Offerta di lavoro dal 5 giugno 2026
Symposium
Venice, Ca’Foscari and Biennale
28–29 September, 2026
“The world is its own best model”, wrote roboticist Rodney Brookes – “always exactly up to date and complete in every detail”. This was the central claim of Nouvelle AI, which rejected the idea that autonomous systems should maintain and update an explicit representation of their environment – a world model, a term borrowed from human psychology. Thirty years later, world models have become a crucial topic in artificial intelligence research, as well as many other disciplines in conversation with it (e.g., computational neuroscience and Earth system science).
Whether through OpenAI’s video generation systems described as ‘world simulators’, Google’s Genie synthetic environments built to train agents from the visual language of videogames, or neuroscientific understandings of brains as encoding internal parameters of an outside world, world models seem to be the apotheosis of the drive in machine learning to absorb, and ultimately govern, the entire world. What unites these projects is therefore not a single technical architecture, nor even a single epistemology of modelling, but a shared pressure placed on the concept of the model itself. In each case, “world model” names a different relation between representation, prediction, compression, and control to produce differentiable simulations of the world, in and through which an agent might learn and be trained. The apparent coherence of the phrase conceals an epistemic and technical instability: sometimes the world is a physical environment, sometimes a video stream, sometimes a latent state-space, sometimes an affordance structure for action. World models thus do not simply extend the old scientific ambition to render the world legible. They expose a crisis in the very concept of modelling: the more total their claim to world-ness becomes, the less clear it is what kind of world, and what kind of model, is actually at stake.
The symposium Dysphoria Mundi seeks to engage with the promissory ambitions of world models, and the estrangement they produce. Models built to understand, know, and inhabit the world often do not fit the beings that should be living through them. Following Preciado, this misfit presents itself as a dysphoria. Rather than interpreting such dysphoria as mere errors to be corrected, or a condition to either accept or resist, Dysphoria Mundi asks what this very misfitting reveals about the structural exclusions operated by these models, the dimensions they 'shallow out', and the possibilities for praxis their opacity, paradoxically, opens up.
Dysphoria Mundi seeks to explore such limits and possibilities by thinking through four intersecting strands of inquiry concerned with world models, and world modelling more broadly. First, we explore the technical and epistemic genealogies of world models within the history of artificial intelligence research. Secondly, we engage with artistic and experimental practices that make said architectures visible, and, crucially, contestable. We then discuss the confluence of computation and cognition, which sees brains and machines as world-modelling devices, asking what kinds of worlds are being worlded through such operative schematics, and what may lie beyond. Fourth, we examine planetary modelling, and the paradoxes of a form of knowledge that burns what it is trying to know, and, allegedly, save.
We thus seek short presentations that relate to one (or more) of these themes. The event will be split across 2 days and will take place both at Ca Foscari and the Venice Biennale.
Conference Strands
I. From Representation to Simulation: The Shifting Epistemics of World Modelling
World Models have a contested history in AI research. In classical AI and cognitive science, the term named an internal representation through which an agent could model its environment. For critics of symbolic and representational AI from the tradition of “Nouvelle AI”, this was precisely the problem. Phil Agre, who used videogames as experimental environments for AI in his own research, thus described world models as the “epitome of mentalism”.
This strand asks how that older dispute has returned in an apparently logically reversed form. The classical problem was whether an agent needed an internal model of an external world in order to act within it. The contemporary problem might be, instead, whether the agent itself can be produced through a modelled world: a generative, differentiable simulation, most often visual, in which behaviour is trained before it is transferred elsewhere. In this sense, world models no longer name only representations held by agents, but environments that generate agents; not mental pictures of the world, but entire worlds told through pictures (Heidegger’s Weltbild), through which capacities for perception, prediction, and action are formed.
Through this strand, we ask and examine what distinguishes a world model from a predictive model, a generative video model, or even a simulation, and when this shift becomes a claim to understanding, and perhaps a new mode of modelling. The session seeks to conceptualise this shift from representation to simulation, which points to a deeper instability in the concept of modelling itself within AI research.
II. Brains that Model, Models that World: Neural and Algorithmic Frontiers
This strand engages with current approaches that view brains and machines as devices that build and maintain internal, generative models of an alleged outside. From predictive processing and the Bayesian brain to model/energy-based reinforcement and unsupervised learning architectures, these approaches construct models that organise perception, action, inference and expectation in ways that presuppose particular, and particular kinds of, subjects, environments, and modes of being.
We ask what kinds of worlds are being worlded through these theories and models of mind and life, and at the expense of what. On the one hand, this strand seeks to critically examine and historicise claims of brains and machines as world-modelling devices, as well as attending to the modes of being that such frameworks foreground, and the ones which come to be disavowed and pathologised – including forms of perception, affect, and cognition that deviate from the norm these models both assume and reproduce. On the other hand, we aim to highlight precisely the sites through which otherwise modes of worlding may come about, not as a refusal of such paradigms, but as an engagement with their dysphoric consequences.
Contributions may thus engage with: critiques of computational neuroscience; reframings of cognitivist models of feeling; discussions of non-Western world-making practices; critical analyses of contemporary machine learning world models architectures.
III. Dysphoric Imaginaries: Model Contestation In and Through Artistic Practice
The complexity of most models and their architectures in machine learning make them difficult to represent, understand and critically address. In this regard, artists and artistic practices have played a crucial role in making such opacity legible since the emergence of complex computational processes. Through experimental and alternative uses of these systems, artistic practices have challenged conventional understandings of latent space and artificial representations, while disrupting the normative and intended uses of such models to question contemporary concerns on visibility, classification, and social norms. Crucial examples of such disruptive encounters with modelling techniques come, for instance, from Chatonsky’s work on autonomous imagination, and Anadol’s research on latent spaces and machinic memory aesthetics.
As the field of research around World Models (WMs) remains highly experimental, artistic practices offer critical perspective on their possibilities, limitations, and ways of perceiving, interpreting, and simulating realities, and the worlds they presuppose and foreclose. We are particularly interested in artistic practices that engage with these models not only as tools, but also as collaborators and catalysts of specific visions of the world, aesthetics and forms of artificial imagination, as well as tools for contestation. Through this strand, we aim to bring together artists whose work challenges dominant narratives surrounding WMs, and whose experimentation opens critical inquiries into representation, perception, and agency within increasingly sophisticated artificial systems.
IV. Modelling the World, Burning the World: Planetary Computation, Control, and Collapse
This strand looks at world models as techniques for knowing, predicting and governing planetary life (e.g., Earth System Models, economic and epidemiological simulations, insurance and risk prediction) in the history of ecology and geosciences. It foregrounds both their technical dimensions — complexity, the interplay of historicity and lawlikeness, granularity — and their epistemological foundations, rooted in colonial and extractive histories, while also attending to socio-economic and radically emancipatory counter-traditions within the field. It further highlights their normativity, grounded in the imperative of prediction that defines Earth System science; the paradox of expending energy and resources to model the very systems being destroyed; and the tendency to defer concrete intervention through implementation models that presuppose the reversibility of climate change through technological means. Taken together, these threads invite a reflexive, second-order critique: not just of what models of the world represent, and of which image of the planet is represented, but of the practice and politics embedded in modelling itself. It welcomes work on radical epistemologies of Earth System science, decolonial and Indigenous critiques of planetary modelling, as well as radical and anti-hegemonic alternatives to the ESS paradigm; operational images of the Earth (remote sensing, digital twins); and the politics of anticipatory governance, resilience, and adaptation.
To participate, please send your expression of interest (including a short bio and a paper proposal or artistic portfolio) by filling in this form: Google Docs
Deadline for submissions of abstracts is the 22nd of June, 2026. We aim to confirm acceptance by mid-July.
Organised by AI Models (Ca Foscari), Cambridge Digital Humanities (University of Cambridge), and Machine Visual Culture (Bibliotheca Hertziana - Max Planck Institute for Art History), as part of Venice Biennale Sessions 2026.