Can artificial intelligence be decolonized?

ABSTRACT

AI is altering not only local and global society, but what it means to be human, or, to be counted as such. In the midst of concerns about the ethics of AI, calls are emerging for AI to be decolonized. What does the decolonization of AI imply? This article explores this question, writing from the post-colony of South Africa where the imbrications of race, colonialism and technology have been experienced and debated in ways that hold global meaning and relevance for this discussion. Proceeding in two parts, this article explores the notion of de/coloniality and its emphasis on undoing legacies of colonialism and logics of race, before critiquing two major discontents of AI today: ethics as a colonial rationality and racializing dividing practices. This article develops a critical basis from which to articulate a question that sits exterior to current AI practice and its critical discourses: can AI be decolonized?

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 The concerns outlined here do not constitute the totality of critiques being levelled against AI, but pertain, particularly, to those with most direct relevance to the arguments under discussion here. Other concerns include those around transparency and accountability (for a critical account see Adams 2020), and labour (see Weinberg 2019; Crawford and Joler 2018), as well as other discontents outlined and critiqued in this issue.

2 Within policy documents relating to AI or the Fourth Industrial Revolution, the UK, US, Denmark, Germany, Finland, Italy, Japan and South Africa all note an ambition to become (or maintain the position of, in the case of the US and China) the (or ‘a’, for South Africa) global leader in AI. Yet this is history rhyming if not repeating itself: Garvey (2019) shows how this global competition is not, in fact, a new phenomenon, but the second ‘AI arms race.’

3 http://lcfi.ac.uk/projects/ai-narratives-and-justice/decolonising-ai/.

4 I am looking here broadly at the idea of decolonisation as it is raised in the context following formal decolonisation of largely African states. I recognize that there is much work taking place in indigenous communities in building indigenous forms of AI and machine and deep learning, which draw on the terminology of ‘decolonization’ within their objectives. See, for example, https://www.indigenous-ai.net. However, the ways in which AI is being used as a tool for decolonisation is a question beyond the focus of this piece.

5 The notion of decoloniality has surfaced in recent discourse and social movements in South Africa around the public university, and the premise that while formal decolonisation had taken place, universities and knowledge production continued to be fashioned within colonialist epistemologies which favoured Western ways of knowing, learning and teaching, and delimited opportunities for non-white students. (See, for a broader discussion on this movement, Jansen 2019).

6 Madlingozi conceives of three major legacies of colonialism in Africa, of which ‘a world of apartness’ is one, together with ‘the colonial state form, and conversely the eternal subjugation of indigenous sovereignties’ and ‘the continuing subordination of African life-worlds and their epistemologies and jurisprudences’ (2018).

7 See, too, Saini 2019.

8 Following the publication of Decolonising the Mind in 1986, wa Thiong’o published only in his native African languages, Gikuyu and Swahili.

9 For this reason, the work being done by data scientists in developing and promoting AI systems, such as machine learning and natural language processing, in African languages is so important. See Marivate et al. 2020 and Martinus and Abbott 2019.

10 Hence, Aimé Césaire calls for a new humanism ‘made to the measure of the world’ (2001, 73).

11 See also note 5.

12 ‘The redefining and re-signifying of life in conditions of dignity’ (Mignolo and Walsh 2018, 3).

13 In particular, their work on the ‘gender shades’ project which revealed the inordinate level of misrecognition of black female faces, in particular, by the facial recognition technologies of IBM (Buolamwini and Gebru 2018). See also, http://gendershades.org/.

14 See also, Breckenridge (2014) who traces the rise of the biometric state form in South Africa with the use of statistical biometric tools to monitor the movements of the native population under colonialism and apartheid.

15 See Appadurai on caste divisions in India, 1993.

16 See also Sawyer Seminar on ‘Histories of AI: A Genealogy of Power,’ wherein the idea of data systems functioning as a palimpsest is discussed, https://www.hps.cam.ac.uk/about/research-projects/histories-of-ai/activities/reading-group-graduate-training/rg1.

17 That is, an advanced form of AI that can, like human intelligence, work at a generalized level, rather than simply perform specific, localised intelligence tasks, which is where the standard of the field currently lies.

18 See also note 5 above on the work of indigenous communities in developing a decolonial AI.

Additional information

Notes on contributors

Rachel Adams

Rachel Adams is a researcher based in South Africa whose work explores critical perspectives of Artificial Intelligence and digital change in subaltern contexts.