The Oppositional Gaze

In the AI Story

The essay emerged from hooks's experience of watching films in segregated and then integrated theaters, and her observation that Black female spectators often brought a different quality of attention to the screen than white audiences. They noticed absences—where were the Black women?—and distortions—why were Black women presented only as mammies, maids, or objects of white male sexual fantasy? The noticing was not academic. It was survival. To internalize the cinema's vision of Black womanhood was to accept your own negation. The oppositional gaze was a practice of self-preservation that required vigilance, collective discussion (the conversations Black women had after the film, naming what they had seen), and the construction of alternative frameworks for understanding Black female identity.

hooks extended the concept beyond cinema to every domain of representation—advertising, journalism, academic discourse, and now algorithmic systems. The gaze required is not merely skeptical—skepticism doubts everything equally—but critical in the specific sense of being attuned to power, to whose perspective is centered, whose is erased, and what the centering and erasure accomplish. Researchers studying AI image-generation and captioning systems have found exactly what hooks's framework predicts: when large language models describe archival photographs of colonized peoples, they reproduce the colonial gaze—essentialism, dehumanization, othering, infantilization—because the training data from which they learned to see carried those patterns. The models learned colonial ways of seeing and now reproduce them with computational consistency.

The oppositional gaze in the AI age requires the capacity to read generated outputs against the grain, to ask not merely 'is this accurate?' but 'whose accuracy?' A large language model asked about the history of colonialism will produce an answer. The answer will cite facts, present perspectives, appear balanced. The oppositional gaze asks: whose facts? Which perspectives are centered, which are quoted but not privileged, which are absent? How would this answer differ if the training data had included more texts produced by colonized peoples in their own languages, according to their own frameworks? The question is not answerable by the model itself. It can only be answered by a person who has developed the critical consciousness to perceive the watermark of the dominant culture even—especially—when the output appears neutral and authoritative.

Origin

hooks wrote the essay in direct response to feminist film theory, particularly the work of Laura Mulvey on the male gaze, which had argued that classical Hollywood cinema structured spectatorship around a male heterosexual perspective. hooks affirmed Mulvey's analysis and extended it: the gaze was not only male but white. Black women were positioned in cinema not as subjects of desire or identification but as absences or Others. The development of the oppositional gaze was Black women's collective response to this structural positioning—a learned practice of critical looking that refused the camera's way of seeing.

Key Ideas

Looking back as resistance. The act of looking critically at representations that degrade or erase you is itself a political act, a refusal to internalize the dominant culture's vision of who you are.

Gaze must be learned. The oppositional gaze is not a natural capacity but a cultivated practice requiring community, collective discussion, and frameworks for understanding whose interests representation serves.

Representation encodes power. Every act of representation—cinematic, textual, algorithmic—makes choices about whose perspective to center, and those choices are political whether or not they are acknowledged as such.

AI reproduces colonial gaze. Large language models trained on archives that themselves carry colonial assumptions reproduce those assumptions with statistical consistency, presenting colonial ways of seeing as neutral description.

Critical reading as infrastructure. The capacity to read AI outputs oppositionally—asking whose knowledge, whose framing, whose exclusions—must be taught as foundational literacy for the AI age, not as an advanced critical skill.