative Prototyping Lab, which ran from January 24–February 10 2024. Our heartfelt thanks to Alan Warburton, Patrick Harrison, Irma Pužauskaitė, Laura Forlano, Tim Cowlishaw, Vasundhra Dahiya, Rohit Gupta, David Hayward, Valdis Silins, Charlie Loyd, Michelle Kasprzak, Marbles (a cat), and Murphy (another cat).


Andersen, C. U., & Cox, G. (2023). Toward a Minor Tech. A Peer-Reviewed Journal About, 12(1), 5–9.

Bogost, I. (2012). Alien Phenomenology, or, What It’s Like to Be a Thing. University of Minnesota Press.

Fartan, T. S. (2023). Rendering post-anthropocentric visions: Worlding as a practice of resistance. A Peer-Reviewed Journal About, 12(1), 43–60.

2 colaborators







Project status
Joint valuation