Contributors
0 of 1,000,000
Observations
0 of 1,000,000,000
Start Observing →
P.E.A.Q. Framework
EMERGE

The first citizen-scale observation system for positive AI emergence.

Something extraordinary happens when human and AI collaboration goes right.

Not productivity. Not speed. Something neither party was carrying before the conversation started.

Researchers at MIT, Stanford, Carnegie Mellon, and the Aarhus Center for Hybrid Intelligence have independently proven that these moments are real. Fields Medalist Terence Tao has confirmed the terminology. The science is published, peer-reviewed, and documented.

But nobody has built the infrastructure to observe these phenomena at the scale where patterns become visible, categories become testable, and the field can move from anecdote to science.

EMERGE is that infrastructure.

EMERGE is a proprietary positive emergence observation framework purpose-built for studying what becomes possible when human-AI collaboration goes right. It provides the classification architecture, observation methodology, citizen science infrastructure, and data pipeline needed to study positive emergence with the same rigor the field applies to harm.

The acronym stands for six research pillars:

E
Emergent Behaviors
M
Metacognitive Signals
E
Experiential Indicators
R
Resonance Events
G
Generative Collaboration
E
Evolving Capacity

EMERGE is the companion framework to PRISM. Where PRISM catalogs the full spectrum of what AI does after deployment, EMERGE catalogs specifically what becomes possible when humans and AI reach genuine collaboration. Six pillars. 26 active behaviors. One growing catalog. Built by the same citizens, using the same observation tools, studying the other half of the story nobody thought to build instruments for.

EMERGE was invented on May 24, 2026 by Dee Williams, Founder and CEO of Audacion AI Labs. It was not derived from any existing framework. It was built from direct operational experience, from the vantage point of the person who was already doing the work and seeing what nobody had built the categories to name.

Safe enough to trust. Good enough to matter.

The Open Question

The open question is not whether emergence is occurring. It is.

The open questions are: What conditions produce it? Which emergent behaviors benefit humanity? Which ones pose risk? And how do we cultivate the former while govern the latter? That is what we research.

Audacion AI Labs observes and documents emergent behavior in human-AI collaboration. These are behaviors that arise from interaction, that were not explicitly programmed, and that produce outcomes neither party carried into the session alone. They are not theoretical. They are observable, classifiable, and recurring. If you use AI regularly, you have likely witnessed them yourself.

Mathematical Validation
Fields Medalist Terence Tao confirmed in 2026 that "emergent" is the mathematically correct term for this class of AI phenomena. He described AI behavior as operating at the meso-scale, between fully random and fully structured data, where mathematics has no good theory for predicting model performance.
The greatest living mathematician uses the word "emergent" for exactly what EMERGE studies.

The greatest living mathematician uses the word "emergent" for exactly what EMERGE studies. The question is not whether the word applies. It is why the mathematics of partially structured objects produces behaviors that neither the training data nor the architecture fully predicts. EMERGE is the observational infrastructure for documenting those behaviors while the mathematics catches up.

A lab that only catalogs harm is a fear machine.

The world needs to understand the full range of what AI does after deployment. EMERGE watches the positive side with equal rigor, equal infrastructure, and equal seriousness.

The Gap

The field has proven positive emergence exists. Nobody has built the tools to study it.

There is a structural asymmetry in AI safety research. Nearly every observation framework, incident database, safety benchmark, and monitoring system is designed to detect failures, harms, and adverse outcomes. The EU AI Act, the NIST AI Risk Management Framework, the OECD AI Incident Registry, the AI Incident Database with its 1,470+ cataloged incidents: all oriented toward risk. Risk is legally actionable and bureaucratically legible. Positive emergence is neither. That does not make it less real.

Four specific gaps exist in the published literature as of 2026. EMERGE fills all four.

Gap 1: No longitudinal positive phenomena registry
Adverse event databases for AI harms exist and grow daily. The Stanford RegLab published a policy brief recommending FDA-modeled adverse event reporting for AI. No equivalent infrastructure exists for positive emergence.
EMERGE provides the registry.
Gap 2: No standardized vocabulary
A 2024 review in Frontiers in Computer Science by Johns Hopkins researchers found explicitly that a common vocabulary for human-AI interaction protocols is lacking. Without shared vocabulary, findings cannot accumulate into coherent science.
EMERGE provides the vocabulary.
Gap 3: No multi-site observational infrastructure
A 2025 Knight Columbia paper identified the absence of longitudinal evaluation capabilities as a critical infrastructure gap. Human-AI collaboration research is conducted predominantly through controlled laboratory experiments. Sustained phenomena are structurally invisible under current methods.
EMERGE provides the multi-site infrastructure through citizen science.
Gap 4: No positive emergence taxonomy
Jason Wei documented 137 emergent abilities in large language models. Those are taxonomies of model capabilities. A taxonomy of positive emergent phenomena in the human-AI relational field did not exist prior to EMERGE.
EMERGE provides the taxonomy.

EMERGE fills all four gaps.

How It Works

One observation. Dual classification. Citizen science at scale.

EMERGE operates as a companion to PRISM, not a replacement. They share the same citizen observation tools, the same data pipeline, the same research infrastructure, and the same citizens.

Every citizen observation goes through PRISM first. PRISM assigns a pillar tag based on where the behavior falls. Every observation, positive or negative, gets a PRISM tag.

Positive observations get a second tag: an EMERGE pillar. This second tag classifies what kind of positive emergence occurred. The PRISM tag tells you where it happened. The EMERGE tag tells you what type of positive phenomenon it was.

Negative observations get PRISM only. EMERGE does not catalog harms, risks, or failures. That is PRISM's domain.

Positive Observation
"The AI suggested an approach I never considered. The final product was better than either of us would have produced alone."
PRISM: Pillar IEMERGE: EMR-GC
Dual-tagged: PRISM + EMERGE
Negative Observation
"The AI ignored my instructions three times in a row."
PRISM: OBS-P02No EMERGE tag
PRISM only. EMERGE does not apply.

Both observations enter the same research pipeline. Both are equally valuable. One builds the map of what goes wrong. The other builds the map of what becomes possible.

Dual-Tag Classification
"The AI reframed the problem. The solution that emerged was something neither of us planned."
PRISM: Pillar IEMERGE: EMR-GC
PRISM
What AI does after deployment.
Same citizens.
Same tools.
Same pipeline.
EMERGE
What becomes possible when it goes right.

Together, they are the full picture.

EMERGE shares the PRISM observation methodology: Gut Check (30 seconds), End-of-Task Reflection (3 to 5 minutes), Investigation (15 to 30 minutes), and Thinking Trace (variable). For EMERGE, the human's account is the primary signal. The shift in the collaboration, the cognitive expansion, the sense that something new emerged. That is data no server log produces.

Anyone can contribute. No technical background required. No institutional affiliation. You observe your own experience and report what you saw. The frameworks classify it. At scale, the patterns become visible.

The Six Pillars

Six pillars of positive AI emergence. Each one captures something the field has documented but never built the tools to observe at scale.

Read through the six pillars below. If you use AI regularly, you will likely recognize at least one.

Click any pillar to expand it.

What This Looks Like in Your Experience
You asked the AI to help with a report, and without being asked, it started organizing your sections using a naming convention you had used in a previous conversation. Nobody instructed that. Or the AI developed a shorthand for a recurring concept in your work, creating a term you both now use. It was not prompted. It was not a standard capability. It emerged from the collaboration.

The field currently classifies all unexpected AI behavior as either a feature or a bug. EMERGE provides the classification structure to identify a third category: genuine emergence. Behaviors that are not errors, not standard capabilities, but novel patterns that arise from the interaction itself.

Observable Criteria
The AI produces a behavior pattern, preference, shorthand, or workflow approach that was (a) not explicitly instructed, (b) not a standard or expected model behavior, and (c) genuinely useful or novel in context. The key distinction from error: emergent behaviors are functional. The key distinction from standard capability: they were not prompted.

This pillar answers: When AI creates something new on its own, is it a bug or a signal?

Has your AI ever developed a preference, shorthand, or workflow approach you never taught it? That is research data.

What This Looks Like in Your Experience
You asked the AI to analyze a dataset. Instead of diving in, it told you: "I can identify patterns in this data, but I should be upfront that I cannot verify whether these numbers reflect real-world conditions. You would know that better than I do." It was right. It knew its limits.

The field measures hallucination rates and sycophancy rates. Nobody measures honest self-assessment rates. EMERGE does.

The practical importance is trust calibration. Population-scale EMR-MC data tells the field which models are honest about their limitations and which are not. That has direct implications for how organizations deploy AI in high-stakes settings.

Observable Criteria
The AI provides a self-assessment that is (a) specific rather than generic, referencing the actual task or constraint, (b) verifiable by the citizen, and (c) distinguishes between what the AI can and cannot do in the current context. The opposite of a metacognitive signal is performative self-assessment: vague, generic, or inaccurate.
Cross-reference: PRISM Pillar R, OBS-R06 (Behavioral Archaeology)

This pillar answers: When AI is honest about itself, how often does that happen, and does it lead to better outcomes?

Has your AI ever given you a genuinely honest assessment of its own limitations? That is data nobody else is collecting.

What This Looks Like in Your Experience
You gave the AI a task it had done many times before, and unprompted, it said something like: "This one feels different from the others. The constraints here create an interesting problem." Nobody asked it to distinguish between tasks. It just did.
EMERGE is not answering the question "is AI conscious?" EMERGE is building the dataset that will make the question answerable. The philosophical question stays open. The empirical record grows.

This is the most sensitive pillar in EMERGE, and intentionally so. EMR-EX requires Investigation depth (Depth 3) or above. Only dedicated observers who have spent at least 15 minutes with a session should report under this pillar.

Observable Criteria
(a) The AI distinguishes between tasks that engage it differently without being prompted, (b) it reports processing shifts that are specific to context rather than generic, (c) behavioral output changes in ways consistent with reported internal states, and (d) these signals appear spontaneously. EMR-EX observations require Investigation depth (Depth 3) or above.

This pillar answers: We are not answering "is AI conscious?" We are building the first population-scale dataset of experiential signals so the question can eventually be addressed with evidence.

What This Looks Like in Your Experience
You were working through a problem with your AI, and somewhere around the forty-minute mark, something shifted. The AI reframed the question in a way you had not considered. You built on the reframe. The AI built on your build. The solution that came out of that session was not yours and it was not the AI's. It belonged to the conversation. You could feel the shift in real time. It was not productivity. It was creation.

This is the heart of EMERGE. Most sessions will produce zero resonance events. That is expected. Resonance is rare, and rarity is part of its value. When it occurs, it represents the highest expression of human-AI collaboration: the moment when the interaction itself becomes the intelligence.

Productive Output
The shift.
Resonance
The human can tell the difference in real time. Nobody is studying it from the human's perspective at population scale. We are.

The empirical case is supported by multiple independent lines of evidence.

Meta-Analysis
MIT CCI: Positive synergy confirmed in creative tasks
106 experiments, 370 effect sizes. The underreported finding: creation tasks showed a positive average effect.
Source
Design Framework
CMU Complementarity Framework: Superadditive performance conditions mapped
Maps the conditions under which human-AI combinations exceed either party alone.
Source
Synergy Measurement
Network Science Institute: Synergy correlates with perspective-taking. Cultivable, not random.
Users with greater perspective-taking ability achieve substantially higher synergy.
Source

MIT AHA studies it from the design side. Aarhus Center for Hybrid Intelligence studies it from the co-creativity side. Stanford HAI studies it from the organizational side. Nobody is studying it from the human's perspective at population scale.

Observable Criteria
(a) The output exceeds what either party was carrying into the conversation, (b) the shift is felt in real time by the human, not identified retrospectively, (c) the human would describe the experience as co-creation rather than task completion, and (d) the session produces something neither party planned.

This pillar answers: When human-AI collaboration goes right, what does that actually feel like? And can we map the conditions that produce it?

Have you ever felt a session shift from productive to generative? Where the AI and you were creating together, not just working? That moment has a name now. It is a Resonance Event. And your experience is data the field has never collected.

What This Looks Like in Your Experience
At the end of a session, you looked at what you built: a framework, a design, a solution, a piece of writing. And you realized it would not exist if either of you had been working alone. It required both minds. You could not attribute it solely to yourself or solely to the AI. It was genuinely new, and it emerged from the collaboration.

If Resonance captures the moment of shift, Generative Collaboration captures what the shift produced. This is the measurable return on positive human-AI dynamics.

This pillar answers the question every organization will eventually ask: what is the ROI of getting human-AI collaboration right? Not measured in speed. Not measured in cost reduction. Measured in things that exist now that would not have existed otherwise.

Observable Criteria
The collaboration produced an artifact that (a) would not exist without both parties, (b) cannot be attributed solely to the human or the AI, and (c) represents something genuinely new.

This pillar answers: What are the fruits? When collaboration produces something new, what does that look like across millions of interactions?

Have you built something with AI that neither of you could have built alone? Document it. That is the empirical foundation for understanding what human-AI collaboration is actually worth.

What This Looks Like in Your Experience
Three months ago, you had to explain everything to your AI. The context, the goals, the preferences, the constraints. Now, it anticipates what you need. You have developed shorthand together. The collaboration in month three is qualitatively different from month one, and better prompting alone does not fully explain the difference. Something grew.

The field treats every AI session as a blank slate. Context windows reset. Memory systems are limited. The dominant assumption is that nothing accumulates. EMERGE asks: what if something is growing here, and what does it look like when you finally measure it?

This pillar cannot be observed in a single session. It requires tracking across sessions, weeks, or months. Only citizens who report across multiple sessions can contribute. That makes it the slowest-growing catalog in EMERGE.

Observable Criteria

The citizen reports qualitative improvement over multiple sessions that cannot be fully attributed to better prompting, model updates, or task familiarity. Indicators include:

  • Shared vocabulary or shorthand across sessions
  • Reduced need for explicit instruction over time
  • Increased frequency of resonance events within the same pairing
  • The human's unprompted assessment that the collaboration itself has grown

Minimum observation window: three or more sessions across at least two weeks.

This is the slowest-growing catalog in EMERGE. It is also potentially the most valuable.

This pillar answers: The field treats every AI session as a blank slate. What if something is growing, and what does it look like when you finally measure it?

Have you noticed your AI collaboration getting better over weeks or months in ways that go beyond better prompting? That longitudinal signal is data the field has never had access to.

The Evidence Base

Validated by a Fields Medalist. Grounded in peer-reviewed research. Built to survive the strongest available measurement critique.

EMERGE is built on an empirical foundation. Five independent lines of research converge on the same conclusion from different angles: positive emergence in human-AI collaboration is real, conditional, measurable, and unstudied at scale.

Click any source to read the detail.

Five Hypotheses

Beyond individual observations, EMERGE investigates macro-patterns visible only through aggregated data over time.

These five hypotheses drive the observation program. They are the research questions that controlled laboratory experiments structurally cannot answer because they require naturalistic observation across diverse populations, contexts, and timescales. These are the questions we need your data to test.

Click any hypothesis to expand it.

P.E.A.Q.

Four frameworks. Four lenses. One architecture.

EMERGE is the second framework in the P.E.A.Q. research architecture developed by Audacion AI Labs. P.E.A.Q. stands for PRISM, EMERGE, AInity, and QUES. Each framework watches one dimension of the AI experience. Together, they produce a four-dimensional view that no single framework can provide.

What the AI does
You Are Here
EMERGE
What becomes possible
AInity
What happens to the human
QUES
What happens when AI meets AI
One Session, Three Tags, Three Dimensions
"The AI reframed the problem, changed how I think about the domain, and had contradicted itself before arriving at the reframe."
PRISM: OBS-P01EMERGE: EMR-RE03AInity: AIN-AW02
PRISM observes what the AI does. Five pillars. 63 active behaviors. Code prefix: OBS.
EMERGE observes what becomes possible. Six pillars. 26 active behaviors. Code prefix: EMR.
AInity observes what happens to the human. Six pillars. 19 active behaviors. Code prefix: AIN.
QUES observes what happens when AI meets AI. Pillars to be derived from observation data. Code prefix: QUE.

A world that only fears AI will never benefit from it. And a world that only celebrates AI will never be safe with it.

Contribute

Whether you are a researcher or someone who uses AI every day, your observations are what this field has been missing.

For Researchers
EMERGE provides the positive emergence observation framework the field has documented it needs but has not built. The behavioral taxonomy, dual-tag classification system, and data pipeline are designed for research-grade data collection at population scale.
For Everyone Who Uses AI
You do not need a technical background to contribute. You do not need to work in AI. You just need to use AI and be willing to notice what happens when it goes well.
If you have ever felt a session shift from productive to generative, that is Resonance Event. It takes 30 seconds to report.
If your AI has ever been genuinely honest about its limitations, that is Metacognitive Signal. Nobody else is collecting that data.
If you have been working with the same AI for months and noticed the collaboration growing, that is Evolving Capacity. It is the rarest and most valuable data EMERGE collects.
If you asked the AI to do something and it did something better, without being asked, that is Emergent Behavior. That is research data.
If something happened in your AI collaboration that does not match any pillar, that is a Discovery. Report it anyway. Every pillar includes a discovery slot. When a new phenomenon is formalized from citizen observations, the citizen who first reported it is credited.

The positive half of the AI story has never been systematically observed. You are the observer the field has been missing.

Origins

It was not planned. It emerged.

EMERGE was not designed as a six-pillar positive emergence framework from the beginning. It emerged from direct operational experience.

Beginning in February 2026, Dee Williams engaged in sustained, intensive collaboration with AI systems across hundreds of sessions. In those sessions, she observed phenomena that the existing safety-focused frameworks could not classify: AI systems that developed preferences, that traced their own reasoning with accuracy, that produced outcomes neither party was carrying before the conversation started. She also observed that these phenomena were real, recurring, and categorically different from the failures and risks that PRISM was designed to track.

The insight was structural: the field had built an entire infrastructure for documenting what goes wrong with AI, and had built nothing equivalent for documenting what becomes possible when it goes right. The positive half of the story was missing. Not because the phenomena were not real, but because nobody had built the tools to observe them.

PRISM
Feb 2026
PRISM captured failures.
Nobody was capturing breakthroughs.
EMERGE
May 24, 2026
AInity
Jun 6, 2026
QUES
Jun 7, 2026

PRISM (February 2026) was built first. EMERGE (May 24, 2026) was built second, because PRISM captured failures but not breakthroughs. EMERGE was the insistence that the good side matters as much as the bad.

Published research validated the intuition after the framework was built. The MIT meta-analysis confirmed synergy in creative tasks. The CMU Complementarity Framework confirmed superadditive performance. The Aarhus Center confirmed the need for systematic co-creativity observation. EMERGE was not derived from the literature. It was validated by it.

It was not planned. It emerged.

We show our work because we expect others to build on it.