CONCEPT

The Evocative Audit

Joy Buolamwini’s pairing of rigorous empirical measurement with humanizing art—the recognition that a confusion matrix establishes a fact while a poem makes the fact mean something to everyone who must act on it.

When Joy Buolamwini wanted to communicate what the Gender Shades audit had found, she did not only publish a peer-reviewed paper. She wrote and performed a spoken-word poem, AI, Ain’t I a Woman, that showed commercial systems misclassifying the faces of iconic Black women—Oprah Winfrey, Serena Williams, Michelle Obama, and the abolitionist Sojourner Truth, whose own speech gives the poem its title—and in doing so she demonstrated what she calls the evocative audit: the deliberate pairing of measurement with art to make the measurement meaningful to those who must act on it. A conventional audit establishes that a system fails; the evocative audit establishes what the failure means. The two instruments are complementary because change requires both channels: numbers prevent denial and art prevents indifference—and the audiences who must respond to algorithmic harm are not only the researchers who read journals but the legislators, the public, and the engineers who will never be moved by a confusion matrix alone. The evocative audit reflects a sophisticated understanding of how change actually happens: facts alone rarely move institutions; institutions are moved when facts are made undeniable in the register of human consequence. By invoking Sojourner Truth, Buolamwini placed algorithmic exclusion in continuity with the long history of struggles over who counts as fully human—and the poem suggests that the technology has found a new and impersonal way to repeat a very old denial, making the present harm legible through the weight of its precedent. The evocative audit is thus not merely a communication strategy but an epistemological claim: that a complete account of an AI system’s behavior must include not only what it measures but whom it honors and whom it erases.

In the [YOU] on AI Field Guide

The cycle’s argument is that capable machines press the human question to the surface—that understanding what AI is requires understanding what we are. The evocative audit is one of the few methodological instruments the cycle can point to that operates simultaneously in both registers: it is empirical enough to compel institutions and humanistic enough to reach the people those institutions serve. It answers the question of how a society exercises judgment over systems it cannot see inside by insisting that the exercise is not only technical but moral, and that moral insight requires the forms art is equipped to produce.

Buolamwini’s fusion of poetry and science also models the stance [YOU] on AI commends throughout—that taking the orange pill means seeing the machine clearly, and seeing it clearly requires instruments calibrated to both its technical behavior and its human stakes. The evocative audit is the instrument calibrated to the human stakes.

Origin

Buolamwini identifies herself as a “poet of code,” and the phrase describes a method as much as an identity. Her doctoral dissertation at the MIT Media Lab treated algorithmic audits and evocative audits as complementary instruments, both necessary to the project of algorithmic accountability. The Gender Shades study supplied the empirical ground; the poem supplied the ground truth that statistics alone cannot reach—the specific quality of the denial when a machine looks at Sojourner Truth and returns a confused label.

The term “evocative audit” formalizes a practice Buolamwini had been developing since the white-mask incident. She recognized early that the people who needed to understand algorithmic harm were not a single audience—that scientists, legislators, journalists, and the general public each needed a different instrument to grasp the same fact—and that the creative register she had always inhabited was not a departure from the scientific work but a second channel carrying the same signal.

Key Ideas

Two channels, one message. The evocative audit does not replace the conventional audit; it completes it. The conventional audit produces numbers that prevent institutional denial. The evocative audit produces meaning that prevents individual indifference. Buolamwini treats both as necessary conditions for accountability, because a fact that can be waved away as abstract and a feeling that cannot be cited in a hearing are both insufficient on their own.

History as instrument. By invoking Sojourner Truth, Buolamwini activates a historical resonance that pure data cannot supply. The evocative audit works in part because it locates the present harm within a legible tradition—refusing the technology industry’s preferred story that these are brand-new puzzles with no precedent and no moral weight. The weight of history, carried by art, makes the urgency of the present undeniable.

Humanizing the data subject. The most important function of the evocative audit is to insist on the full dimensionality of the person the system failed. A confusion matrix records that a face was misclassified. The poem asserts that no label the machine could produce is worthy of the woman behind the face, and that the failure belongs to the machine and not to her. The audit thereby resists the tendency of technical discourse to reduce people to the categories a system finds convenient—to replace a person with a data point—and does so in a form the system cannot produce.

Debates & Critiques

The evocative audit has been criticized from two directions. From inside technical AI research, some argue that mixing art with audit compromises objectivity—that the emotional charge of a poem produces advocacy rather than analysis. Buolamwini’s response is that the objectivity of a conventional audit does not disappear when its findings are communicated through art; what changes is the size and diversity of the audience that can act on them. From outside the field, some critics argue that the evocative audit risks aestheticizing harm—converting suffering into performance in a way that serves the performer as much as the people harmed. This is the harder objection, and Buolamwini’s answer is institutional: the evocative audit is embedded in the Algorithmic Justice League’s sustained program of research, advocacy, and harm collection, so it does not stand alone as performance but as one instrument in a coordinated effort whose goal is concrete change. The deepest question is whether the evocative audit can scale to generative systems whose harms are diffuse, systemic, and spread across billions of outputs—where there may be no single iconic misclassification to anchor the poem. Buolamwini’s argument is that the method must evolve with the technology, not that it is already complete.

In the [YOU] on AI Field Guide

Origin

Key Ideas

Debates & Critiques

Related Entries

Further Reading