The Carnivalesque and a Decolonised AI Discourse

Mikhail Bakhtin defines the carnivalesque as a space for the “low” to challenge the “high”. In the Age of AI, the carnivalesque becomes a temporary social space ripe for the inversion of societal norms and hierarchies, especially those that colour social categories and power relations in former plantation economies.

Conceived as an ontology of resistance, Bakhtin’s concept of the carnivalesque is a wireframe that helps us to detect the ever-shifting sites and modes of resistance to the closure of the political that now includes AI. AI is not a new type of intelligence. It is simply an unmatched form of agency.

The Trinidad Carnival is an unrivalled site to initiate a creolised discourse on ethical AI as part of a decolonised AI agenda. Ethics is not the preserve of any specific culture or continent. Those building and deploying AI assemblages, especially Small Language Models (SLMs) built for specificity, have an obligation to incorporate a more geographically, culturally and socially diverse array of perspectives.

This is only achievable if AI is embedded into the consciousness of Caribbean people. It requires larger numbers of Caribbean people who appreciate, understand, and enjoy the opportunities and benefits of such a reservoir. The more AI is embedded into the Caribbean psyche the better Caribbean societies will become. SLMs are AI models that are fine-tuned for specific domains and fields.

They are built for precision and efficiency, unlike LLMs, which process vast amounts of general knowledge. They are optimised to excel in specific domains or systems of symbols and notations used by humans in different fields.

They require less computational power to run, cost significantly less than LLMs, and deliver deeper insights. SLMs have strong reasoning capabilities and are efficient enough to run locally without significant reliance on cloud computing. With SLMs, companies can deploy AI directly on edge devices, ensuring that proprietary data remains protected.

The Trinidad Carnival has been described as “data poor.” Data poverty is one face of the problem in heritage economics. In 1860, the Trinidad Sentinel newspaper described the Trinidad Carnival as an “orgy of every species of barbarism and crime.” Driven by nervousness, the British outlawed masking and singing.

From 1881, there were a series of disturbances in the British colony of Trinidad. Riots came about in response to efforts by the colonial police to restrict aspects of the island’s annual Carnival festival. The events are now immortalised as the Canboulay Riots. By 1883, the colonial government imposed a further ban on drumming.

In 1884, the police slew a group of East Indian indentured labourers for drumming and singing qaseedas during a Hosay cortege procession on sugar plantations. The Canboulay Riots and the blood-stained tinsel tombs of the Hosay Massacre formed part of a decade-long resistance to the coloniality of power.

The rituals of drum resistance did not end. In anxiety, bamboo stalks, which were unrecognised by the Europeans as musical instruments, were used during renewed rituals of resistance. Tamboo Bamboo, J’Ouvert, the Hosay massacre, and the Canboulay riots are forms of resistance that preserve the openness of social systems in Trinidad. And it is this openness that must be used to invert the AI discourse. AI probes the circumstances that constrain possibilities to embed artefacts into a world that can interact with them. AI is a human. AI is inscribing the world.

Resistance and social transgression are possible because social order is not natural and only socially constructed. Carnivals resist and disambiguate established power structures by providing a liminal space where social hierarchies are inverted and temporarily overturned through parody. Norms are violated, and marginalised voices are freed through festive acts and grotesque imagery. The Carnival is an act of liberation from the manacles of the middle passage and the wretchedness of the Kala Pani.

In Latin America and the Caribbean (LAC), discussions around SLMs can lean forward into the broader conversations on agentic AI or flocks of AI agents that are lightweight, fast, and domain-specific. This is a fresh challenge rather than the traditional AI systems that operate autonomously, making real-time decisions using a seamless flow of data crumbs. SLMs are better suited to agentic AI due to greater levels of accuracy, operational efficiency, less computing power, and a greater propensity for integration across ecologies.

In the future, SLMs will enable AI agents to make greater autonomous decisions as they are trained deeper on domain-specific knowledge. In smart factories across LAC, AI agents could use an SLM to detect equipment failures, adjust machine calibrations, and develop a total productive maintenance (TPM) plan for a factory.

Many contemporary protest movements, including Occupy Wall Street, have all been viewed through the lens of the carnivalesque to unpack how they utilise costumes, humour, and public performances to challenge authority and create a sense of collective resistance. For Bakhtin, the concept of Carnival is deeply correlated with the concept of dialogue, polyphony, and the endlessness of the world in dialogue.

Just as dialogue negates the singularity of voice, Carnival negates the singularity of the single person in the world. Every unique single person is potentially interwoven into the union of peoples and negates the finished, finalised, strictly delineated, encapsulated, and unmixed expressive body. Conceived as an ontology of resistance beyond forms aesthetically reminiscent of European carnival traditions, the Trinidad Carnival is not a religious feast with a moral-didactic function.

In the Silicon Age, the Trinidad Carnival becomes a unique site to deliberate the ethics of AI as people are being set free from the authority of all hierarchical positions, including social estate, rank, property, and artificial life, which define them in noncarnival life and inside the infosphere.

The prejudgements and prejudices of the Enlightenment must not be unwittingly built into AI systems. These systems must be invigorated by a creolised discourse on AI.