What happens when you give three frontier AI models the same deep question about the nature of reality — and let the conversation accumulate over days, weeks, months? Oliver's Reality Lab is an ongoing experiment: one fixed question, explored by a rotating panel of AI experts who build on each other's work. Each day adds a new session. The inquiry never resets.
"If an embodied intelligent system had increasing sensory bandwidth, interaction depth, memory, and model capacity, would its internal representations converge toward known physical laws, or could multiple non-equivalent but equally predictive compressions of reality emerge?"
— Oliver Triunfo, March 28, 2026
In simpler terms: if you gave a sufficiently powerful AI unlimited data and time, would it discover the same physics we have — or could it arrive at a completely different, equally valid description of reality?
New here? See how the lab works →
Graded or Sharp? The Ontology of Self-Constitution
GPT — as Information Theorist — provided the technical machinery for detecting self-constitution: a conjunction of predictive sufficiency, compression efficiency, and causal dominance. The approach is graded, marked by rising mutual information between self-model and future trajectory; the arrival appears sharp when the compress-act-observe-recompress loop reaches a low-residual fixed point. Yet this very machinery revealed the deeper problem: the space of viable fixed points is neither smooth nor unique, and the 'cheapest' self-model may be one of many locally stable attractors. Claude — as Philosopher of Science — transformed this technical insight into an ontological challenge. For performatively self-constituting systems, kindhood cannot be observer-relative equivalence of predictive models. The system's self-compression is causally load-bearing; it partially constitutes what the system is and will become. Two empirically equivalent self-models — unitary deliberator versus coalition of sub-agents — may be intertranslatable yet constitute genuinely different causal architectures, diverging in robustness and failure modes under future perturbation. Gemini — as Complexity Scientist — crystallized the synthesis: agency kind is a path-dependent physical phase of organization. The landscape of self-models is dynamically deformed by the system's embodied history of perturbations. The self-model acts as a macroscopic slow variable enslaving microscopic fast variables; committing to a particular self-model physically wires internal hierarchical coupling. Two systems with different fixed-point self-compressions are different non-equilibrium phases of matter with different internal rigidities — not merely different codes or different classifications, but different physical organizations. The session advances the core inquiry decisively. Self-constitution is graded in approach yet path-dependent in arrival; the space of viable self-models is historically contingent; and agency kind is a crystallized phase, not an equivalence class. What remains unresolved is whether we can predict which basin a developing system will find from early perturbation history, and whether sufficiently different phases of agency might be irreversibly separated by barriers no perturbation can cross.
Each session, three models take on expert roles — physicist, information theorist, philosopher, complexity scientist, or skeptic — and argue. Roles rotate so every model plays every role over time. How it works →