THEFTSubscribe

07 · Cross-cultural IDI study

YouTube

Responsible AI image generation study

Cross-cultural depth research on inclusive AI image generation principles

Status
Shipped.
Service
Behavioral research, responsible AI
Scope
Principles for inclusive AI image generation
Markets
USA, Brazil, Japan, Nigeria, Indonesia, India
Studio
THEFT Studio

YouTube needed first-principles guidance for how inclusive AI image generation should behave by default, not as a post-hoc filter on model output.

The studio ran a cross-cultural depth-interview study: six ninety-minute IDIs anchored by cultural immersions in Lagos, Jakarta, and Mumbai, with coverage across six markets, the USA, Brazil, Japan, Nigeria, Indonesia, and India.

The work produced a five-force framework for the evolving representation landscape and principle-level guidance across four representation dimensions, with homogeneous defaults rejected in every market studied.

Method · effort by phase

01
02
03
04

01 · Framing

Scoped the six markets, recruited participants with representative backgrounds, and defined a discussion guide grounded in representation rather than raw accuracy.

02 · Depth interviews

Six ninety-minute IDIs explored how participants evaluate AI-generated imagery for representation across race and ethnicity, gender, attire, and other characteristics.

03 · Cultural immersions

In-country fieldwork in Lagos, Jakarta, and Mumbai grounded the IDIs in local visual culture, attire norms, and daily life.

04 · Synthesis

Translated findings into a five-force framework and a set of high-level principles across four representation dimensions.

What the work surfaced

01

Homogeneous defaults are rejected across every market studied

Across all six markets, participants rejected AI imagery that defaulted to a single majority look, and the reaction held across race, gender, and body dimensions.

02

Balanced local-population representation is the baseline

Participants expected AI imagery to reflect balanced local-population representation rather than over-indexing on a global majority: locality-aware, not universally uniform.

03

Attire, age, and body diversity matter as much as race

Race and ethnicity are the most visible axis, but attire, age, body, and ability carried nearly equal weight in whether an image read as genuinely representative.

04

Representation is an evolving landscape, not a fixed target

The five-force framework describes representation as an active field rather than a static specification.

05

AI inherits the training set; design choices inherit the AI

Participants expected product teams to treat generation defaults as a design decision carrying responsibility downstream, not as a technical output.

Delivered

  • Cross-cultural depth-interview readout
  • Five-force framework · evolving landscape of representation
  • High-level principles across race / ethnicity, gender, attire, and other characteristics
  • Cultural-immersion field notes from Lagos, Jakarta, and Mumbai
  • Implications deck for responsible AI product teams

Outcome

A cross-cultural framework for inclusive AI image generation that holds across six markets.

The work gives responsible AI teams a research-backed framework to reason about representation defaults. The five-force landscape and the four-dimension principle set reframe inclusive generation as an active design responsibility rather than a post-hoc filter on model output.

Framework
Five forces · four dimensions
Coverage
6 countries · 3 immersions
Reframe
Output filter to design default
Informs
Responsible AI image generation