Lack of Representation in Generated Content

Lack of representation in AI-generated content occurs when AI systems fail to output diverse and inclusive content (text, images, etc.) that accurately represents all populations. This lack of representation can manifest as over- or under-representation of certain groups in outputs (e.g., never generating images of certain ethnicities) or the reproduction of societal stereotypes (e.g., only generating images of male scientists).
These discrepancies generally arise from biased training data that either does not contain instances of certain groups or primarily contains the stereotyped representation. Without interventions, generative models will reproduce the behavior seen in these training sets.

Poor representation of different groups can impact societal perceptions and reinforce biases. Over-representation of certain groups can create a skewed sense of normalcy and dominance of certain groups over others. Conversely, under-representation can render some groups invisible, denying them recognition and reinforcing societal exclusion. Misrepresentation can result in the reproduction of stereotypes or misinformation related to specific groups.

- AIID Incident 18: Gender Biases of Google Image Search: Google Image returns results that under-represent women in leadership roles, notably with the first photo of a female "CEO" being a Barbie doll after 11 rows of male CEOs.