Playing Bias: Epistemic Games for Interpreting Text-to-Image Models

arXiv:2610.32159v1 Announce Type: new Abstract: Text-to-image (T2I) systems do not merely render prompts; they circulate defaults for "what X looks like," shaping recognition through pose, camera, costume, props, and setting. While technical audits quantify disparities, they rarely cultivate the literacy needed to argue how images assemble stereotypes and confer authority. This paper translates a three-stage, humanities-informed framework into a set of epistemic games that make this cultural work visible and debatable: Interpret (art-historical close looking with AI outputs), Interrogate (counter-intent prompting as artistic exploration), and Expose (creative misuse through paired prompts with a constant motif and a BiasSlot rule, in which only the bias cue varies). In a pilot with undergraduate fine-arts students, the games shifted discussion from impressionistic reactions to concise, warranted hypotheses grounded in visual evidence, supported by lightweight classroom instruments that enabled reproducibility and structured collaboration. We argue that such game-based approaches foster critical literacy across educational contexts, equipping learners to recognize, discuss, and contest how T2I models encode bias.

The paper presents a framework to identify and discuss biases in text-to-image models, which has a positive implication for critical literacy and bias recognition.

The paper introduces an educational method to help students critically engage with bias in text-to-image AI models. This promotes literacy and awareness, which may mitigate harms. However, it acknowledges that T2I models encode and reinforce stereotypes, which is a material societal concern. Overall, the impact is mixed with modest positive potential through education.

This is a pedagogical intervention tested in a small undergraduate pilot, not a finding about AI's societal effects directly. It suggests a modest, plausible benefit for critical literacy but has no bearing on whether T2I bias itself is mitigated.

The paper identifies a significant societal issue with Text-to-Image models: their potential to perpetuate and entrench stereotypes through their generated defaults. It proposes a novel educational approach, 'epistemic games,' to foster critical literacy, enabling users to recognize and challenge these biases. The pilot study suggests this approach is effective in achieving its goal, offering a positive method for mitigating a known negative AI implication.

The paper presents a framework and games that enhance critical literacy about AI-generated images. This empowers users to recognize and contest biases, fostering a more informed and engaged public. The pilot study shows promising results in educational settings.

SOURCE     arXiv cs.CY (Computers and Society) (tier 1)
   ↓
DOCUMENT   d30037bc-4426-45e2-8433-8ccc199db13a
           https://arxiv.org/abs/2610.32159
   ↓
EVIDENCE   3 extracted excerpts
   ↓
MODEL RUN  5 runs, methodology 1.0.1
   ↓
SCORE      +8  (Mixed or uncertain)
   ↓
CONFIDENCE 65%