(Originally published on Overtime Online, 16th November 2020.)
We have given artificial intelligence power and jurisdiction over decision-making processes that can be simultaneously nuanced and personal. We trust that algorithms are a neutral, objective factor in this system, and are less prone to corruption as they are not vulnerable to the effects of human error. But from whose perspective is this the case?
Shalini Kantayya’s documentary film Coded Bias takes steps towards breaking down and exposing the in-built structures of power which feed into machine-learning and AI tech. The director achieves this through conducting interviews with an array of academics and members of activist groups who are monitoring a broadly unregulated area of tech development.
Together, the subjects have all uncovered some uncomfortable truths about rapidly developing technologies which are presently being evaluated through non-consensual or otherwise discrete trials on the general public.
The documentary follows the progress of Joy Buolamwini, an AI researcher at MIT and founder of the Algorithmic Justice League, as she takes her ethical concerns about facial recognition technology to a hearing with Congress in the US.
Buolamwini was mobilised to take action against the irresponsible programming of facial recognition tech based on the incidences of automated discrimination she experienced while conducting her own experiments with AI.
She found that the facial recognition programs she had been encoding into her work failed to register visual information about her face, but when she held up a plain white mask to cover her skin and features the program was able to detect a face – despite the fact that the mask had almost no human qualities about it.
The inaccuracies in the AI that Buolamwini studied were specifically biased against people of colour and, to a significant degree, against women; for example, IBM’s facial recognition tech could identify the face of a white male with 99.7% accuracy, whereas its ability to identify the faces of women of colour was reduced to only 65.3% accuracy.
This is what Buolamwini considers as an embedding of unconscious bias, based on the – likely unintentional, but nonetheless destructive – myopia which resulted from the homogeneous backgrounds and experiences of the field’s initial programmers. To achieve objectivity of an algorithm, we must first be able to assume that all its variables are equal.
But as this has not historically been the case, what real-world effects does this inequality in information have on the lives of everyday people? Kantayya explores this in Coded Bias by focusing in part on the work of Big Brother Watch, a UK-based activist group led by director Silkie Carlo, who are campaigning against the exploitation of data rights and civilian privacy, which the group consider to be violations of civil liberties.
Kantayya’s film captures an important moment where a 14 year old schoolboy is accosted by police, fingerprinted and added registered on their criminal database as a response to a suspected match in facial profile detected by cameras police stationed in the area capable of facial recognition.
The match was a misidentification—entirely incorrect, but the boy’s data was still logged on file. The use of an inaccurate and biased profiling algorithm resulted in harassment of an underage child by police.
Coded Bias sheds light on a number of other case studies which are equally as important to stimulating discussion over the moral culpability of big tech in cases of digital injustice. The film also does an excellent job of demystifying AI development as a discipline; as Cathy O’Neil says in her interview segment, it really is “just math” [sic].
With present-day events being intercut with our glamorous visions of intelligent futures—clips from the cinematic adaptations of Minority Report, 2001: A Space Odyssey and 1984—Kantayya has constructed an enjoyable and broad-ranging study of pressing ethical issues contemporary tech development, while giving its audience space to appreciate the women on the front lines of tech research who are championing the defence of our right to privacy as global citizens.
Coded Bias is available to stream instantly via Eventive until 22 November as part of the AMPLIFY! film festival.
- AMPLIFY! is an online film festival running November 6-22, with digital programming curated by four of the UK’s biggest film festivals: CINECITY (Brighton), Cambridge and Cornwall Film Festivals, and FilmBath.