Hyperrhiz 30
BabyHex
David Ciccoricco
University of Otago
Mez Breeze
Mez Breeze Design
Marina Cone
University of Otago
Citation: Ciccoricco, David, Mez Breeze and Marina Cone. “BabyHex.” Hyperrhiz: New Media Cultures, no. 30, 2026. doi:10.20415/hyp/030.m01
Abstract: BabyHex is a developmental language learning simulator embodied in a virtual toddler. Or rather, it’s a story masquerading as all that. You are cast as a volunteer training the simulation, pretending to engage BabyHex through basic principles of imitation, where she repeats simple words she hears, or association, where she tries to identify images and pairs image with words together as concept. She also has a motor babble function, which you can initiate and offer praise for syntactical coherence. Inevitably, the system will break down – and descend into story.
Keywords: participatory, narrative, conceptual, poetry, literary, interactive fiction, simulation.
Instructions
Engage BabyHex through:
- IMITATION – Choose simple words to say aloud for her to practice repeating. Ensure your audio is enabled.
- ASSOCIATION – Hold up an image or object for her to identify (be sure to align it with her HexCam).
- GENERATION – Initiate her motor babble sequence; praise emergent syntactical coherence.
- CALIBRATION – Adjust the rate of her learning so that her output follows language use in a consistent and realistic progression.
Artist Statement
BabyHex is a work of electronic literature designed to broaden the discourse of simulation in the humanities beyond reactionary alarmism and moral panic. More specifically, it recognizes simulation as a dominant cultural form in the 21st century – one that calls for greater recognition as a critical, epistemological, and aesthetic mode. BabyHex enacts simulation as critique.
BabyHex is technically a simulation of a simulation, and a pseudo-simulation in terms of its functionality (she does not actually listen through your computer mic or see through your computer camera). Our BabyHex is based on BabyX, an autonomous virtual toddler with a virtual brain and nervous system developed by Mark Sagar and his team in 2013, first at the University of Auckland Bioengineering Institute and then at Soul Machines Ltd. Their BabyX simulation is designed to model human cognitive development and language learning, through interactions with her carers. In addition to state-of-the-art graphics and animation, it uses a custom-built neural network for this purpose. This virtual child is also based on an actual biological child: Soul Machines’ BabyX was created in the likeness of Mark Sagar’s own daughter when she was around two years old. BabyHex critiques BabyX as it reanimates and stories her. (The fact that BabyHex is voiced by my own daughter, who was roughly the same age as Soul Machines’ BabyX at the time, adds further layers of complexity to the project.)
For us, Soul Machines’ BabyX was an opportunity to study an internal contradiction in simulation studies, what we might call “simulation’s temporal paradox.” On the one hand, we demand ever-speedier systems in order to see what their projections will yield; we want the results, and we want them now. On the other hand, we demand modulation and temporal control so that our systems can simulate their target in the most realistic way possible, right down to the temporal flow of the behaviour being modelled. In short, there is a trade off between speed and realistic-ness; two competing measures of accuracy and success – for judging what makes a simulation successful – emerge in turn. For example, we do not want a 100-year seismic activity projection to take 100 years to run; the map would become the territory in the classic Borgesian bind. But perhaps we do want a child language learning simulator that actually runs from age one to age five. We wanted to question how a developmental AI language learning system might adhere to a real-time developmental trajectory. To what extent can or should it? The first answer is objective. The second is a general one: in operational terms, what is possible? The third of course is ethical.
Our BabyHex dramatizes these questions in the mode of aesthetic simulation, under the auspices of a fictional company (Infodome Incorporated), while readers are cast as volunteers training the simulation. Specifically, users pretend to engage BabyHex through basic principles of imitation, where she repeats simple words she hears, or association, where she tries to identify images and pairs image with words together as concept. She also has a motor babble function, which users can initiate and offer praise for syntactical coherence.
Simulation is a matter of selection – deciding, from the target system, what elements get simulated. We endeavored to add in the curious communion of creepiness and beauty in crafting our aesthetic simulation (hence the heavy focus on eyes and faces in both the visuals and the story). In this vein, BabyHex serves as a provocation: perhaps the relentless drive to make AI more and more diverse, more inclusive, and more human is ultimately misguided. What is the point of making AI so much like us that it becomes experientially indistinguishable? Might we do better to keep it as creepy as humanly imaginable?
With regard to the interrogation of temporal scales, in the context of language acquisition, we can either simulate as quickly as humanly (or inhumanly or inhumanely) possible, and with seemingly infinite iterations, in order to reveal how we (both humans and AI) learn. Or we calibrate the output of the simulation so that it faithfully represents the language learning progression of a young mind, replete with those moments of revelation through chance recombination and all too human error.
The overall premise of BabyHex is, however, the breakdown of temporal calibration: this is the story of a toddler whose linguistic precocity runs amok. The system is broken by design, from the start, at which point it breaks into story. BabyHex is about the relationship between a scientist-father and his experiment-daughter (that acknowledges the literary precedents but leaves the hopelessly gendered erotic power fantasies behind). As Mez Breeze describes it in her “Digital T[H]Readings [1994-2025]” collection:
What begins as orderly psycholinguistic conditioning slides into runaway sentience: the toddler races through syntax, questions her own embodiment + hints at fault-lines between paternal scientist + experiment daughter. (archive.mezbreezedesign.com/threadings/)
In this respect, the project also aims to lay bare the interdependence of simulation and narrative as modes of cultural expression in the posthuman era, and frame narrative itself as a vital time management resource both despite and because of the machine-generated language proliferating around us.
- David Ciccoricco (on behalf of the Infodome team)
Credits & Production Notes
Conceptual and narrative design, and text for the Calibration narrative by David Ciccoricco. Imagery, graphical design, and text for the Generation segments by Mez Breeze. Programming and text for the Association bites by Marina Cone. Special thanks to Roa Ciccoricco (voice acting), who was compensated overgenerously with Maltesers, and to Mark Marino, BabyHex’s adoring Godfather.
The philosopher-historian referenced in the opening segment is Jean-Pierre Dupuy. In his 2000 book, The Mechanization of the Mind, Dupuy writes about the origins of cybernetics at the so-called Macy Conferences. Here’s the full passage, which indexes a formative moment in the intellectual history of simulation:
Reality had become a mere means to higher end: model making. […] But more often, and then usually in an unreflective way, object and model exhibited an awkward tendency during these discussions to constantly switch roles. The ambivalence of the very word model […] was thus revealed: the model may be either that which imitates or that which is, or deserves to be, imitated.
In segment IV, we cite a line from Siobhan O’Flynn’s wonderful “Ten Maxims for Guiding Shy, Reluctant, or Unruly AI” (2022).
Other aesthetic inputs include Erik Loyer’s Lair of the Marrow Monkeys and Chroma and in particular his wildly inventive “mnemonic membrane.” And in terms of its visual allure, there is the me[z]merizing [Por]TrAIts (2022) by Mez Breeze. The project also calls back to Breeze’s The MALFI Trials (1996) with “Infodome,” which is derived directly from The MALFI Trials Scientific Research Organisation.
Disclaimer: this is a work of fiction. Any resemblance to actual persons, living or dead, virtual or actual, is purely incidental.
AI disclosure: images were generated in collaboration with MidJourney; version 2 of BabyHex used Claude AI to format the interface for mobile devices.