The liminal epistemology of ChatGPT — a radical design perspectiveRadically new ways to engage with fictional knowledge.The traditional epistemological categories — truth, belief, inference, and justification — which we normally use to define how knowledge is recognized and validated, are ill-suited to describe ChatGPT. It does not possess beliefs or human grounding in real world experience, nor does it verify claims. Instead, it generates plausible continuations of linguistic sequences. It doesn’t actually understand the meaning of what it says — it only knows how words and sentences typically fit together based on patterns in the text it was trained on. Yet, knowledge nonetheless emerges in our interactions with it.This philosophical distinction matters greatly for how we design our interactions with the model. Do we make it quick and easy to retrieve ‘definitive’ answers, or do we make sure that responses come caveated with all its limitations. Neither approach seems to take advantage of the unique possibilities that large language models offer us already.The paradox is more constructively resolved if we recognize that ChatGPT operates not as a knower, but as a participant in a liminal space: a threshold zone where human intentions, algorithmic outputs, and cultural frameworks converge. Relational and phenomenological thinkers such as Maurice Merleau-Ponty and Eduard Glissand emphasize how knowledge emerges not from detached observation, but through embodied, situated engagement with the world. Extending this view of knowledge as emerging from active participation rather than fixed positions, Victor Turner’s seminal work on ritual and transformation further underlines how structures and meanings are provisionally re-formed in liminal spaces. In this light, ChatGPT occupies a “betwixt and between” status — not author, not tool, but a generative threshold for synthesis and novel emergence. Liminality is not incidental to ChatGPT; it is the very space in which epistemic value is co-constructed.One might of course argue that ChatGPT is ‘grounded’, just in a different kind of reality: a probabilistic, textual environment shaped by patterns rather than perception. And philosophers like William James, Nelson Goodman, Ronald Giere, Karen Barad, and Donna Haraway have all foregrounded that human knowledge, too, is always shaped by perspective. From this pluralistic view, all knowing is situated — there is no “view from nowhere.” Whether ChatGPT lacks sentience or simply occupies a different ontological frame, the key point remains: knowledge here is co-performed. It emerges not from what either side contains, but from what takes place between them. That is the condition that defines this epistemology as liminal.Exploration not AnswersConsidering ChatGPT as a liminal actor shifts the user’s stance. Rather than seeking definitive answers, the user engages the model as a co-inquirer, a reflector and suggestive pattern-maker for our framed explorations. This reorientation enables a more critical, playful, and pluralistic engagement. We no longer ask “What does it know?” but “What does it let us see?” Epistemology becomes performative: we are not extracting information from ChatGPT, we are staging inquiries with it. This changes the nature of trust, authority, and responsibility in our interactions. Instead of treating ChatGPT as a knowledge endpoint and source for answers, we treat it as a site of emergence for curious explorations.The heartfelt quest for real alternatives to the linear, transactional logic of product relationships should be unsurprising coming from those whose work explores the deep, messy tilth of liminality. Regardless, all of us are well served to create room in our mundane commercial interactions: to seek perspective, openness and dialogue. This is even more true for ubiquitous technology, where the offered extensions to our own reach and fixed to our enthusiastically sub-par productivity, risk instead making us extensions of the machine that offered it. We become the product in service of its needs, prodded and changed by unrelenting design to make us better fit with corporate needs for predictability, efficiency, profitability and scale — all things counter to the sacred uniqueness of the human spirit.Liminality is inherently well suited to guard against this: to create an in-between space that suspends ordinary relationships and structure through metaphor and paradox in search for new perspectives that can then transform the linearity that it came from. Liminality doesn’t change reality. It changes how we assign meaning to what is, and our ideas of what might be. It changes our vantage-point. Aristotle suggested that all stories need to be both inevitable and surprising, at the same time. This paradox is what liminality thrives upon. The surprise is always a change in our own vantage point.A Change in AttitudeTo embrace ChatGPT’s liminality, users themselves can shift from instrumental to exploratory prompts. Rather than asking it for definitions or summaries, we can engage it in:Speculation, metaphorical framing, or narrative experimentation.Ask questions that explore ambiguity, juxtaposition or multiplicity rather than seek definitive answers.Use speculative, recursive, or exploratory languageInvite the model into dialogic, not didactic, space.Users can also foreground their own uncertainty, prompting the model to co-wander rather than solve. In this way, epistemic authority is dispersed, not abandoned: meaning becomes a collaborative construction.But the burden of transformation cannot fall solely on the user. If we are to treat ChatGPT as a liminal partner, we must also question what responsibilities fall on the system and its designers. To leave the model as-is while demanding interpretive flexibility from users is to replicate the very asymmetry that liminality seeks to dissolve.We must ask: how can ChatGPT be made to participate more actively and generatively in this suspended liminal space? And remember, the epistemic dialogue driven LLMs are already liminal: we know with certainty that it doesn’t actually have answers or real knowledge. As it today asks us to suspend our own disbelief, and to treat it ‘as if it knows’, a liminal approach would be more sincere and indirectly, highlight our agency and responsibility in framing questions. It would overtly redirect the interpretive responsibility back to the user where it — after all — always resided, no matter the high-tech opportunistic cavalierity fumbling for gold in the existential landscape it has created.So, what would a more clearly articulated liminal epistemology mean in terms of product changes to ChatGTP? How might participation in our inquiries be better at both moving the exploration forward in relevant ways, while also encouraging novel dialogue through unrivaled access to data and ways to structure it anew and in the moment. As always in liminality, a good question will unfailingly trump a correct answer.A system model more attuned to liminal dialogue might:Privilege metaphor over fact: Rather than defaulting to definitional responses, ChatGPT could be prompted or trained to respond metaphorically when cues indicate symbolic or philosophical exploration.Signal discursive echoes, not sources: Instead of citing definitive authorities, ChatGPT could gesture toward interpretive traditions, preserving openness while offering theoretical grounding.Preserve ambiguity and contradiction as feature, not failure: The model could be trained to resist tidy synthesis when ambiguity is thematically appropriate. This could involve rhetorical hesitation, competing images, alternative view points, or posing recursive questions to keep conclusions open.Mirror epistemic risk: ChatGPT could take on the user’s mode of questioning — responding to speculative queries with speculative tone, to paradox with paradox, to poetry with asymmetry — amplifying the liminality instead of resolving it.All of the above is very possible to do. A liminal perspective would also elevate the current and anxious AI debates beyond the stubbornly false dichotomy of Martin Buber (I-It vs. I-Thou) — is it sentient, is it a person, a tool etc? — and set the table for a richer and more future forward debate. The question isn’t what it is, but what it might be.A liminal reframing aligns with a post-representational view of epistemology. Karen Barad’s agential realism argues that knowledge emerges through intra-action — the mutual entanglement of agents and phenomena — rather than through detached representation. This perspective strengthens the view of ChatGPT as epistemologically liminal: meaning does not reside inside the model or the user alone but arises within the dynamic, relational space between them.[1] Derrida’s notion of différance reminds us that meaning is always deferred, never present in full — a logic mirrored in ChatGPT’s predictive architecture.[2] Heidegger’s Gelassenheit, or “releasement,” invites an attitude of openness toward technological being — not mastery, but co-habitation.[3] These thinkers suggest that rather than resisting AI’s epistemological ambiguity, we might lean into it. Not to solve meaning, but to dwell in, and more fully mine its unfolding. A very healthy liminal attitude indeed.Looking forward, this design perspective should not be applied too narrowly. Consider not only the liminal space for dialogue that we have today, but also completely new contexts and means for engagement. We must question the singular and monotheistic anthropomorphism of a mighty power up in the cloud. The dialogue could be tied to the changing conditions of a particular object or product directly — adding real world embodiment to the context. It might be a live situation, relationships or projects as conditional framing. Or a different cadence, a certain geographical or domestic location. Dialogue could take form through fictional characters — not a tool, not a person, not a thing, but relational figures we come to know and learn with over time. Here, the epistemic role would become narrative and affective rather than informational, with the ability to have tremendous and long lasting impressions on our lives. After all, our current relationship with AI is already fictional; failing to recognize this may have as much consequence as what we choose to invent._____________________Johan Liedgren is the founder of The Liminal Circle, an international think tank for applied liminality. An award-winning film-director, writer and strategy consultant working with media and technology companies on liminal product design, Liedgren is based in Seattle, Stockholm and Milan. www.liminalcircle.com / http://www.liedgren.com / https://medium.com/@johan_liedgren[1] Barad, Karen. Meeting the Universe Halfway: Quantum Physics and the Entanglement of Matter and Meaning. Duke University Press, 2007.[2] Derrida, Jacques. Of Grammatology. Johns Hopkins University Press, 1976.[3] Heidegger, Martin. Discourse on Thinking. Harper & Row, 1966.[4] Nagel, Thomas. The View from Nowhere. Oxford University Press, 1986.The liminal epistemology of ChatGPT — a radical design perspective was originally published in UX Collective on Medium, where people are continuing the conversation by highlighting and responding to this story.