The same message builds trust with one audience and breaks it with another. That variation has structure.

Humor Genome is a research program that maps how audiences interpret the same message differently, and turns that variation into something you can measure and design for.

Sponsor a research theme Read the model

Source material: NBC broadcasts, Comedy Central specials, NYC clubs including the Comedy Cellar.

The Model

Audience-dependent meaning

Most communication models treat meaning as something a sender encodes and a receiver decodes. When a message fails, the explanation is usually noise, ambiguity, or poor targeting. But the pattern we keep seeing is different: the same message, delivered identically, produces structurally different interpretations across audiences. Not random noise. Predictable divergence.

Humor Genome starts from that observation. We treat interpretation not as a property of a message, but as a function of the relationship between a message and its audience. The same joke kills in one room and dies in another, not because one room "got it" and the other didn't, but because each room brought a different interpretive frame to the same material.

The research program maps these frames. We identify the structural features of messages that make them flexible (landing across many audiences) or brittle (only working for one). We model the audience variables that predict divergence: shared references, status dynamics, emotional priors, cultural context. And we build instruments that make these patterns visible in real time.

Comedy is the proof surface because laughter is the fastest, most measurable signal of interpretation we have. A laugh is a binary, involuntary event. Silence is equally legible. This makes live comedy the highest-bandwidth environment for studying how meaning shifts across listeners.

The Program

What we're investigating

Sponsors fund research themes here. Those themes are investigated through live performances, hackathons, and experiments at Midtown Show. Systems generated by the lab ship on sound.fan, where partners take them further.

See the callback experiment tested live at Midtown Show · See Backlash Predictor, shipped from this research

Interpretation Divergence
Why the same message builds trust with one audience and fails with another. Measuring alignment, disagreement, and the shape of audience splits.
Example

We ran the same 5-minute set for three audience configurations and tracked where callbacks landed. Audiences with shared cultural context caught callbacks 3x more often than mixed rooms. The divergence wasn't noise — it followed the fault lines of shared reference.

See the full finding →
Mixed Shared context 1x 3x
Persona Modeling
Defining and simulating distinct audience types. What different groups pay attention to, react to, and optimize for when they interpret the same input.
Example

Room Sense Live modeled 200 simulated sets against distinct audience personas. Comedians who pre-tested against these personas had 40% fewer flat segments on stage — the personas surfaced weaknesses real audiences would punish.

See the full finding →
Baseline Pre-tested 100% −40%
How It Lands
Real-time observation of audience reactions as content unfolds. Live environments like Room Sense where interpretation is simulated, measured, and compared across audience types in the moment.
Example

Using Laugh Spikes, we mapped laughter timing across age-segmented audiences for identical material. Peak laugh moments diverged by up to 12 seconds between cohorts — younger audiences laughed at subversion, older audiences at recognition.

See the full finding →
Meaning Structure
The internal architecture of an idea that determines how it can be interpreted. Mapping the structural features that make some messages flexible across audiences and others brittle.
Example

Example forthcoming.

Decision Surfaces
How interpretation affects downstream decisions. When a message triggers trust, action, hesitation, or rejection across different evaluators.
Example

Example forthcoming.

About

About

Humor Genome is an independent research effort exploring how the same joke lands differently depending on who's listening. We break down stand-up clips from NBC, Comedy Central, and rooms like the Comedy Cellar, then model audience reactions to see where meaning holds and where it fractures. It's part analysis, part tooling, part experiment.

Method
Transcript + clip analysis, persona-based audience modeling, reaction classification
Data
Public stand-up performances (broadcast + live venues)
Status
Early-stage, open to collaborators

Who's building this

Michi Yamamoto builds systems that make audience perception visible. His background spans comedy production, software engineering, and AI research, a combination that produced Midtown Show, Room Sense Live, and dozens of hackathon builds exploring how different audiences read the same signal. Humor Genome ties this work into a single research question: how does meaning shift across listeners, and how can we design for it?