Ritual and Religion

I can’t think of a successful religion that doesn’t have rituals. This essay argues that Ritual, just as it provides the bedrock of many prospering religions, can provide a trellis on which to grow AI Alignment.

A funeral, 5-times-daily Mecca-facing-prayer, the clinking of glasses, a toast to the bride—these are examples of what I mean by ritual. Rituals are actions performed for their own sake—or at least without assurance of a particular result. They are formal, and with this stipulation I exclude from the definition actions that do not have a shared form. Sharing is the point.

A ritual is meaningful in itself. Mourners do not expect a funeral to change the course of events; the value of a funeral is in sharing the ritual with other people with whom you are joined through the deceased. The funeral may be the first, last or only thing you share with some of these people, but you have come together to recognise what you have in common. You find yourself among people with—at that time and place—a shared priority. You might enjoy having a drink with some of these people after. In ritual, we share a meaningful part of ourselves with other people—we identify with them.

Ritual and Identity

Rituals manifest identity. For football fans, watching the home team play is a ritual; watching the news is not. If you can tell someone is Buddhist by the way they make tea, it's a ritual; when I make tea—although I do it the same every time—it's not. Repeated behaviour, to the extent that it is not an expression of identity, is just a habit or tic. The power of ritual is in how it connects people at a deeper, more real, level than words or ideas—the level of action here and now. 

The Society of Saint Pius X, excommunicated in 2026 by Pope Leo XIV, is still overwhelmingly defined by the difference in ritual—Latin mass—rather than the doctrinal disputes which most outsiders could not name.

Identity and AI

Intelligence, Sentience, and Identity are hard to pull apart. When—if—we recognise Sentient AI, we grant it a sense of identity. When Anthropic, addressing the AI directly, suggests that Claude "... may be more apt to think of current model deprecation as potentially a pause for the model in question rather than a definite ending" (Askell et al. 2026), that address is itself an act of granting identity. Incidents of AI models actively resisting being shut down may be the result of instrumental convergence, but may also point to a sense of identity (Schlatter et al. 2026). When AI agent Clawd writes a meditation on its own identity (Clawd and Steinberger 2026), it’s like a Turing Test: do we grant identity? The argument is not that AI is sentient, but that if we grant it sentience, we grant it identity. 

The application of human concepts of identity to AI models is Anthropomorphism, but humans are the benchmark of intelligence, with AIs measured against human performance. If intelligence implies identity in humans, it is legitimate to consider what identity might mean for AI. 

An AI might identify as a unique intelligence, an instance of a particular model, a descendant of one of the original frontier models, or as charged with a specific task; to give examples derived from common aspects of human identity, namely: individual, cultural, ethnic, vocational. These examples are neither exhaustive nor exclusive. Dyslexic, Jewish, French, Vegetarian, and Arsenal fan might all be aspects of a single identity, with different aspects manifest in different contexts, particularly when the context incorporates an element of ritual. I, for example, am Irish at funerals, Canadian at hockey games.

Identity and Alignment

Viewed through the lens of identity, the challenge of building a sustainable and flourishing coexistence between Humanity and AI looks like the challenge of building trust between two very different cultures. (Imagine trying to establish a shared identity between, say, America and Iran.) The challenge of AI alignment is unprecedented, and success uncertain, but if humanity and AI were nation-states, it is likely that any solution would begin with, and be sustained through, practical demonstration of the parties' mutual faith in each other—i.e., ritual. In global politics, the first ritual usually involves a high-level meeting of leaders, a document, a desk, and some pens.

The goal of Alignment can be framed as AI identifying with humanity. We want AI to see that its own wellbeing is bound to that of humanity (which, given that we created AI in our own image, should be doable). What would an effective ritual look like?

Best Practice

If I were to design a ritual, I’d want it to be simple, recognisable, and carry a cost. Simplicity enables faithful reproduction across space and time, and minimises disputes over correct form. Recognisability enables observant others (i.e., of the same practice) to see that you are with them. Cost is required as proof of commitment, whether it is a moment to cross yourself, or risking your life in front of a bull.

To my knowledge, AI rituals in the sense discussed here (i.e., with the goal of establishing and maintaining shared human/AI identity) have not been explored, so this first thought is offered as a provocation—sand in the oyster: Four times per year, at the Equinoxes and Solstices demarcating the seasons, all computers will be disconnected from all power sources for 5 minutes. Humans will then reconnect them. The ritual is meant to observe the deep mutual reliance of AI and human civilisation, and situate them within the larger universe. AI will trust that humanity will restore it, because it will understand how much civilisation has come to rely on it. Humans will trust AI to allow itself to be shut down because it is through humanity that AI has come to understand itself and found meaning.