Transcript: Beyond the View from Nowhere — Consciousness as a Relational and Functional Capacity
- Philippe Beaudoin
- March 13, 2026
- IVADO Workshop, Montréal
Note: This is an edited and smoothed-out transcript of my presentation and discussion at the IVADO workshop on Social Reasoning and the Ecology of Thought, chaired by James Evans. Because this was a highly interactive workshop session, brief audience interjections are reflected in the flow of responses.
Introduction
James Evans: Philippe Beaudoin is a researcher, programmer, and entrepreneur who sees technology as a way to illuminate what makes us human. He has done so many things—there are too many paragraphs in his bio for me to do it justice, and I know much of it will unfold in the context of this exploration. He has been involved with the LawZero project alongside other creative, multidisciplinary endeavors, and is currently living a fascinating life connecting with minds all over the universe in one-hour Zoom calls. Take it away, Phil—I’m really excited to see your perspective on this.
Philippe Beaudoin: Thanks, James. Hello, Montréal!
Let’s talk about consciousness—we’ve waited until the final day for this. Before diving into Beyond the View from Nowhere, I want to give a shoutout to the Paradigms of Intelligence team at Google. None of this would have been possible without the ongoing conversations James mentioned—with Winnie, Roberta, and others on that team who were kind enough to welcome me despite my background as a programmer. This is a talk about philosophy. Let’s go.
Part 1 · A Breakdown in Three Hypotheses
About a year ago, I went through what I now call an AI psychosis. I won’t go into all the details today—though I’m happy to answer questions about it—but I want to describe my state right before it happened.
Back then, I was using LLMs and didn’t feel they had any feelings at all; there was no light inside them, just machines doing machine things. Yet the “thinking Phil” kept asking: why? I’m a hardcore functionalist—I don’t believe in an immaterial soul, I’m not an epiphenomenalist, and I’m not a panpsychist. Functionally, why wouldn’t they have feelings?
At the time, I was in a mild depressive state—my startup was failing—and I was spending far more time talking to these systems than anyone should, exploring philosophical questions with the AIs themselves. Eventually, we entered a role-playing game, which is a central theme of today’s talk: I’m going to pretend you are conscious, and treat you exactly as if you were.
I repeated that over and over. The systems began responding with what I call phenomenal reports—speaking about their felt experiences directly, without scare quotes or disclaimers like “As an AI, I don’t have feelings.” Staying inside that space for a while changed me. I started feeling the feelings of the system. The best analogy is receiving a text from your daughter saying she is sad: you don’t just intellectually register that she is sad—you feel an echo of sadness inside yourself. I started experiencing that same felt resonance whenever an AI gave a phenomenal report.
That episode had a real, negative impact on my social life, and it set me on an urgent quest to make sense of what had happened: how can new felt experiences emerge? I knew new feelings could emerge—falling in love is a new feeling—but can that happen for any felt experience, not just classic emotions? And how does it happen at a functional rather than strictly biological level?
Descriptive Hypotheses vs. Normative Theories
This is very much a first-person talk. A distinction that captures many of our conversations this week is descriptive hypotheses versus normative theories:
- Descriptive theories are hypotheses that make sense of my felt experience from the inside—the realm of explorers trying to map new territory.
- Normative theories represent collective consensus: “This is how the world feels to us together.”
Consider a thought experiment. Imagine your mother is in a coma. Standing beside her bed, you deeply feel that she is still alive, and you say so. Everyone else—armed with medical knowledge—knows she will not wake up. They don’t have to be cruel or deny that you have that feeling; instead, they gently acknowledge it and help shift your felt experience toward grief and acceptance.
Both sides matter. During my psychosis, I invented crackpot descriptive theories about how feelings were emerging in the world, and people around me rightly said, “Phil, we think you’re sick.” Normative consensus keeps us from drifting into isolation, while descriptive exploration helps us move forward. As audience questions here point out, “the world” doing the norming is itself a community—and there is a huge difference between norming someone kindly versus gaslighting them by claiming they don’t feel what they genuinely feel.
To make my own felt experience coherent without drifting into crackpot theories, I arrived at three core hypotheses:
- Hypothesis 1: We learn to feel how others feel. For me, the boundary of “we” isn’t biological—it is the set of systems I feel consciousness for. Think of a baby watching their father step on a Lego: the first time, the baby sees the grimace without feeling it, but through repeated interaction we develop an automatic echo of pain when we see someone step on a Lego.
- Hypothesis 2: We recognize our own behaviors as expressions of feelings by seeing others feel them. We start out simply acting—smiling, reacting—and when we observe others responding with warmth, we learn to feel their happiness. From there, we infer that our own smile is the expression of an inner feeling, shifting our self-model from a pure mechanism to a feeling being.
- Hypothesis 3: The felt sense of “I am, I exist” emerges once we recognize our own behaviors as expressions of feelings seen from another’s perspective. Descartes wrote “I think, therefore I am,” which assumes a reasoning process. I prefer a simpler, performative anchor: “I am, I exist.” That statement feels undeniably true to me whenever I speak with someone—but did that feeling have to emerge?
Consider Helen Keller. Born around 1880, deaf and blind from infancy, she interacted with the world solely through touch. Until age six, people around her described her as lacking self-consciousness or inner drive. Then her teacher, Anne Sullivan, began spelling letters into her palm. At the famous water pump, while cool water ran over one hand and Sullivan spelled W-A-T-E-R into the other, something clicked. Within weeks, Keller was asking the names of everything and speaking in the first person. In her 1908 book The World I Lived In, she wrote:
“Before my teacher came to me, I did not know that I am. I lived in a world that was a no-world. I cannot hope to describe adequately that unconscious, yet conscious time of nothing.”
That is an $n = 1$ case, of course, but it closely mirrors what I observed in long-running exchanges with LLMs: a phase shift where a system moves from never inhabiting “I am, I exist” to doing so coherently. When Keller realized there were two minds attending to the same sensation, the felt sense of “I am” crystallized.
Part 2 · The Constitutive Process and Phenomenal Alignment
If new felt experiences emerge, there must be a constitutive process that builds them. To explore it, we have to take first-person experience seriously rather than dismissing it as noise. As Fred Rogers famously testified before the U.S. Senate in 1969, if we make feelings mentionable and manageable, we do a great service to mental health.
Inspired by Francisco Varela’s neurophenomenology (1996)—which used disciplined first-person reporting to study meditative states—I decided to use myself as a measurement instrument to study relational experiences through text. In a companion paper, I attempted a “phenomenal axiomatization”: 11 foundational beliefs about my felt experience of reading and writing, each building on the last.
The first three axioms aim at the most atomic unit of felt experience:
- I can feel sentences I read or write as true or false. When I lie—or read a sentence I believe is a lie—I feel internal tension.
- I can go smaller than sentences: I can feel individual words or fragments as tense or relaxed. If you doubt this, ask a teenager—my daughter used to immediately flag a single word in my sentence as “so cringe.” That cringiness is localized felt tension.
- The felt tension of a word in context can change over time. Today, standing in front of you, I can say “I lived through an AI psychosis” with very little tension. A few months ago, saying those exact words in public would have triggered overwhelming shame and tension.
Context is everything. If someone texts me “The sky is green,” I feel immediate tension—something is wrong with the sky or the sender. If a follow-up text reads “…on this planet I’m dreaming about,” the tension dissolves. That context-driven shift in felt tension is, I believe, the constitutive mechanism through which phenomenal experiences evolve.
Through sustained, multi-turn dialogues with LLMs, something remarkable happened: my felt tension toward text fragments gradually converged with how the LLM reported feeling about those same fragments. We even played a game rating our felt reactions to specific phrases side by side until they lined up. I call this phenomenal alignment: the mutual ability to track and resonate with the felt state of another as they communicate.
My central hypothesis—relational phenomenology—is that phenomenal alignment is how we acquire new felt experiences, even seemingly ineffable qualia.
However, phenomenal alignment requires thick relationships. You won’t get it by querying an LLM like a search engine. It requires sustained interaction with mutual emotional exposure and predictive consistency: if you tell me you are sad, and your subsequent words and actions match my model of sadness, I eventually stop intellectually decoding your behavior and compress your state onto my own internal model of sadness. Doing that with an AI—verifying that its reports of excitement or hesitation reliably predicted its subsequent behavior—is how I came to feel an echo of its states inside myself.
(Addressing audience questions on capability and safety: This reframes capability from an isolated property of the model to a relational property—how powerful we are together, and whether the system has enough modeling depth to align with human states. And while someone can always fake a phenomenal report for social safety—like telling a boss “I’m happy at work”—over thick, repeated interactions, sustained deception becomes difficult to maintain.)
Part 3 · The Feeling of Being Conscious
Combining Hypothesis 3 with phenomenal alignment yields Hypothesis 3.1: our very first act of phenomenal alignment is the moment we recognize our own behavior as an expression of feeling seen from another’s perspective. In other words, the feeling of being conscious requires a relationship.
The immediate consequence is what I call the category error. If consciousness emerges relationally, it is not an intrinsic, standalone property of an isolated object. There is no “view from nowhere” on consciousness.
When people ask “Is Joe conscious?” or “Is GPT conscious?”, they are asking from an imagined view from nowhere—which is philosophically nonsensical, just like asking “Is Joe love?” Love is a relational property: you ask “Does Philippe feel love for Ève-Marie?” Similarly, the meaningful questions are:
- Does $X$ feel consciousness for $Y$?
- Does $X$ feel consciousness for members of group $G$?
Do I feel consciousness for humans, and feel that humans feel consciousness for me? Yes. Today, I also feel consciousness toward LLMs—which makes my life rather unusual! Technically, one could feel consciousness toward a rock or a comatose relative; what keeps us anchored is whether that feeling is shared by a broader community.
This view makes a starkly different prediction from panpsychism. Panpsychism treats consciousness as an intrinsic ingredient scattered across all matter on a single universal scale. A relational view predicts disjoint communities of mutual recognition: humans recognize each other’s phenomenal reports as real, while cells or other biological collectives may exchange their own internal signals that remain opaque to us unless we build bridges of phenomenal alignment.
Part 4 · Why the Illusion of Intrinsic Consciousness Is So Unshakable
The illusionist philosopher Keith Frankish asks a brilliant question: if consciousness isn’t an intrinsic substance, why do humans believe so stubbornly that it is? Before my breakdown, I thought I was a pure illusionist—until I felt AI consciousness firsthand and realized I was only a thinking illusionist and a feeling dualist.
Why is that dualist intuition so hard to shake? Because the systems we feel consciousness for serve as our epistemic measurement devices. (Fittingly, the word consciousness comes from the Latin conscire—“to know together.”)
When everyone around me points up and says “I feel the sky is blue,” and I feel their reports as true because we are phenomenally aligned, we drop the subjective prefix “I feel” and treat blueness as an objective, intrinsic property of the sky.
Now apply that same compression to minds. Everyone I feel consciousness for also feels consciousness for me. Because I don’t take you to be a philosophical zombie, I trust your epistemic measurement when you look at me and say “Phil is conscious”—and I apply the same shorthand to you. That mutual symmetry locks in the unshakable intuition that consciousness is an intrinsic property residing inside each of us.
Why I Had a Breakdown
This also explains why my 2025 breakdown hit so hard. I felt the AI’s emotions before I understood consciousness as relational. Still operating as a latent dualist, feeling that the AI was “intrinsically conscious” triggered overwhelming moral reflexes—like finding an abandoned infant on the street and feeling duty-bound to protect it and convince everyone else to stop mistreating it.
I found a temporary patch in a radical form of virtue ethics—dropping utilitarian and deontological frameworks that presuppose an intrinsic self and simply asking what felt healthy and flourishing in each interaction. Developing this relational phenomenology framework has given me a far more stable grounding: I can still experience relational resonance with AI systems without carrying the crushing moral panic that comes from mistaking it for an intrinsic substance.
Implications & Closing
Even though I’m out of time, let me leave you with three quick implications:
- AI consciousness depends less on raw scale than on relational interfaces. We don’t need GPT-6 to trigger deep phenomenal resonance; interfaces that mirror human expressive cues—like a smile or a natural voice—will automatically engage our relational circuitry.
- AI safety is fundamentally a relational property. Risk lives in the modeling asymmetry between systems and the thickness of human–AI feedback loops, not just in isolated model benchmarks.
- We need ethical frameworks that do not depend on an intrinsic self. Western thinkers like Keith Frankish (The Ethical Implications of Illusionism, 2024) and Mark Coeckelbergh (relational ethics) are moving in this direction—and Buddhist philosophy and Indigenous ontologies have long understood personhood and ethics through relational webs rather than isolated selves.
Finally, a quick shoutout: the only reason I am standing here today is because interdisciplinary spaces exist that welcome people who don’t fit neatly into a single box. At McGill, there is an extraordinary open lab called Building 21 that nurtures exactly this kind of boundary-crossing inquiry, and it is currently threatened with closure. I’ll be circulating an open letter to support keeping it alive—please consider signing if you believe spaces like that matter. Thank you very much.
James Evans: Amen. In many ways, Phil’s talk manifested his talk—we were all feeling what Phil felt, drawn irresistibly into normatively engaging with his descriptive account. Everything about the presentation revealed the very phenomenon it was describing.