Technologies have politics and exercise power (Winner). The discourses in the field of critical data studies and critical algorithm studies have echoed this argument while extrapolating it to the modern technologies operating on Artificial Intelligence. The politics reflected in and enacted by the Artificial Intelligence Technologies are visible through their discriminatory and biased outputs (Nobel; Benjamin). The biases reflected by them comprise of all complex power structures seen in human society, including but not limited to race, gender, complexion, sexuality, nationality, caste, class, etc. These biases are engrained in the AI systems not only through the data and algorithms used to build them, but also through the internal and systemic biases of the humans who interact and work with them, which include both makers and users of these systems (Bucher; Roberge & Seyfert). The biased outputs of these systems are thus used to re-enact the power structures and enable the privileged, which is inherently problematic.
One of the ways of reducing these biases in AI systems is by increasing awareness about their implicit personal presence within humans. But generally, the people associated in building and using these systems are neither aware nor have access to tools which can be used to reflect on the implicit biases shown and faced by them. The project Write to Me, in its first iteration, is an attempt to address this issue. The process of this project is an invitation to reflect on the biases within ourselves (humans).
Write to Me takes a conversational approach of excavating cultural and implicit biases faced and showcased by people. These are people who are somewhere linked with the AI systems, at least in using them, if not building them. This project stemmed as an alternative to Project Implicit (https://implicit.harvard.edu/implicit/) that addresses a similar issue. Project Implicit uses generalized Association test based on quickness of pressing keys on your keyboard to answer multiple choice questions, that isn’t inclusive, and can give test takers an anxiety of not wanting to be falsely associated with biases. Thus, Write to Me decided to steer away from generalization techniques and used personalized conversational method of letter writing (inspired from Anhert & Anhert).
Write to Me asked the participants to reflect on biases experienced by and exhibited within them. The project began with a first general letter as an open call addressed to fellow academicians. After the first set of responses, they were replied to individually, in order to engage in a deeper conversation and request a second round of responses. This process gathered 17 responses from 10 respondents, collected between September to November 2020 through email. Based on the analysis of these responses a final letter was written that included the analysis and conclusion of the project and sent it out to the respondents.
The exploratory analysis techniques included two approaches:
First was quantitative analysis performed by feeding the information from letters to online tools Voyant Tools (https://voyant-tools.org/) and Two Tone (https://twotone.io/) for data visualization and data sonification based on word frequency. This was a frequency-based analysis that just reported on words most frequently associated with the prompt.
Second was a qualitative analysis performed by manually reading the letters and making a concept map using CMAP Tools (https://cmap.ihmc.us/). The concept map was color coded based on different respondents keeping their privacy intact. This was an abstraction of real-life experiences that mapped together human thought and feelings. It did not provide an insight into how biases are formed but unfolded an insight around their reflections, which were mostly summarized as questions.
Write to Me worked as a personalized tool for the respondents to address and raise questions around their own implicit and cultural biases. The results of this first iteration of project are available.
The second iteration of the project, which is still underway, is to extrapolate the project to reflect biases within algorithms. The project Write to Me, does not aim to completely solve the problem of biases and their presence in AI systems; but provides a reflective lens to increase the awareness about their presence, variety, and impact within the users of these systems. It is a collective reflective process of questioning biases as a culture and formulating connections and shared knowledge amongst people. The hope behind this project is that the awareness and questions produced by it prompt its participants to question their biases and thus reduce their infliction on AI systems.
References:
Benjamin, Ruha. Race After Technology: Abolitionist Tools for the New Jim Code. Newark, New
Jersey: Polity Press, 2019.
Bucher, Taina. If ... Then: Algorithmic Power and Politics. New York: Oxford University Press,
2018.
Noble, Safiya Umoja. Algorithms of Oppression: How Search Engines Reinforce Racism. New
York: New York University Press, 2018.
Roberge, Jonathan, and Robert Seyfert. “What are Algorithmic Cultures?” Algorithmic Cultures:
Essays on Meaning, Performance and New Technologies, edited by Jonathan Roberge, and
Robert Seyfert. London, UK: Routledge, 2016, pp.1-25.
Winner, Langdon. “Do Artifacts Have Politics?” The Whale and the Reactor: A Search for Limits in an Age of High Technology. Chicago: University of Chicago Press, 1986, pp. 19-39.
Ahnert, Ruth, and Sebastian E. Ahnert’s article “Protestant Letter Networks in the Reign of Mary I: A Quantitative Approach.” ELH vol. 28, Johns Hopkins University Press, Spring 2015.
