The Recognition Machine (2019—)

The Recognition Machine draws a parallel between colonial and contemporary surveillance technology and the relation to race. In a colonial context, photography was used in combination with pseudo-science to establish false truths about race that fit a racist agenda. Today, artifical intelligence techniques are used in similarly questionable ways that contemporary critics like Ruha Bejamin have described as a "New Jim Code" (referring to the "Jim Crow" laws of segregation based on race in the United States) and Safiya Umoja Noble have termed "technological redlining", referring to a history of racist practices masked by seemingly "neutral" ways of classification.

A "ready made" model is used that has been trained on a data set of images pulled from the Internet labelled with emotions. One reason the emotion model is interesting is that it makes the subjective aspect of these labels more apparent: an image appears "happy" or "sad" by who and in what context? For instance, in the data set several images of Barak Obama have been labelled "angry"; these judgements reflect the biases and prejudices of not only those who placed the images online, but the manner in which these images and labels were (semi) automatically gathered using web search engines and automation using code. Rather than reinforce the claims of these data models, TRM wants to trigger discussions about how and why these algorithms work they way they do, and call their use into question.

Visitors to the photobooth are encouraged to place a copy of their photostrip along with others, the impromptu collection echoing the way modern data sets are gathered, often without the (full) understanding of participants

The photostrip produced by the photobooth shows three pictures, the visitor, the match from the archive, and an image from the data set used to train the algorithmic emotion classifier. Rather than simply giving an "absolute" response (the predominant matching emotional label), the model is used in a relational way. Given an image, the algorithm outputs a set of 7 "weighted probabilities" for each of the 7 possible emotional labels (for instance 33% happy, 27% sad, 8% disgusted, etc.). The archive is then searched to find the image which produced the most similar response. In this way TRM creates a kind of analogical relationship, you are (the visitor is) paired with an image that the model (problematic as it is) (mis)recognizes in a similar way. Combined with the subjectivity of the emotional labels, the idea is to attempt to construct a kind of solidarity between the visitor and the people whose faces are appear in the archive.