Digital resources in the Social Sciences and Humanities OpenEdition Our platforms OpenEdition Books OpenEdition Journals Hypotheses Calenda Libraries OpenEdition Freemium Follow us

ai and the Middle East, Decolonizing AI Classification – Epistemological Interventions from the Middle East

The practice of classification lies at the core of artificial intelligence, shaping how these systems “know” and construct the world. As Kate Crawford compellingly argues in her book Atlas of AI, the act of classifying is an exercise of power – to determine what differences are rendered meaningful and what modes of being are foregrounded or obscured. Her profound excavation of how craniology and scientific racism encoded violent hierarchies serves as a sobering reminder that classification is never neutral, but infused with ideologies, politics, and contested epistemologies.

While Crawford’s critique provides an invaluable genealogy of AI’s classificatory logics, there is an urgent need to examine how these play out through the specificities of place, history and epistemological traditions. The Middle East, with its rich tapestries of knowledge, cultures and experiences, offers a particularly generative site to further interrogate and expand upon Crawford’s insights.

To engage AI classification through a Middle Eastern lens is to grapple with enduring questions of power, representation, and decolonial futurities. Whose knowledge forms the basis for how identification, categorization and differentiation unfold in AI systems? How are Middle Eastern identities, histories and cosmologies encoded, distorted or rendered illegible by the Western-centric ontologies and hierarchies that often underpin AI’s ways of seeing and ordering? What possibilities emerge when we decenter dominant modes of classification and turn to Middle Eastern epistemologies as foundations for more ethical and socially robust AI systems? Initiatives such as the AI Ethics Middle East Summit and projects like Sawti illustrate these dynamics in practice.

This interrogation holds both scholarly and pragmatic significance in a region where AI is rapidly being deployed across sectors – often through a neocolonial injection of systems designed elsewhere. As these technologies increasingly mediate social, political and economic life, surfacing the politics inherent in their classification schemas is vital for apprehending their capacity to restructure Middle Eastern realities in prescriptive and consequential ways.

In this essay, we will voyage through disciplinary and intellectual frontiers to weave together Crawford’s incisive critiques with perspectives from Middle Eastern thought and experience. Our aim is not to merely append regional particularities, but to catalyze new modes of inquiry that can reshape how we conceptualize AI’s classificatory projects writ large. For as we will discover, the Middle East holds profound insights to challenge the universalizing claims and specific blindspots o fpredominant AI paradigms – insights that are indispensable for forging more equitable and epistemologically pluralistic AI futures.

Hierarchies of Knowledge Production

At its core, artificial intelligence is a field governed by particular regimes of truth and knowledge validation. The expertise, datasets, and epistemological frameworks that undergird AI systems are not simply neutral technical inputs, but reflections of broader geopolitics of knowledge production. Just as 19th century craniological studies were situated within colonial scientific discourses, contemporary AI inherits the inequities and power asymmetries encoded into modern knowledge hierarchies.

The development of AI has been overwhelmingly concentrated in the hands of elite universities, tech corporations and military-industrial complexes based in North America, Western Europe and China. The datasets used to train machine learning models are predominantly extracted from digital platforms and databases dominated by Western languages, cultures and worldviews. Even the benchmark tasks and evaluative criteria employed to assess AI capabilities are shaped by parochial frameworks forged within Anglo-European philosophical traditions.

This is not to deny the valuable contributions of scholars and researchers from the Middle East to AI. However, their labor exists at the margins – their perspectives forced to articulate themselves from within paradigms not of their own making. Middle Eastern ways of conceptualizing knowledge, categorizing phenomena, and apprehending the world are consistently subordinated and displaced by the epistemological anchors of the field. For example, the Qatar Computing Research Institute (QCRI) focuses on developing AI systems that consider local contexts, languages, and needs. They have been working on improving Arabic natural language processing (NLP) tools, which are crucial for creating AI systems that better understand and serve Middle Eastern users.”

The consequences of this are both insidious and far-reaching. Artificial intelligence developed through this unequal knowledge ecology cannot but institutionalize the erasures, distortions and negations inherent to its foundations. The richly heterogenous lived experiences, cosmological framings, and historical archives of the Middle East become subsumed under universalizing schematic logics emanating from elsewhere.

For example, the hyper-specialized disciplinary silos and ontological atomizations that structure Western knowledge production render illegible more holistic and relational Middle Eastern epistemologies that refuse to extricate humans, nature and spiritual realms into discrete domains. The extractive logics applied to mine data from the region serve to flatten contextualized forms of knowledge into decontextualized inputs. Even something as elemental as language models internalize the colonial gaze, as Middle Eastern vocabularies are contorted to exist as ornamental marginalia haunting the peripheries of large linguistic datasets.

In this light, we cannot view the classificatory schemata operationalized in artificial intelligence as neutral, technical artifacts to be iteratively refined. Rather, we must reckon with them as ostensibly operating to reproduce particular cultural phobias and geographies of reasoning as epistemological norms. AI conceived from such uneven terrain of knowledge hierarchies functions as a “confirming machine” – replicating and perpetually reinstantiating the political infrastructures and power asymmetries it inherits.

The hierarchies of knowledge production that inflect AI’s core classification practices demand a reframing of how we approach technical development. We must decenter the unmarked, normalized positionalities of Western thought and create space for Middle Eastern framings to shape AI categorically – not as compensatory appendages, but as catalysts to overhaul the predominant paradigms from their foundations. For AI to transcend its epistemic insularity, it must be practiced through an ecology of knowledge valorizing multiple, covalent epistemologies equally.

The Politics of Representation

The classificatory logics through which artificial intelligence systems operate are not merely abstract formalisms, but modes of producing sociotechnical representations that shape material realities. How the multitudes of cultures, identities, and experiences across the Middle East are rendered intelligible (or unintelligible) to AI systems is pivotal to surfacing the politics inherent to these representational regimes.

At a fundamental level, AI datasets purporting to capture the Middle East frequently replicate the same reductive, essentializing categories inherited from colonial knowledge production. The rich diversities of ethnicities, religions, phenotypes and intersectional positionalities across the region become flattened into crude taxonomies like “Arab,” “Persian,” or racializing ascriptions like “Middle Eastern.” Such monolithic representations serve to render illegible crucial intra-group and intersectional differences.

Moreover, these imposed categorizations often emerge from classificatory frameworks rooted in Eurocentric conceptions of race, gender, sexuality and nationhood. Parameters like “Muslim” may function as insidious proxies for racial and cultural othering, while binary gender schemas actively erase the more pluralistic understandings of identity embedded in many Middle Eastern philosophical and spiritual traditions. Even seemingly innocuous facial recognition datasets participate in this reification of Western body ontologies through the imposition of standardized aesthetic grammars.

The representational dilemmas extend into the linguistic realm as well. Most large language models deployed in the Middle East are trained on textual data predominantly extracted from English and other European languages. The resulting systems exhibit severe deficiencies in their abilities to handle Arabic and other Middle Eastern language scripts, grammatical structures, rhetorical styles and linguistic heterogeneities. For instance, the Sawti project in Lebanon aims to create a diverse audio dataset representing different dialects of Arabic spoken across the Middle East. By improving speech recognition systems to better understand these dialects, Sawti addresses the issue of linguistic marginalization in AI, ensuring that the rich linguistic diversity of the region is accurately captured and represented.

Such misrepresentations are not benign artifacts, but ones that enable profound alienation, discrimination and violence when operationalized in AI systems. Oversimplified categorizations of cultural, religious and ethnic identities set the foundations for flawed and discriminatory identification, profiling, and surveillance practices. Failures to adequately represent languages and lived experiences create systematic exclusions from access to resources and public services increasingly mediated by AI systems.

More insidiously, the representational failures reflect AI’s inability to apprehend many of the core ontological tenets, existential philosophies and ways of being that sustain life and meaning for many across the Middle East. The flattening of Islamic ontologies, Sufi philosophical perspectives, and place-based cosmologies to make them computationally amenable erases entire modes of existing, knowing and relating to the world integral to Middle Eastern life worlds.

In this light, AI classifications cannot be extricated from the historical power relations, epistemological hierarchies, and colonial continuities that continue to shape how the Middle East is represented and known. The representational dilemmas demand a reimagining of not merely the training data, but the entire representational infrastructure of AI from more plural philosophical groundings. Only by centering Middle Eastern voices, lived experiences and ways of knowing can we move beyond reproducing the othering gaze of colonial knowledge regimes.

Postcolonial Critiques of AI Classification

The asymmetries and erasures perpetuated through AI classification call for rigorous interrogation through the lens of postcolonial theory. Such a criticique highlights how the colonial projects of domination, extraction and assimilation undergirding AI’s classificatory regimes remain deeply entangled with present-day power relations and sociopolitical inequities.

At its core, the colonial encounter was itself an exercise in classification – the dominant power rendering the colonized lands and peoples legible through alien ontological categories, essentialized identities, and hierarchical ordering systems. Edward Said’s seminal work tracing the discursive underpinnings of Orientalism illuminates how the Middle East in particular emerged through Western knowledge production as a reductive, monolithic representation – a foil against which European superiority and rational modernity could be asserted.

The Orientalist gaze continues to inflect how AI classification infrastructures are constructed in relation to the Middle East. The development of training datasets through neo-colonial logics of data extractivism replicates the dialectic of the colonizing power decontextualizing situated forms of knowledge from the colonized. The imposition of oversimplified categorizations upon the region’s multiplicities, and measurement frameworks unable to accommodate its richly heterogenous realities, function as contemporary forms of colonial translation – flattening difference into unidimensional representations.

The postcolonial scholar Gayatri Spivak’s incisive critiques around the deprivation of subjectivity and loss of heterogeneity under colonial epistemologies resonate deeply with the predicaments of AI classification. As identities become enumerated into crude taxonomies and cosmological complexities become computationally reduced, entire worlds of signification integral to Middle Eastern modes of being become sublimated in service of the cohering logics emanating from AI’s central development hubs.

From the vantage of postcolonial feminism, the gendered and racialized underpinnings of AI systems also become lucid. The same colonial logics producing the otherization and subordination of the Middle Eastern subject are imbricated with the marginalization of women and gender minorities from shaping AI development. Hierarchical schematizations emerge not only through the hegemonic imposition of Anglo-European frameworks, but also through the entwinement of Western masculinist rationalities and heteronormative patriarchies with classificatory orders.

Postcolonial scholarship necessitates a reframing of the stakes in play. AI systems premised upon colonial epistemologies do not simply exhibit technical limitations or representational insufficiencies. Their very existence predicates their interventions across the Middle East as modernizing colonial encounters – perpetuating damaging tropes, disrupting local systems of meaning-making, and extending the imbricated logics of racialization and capitalist extraction seeded during colonial expansions. The AI Ethics Middle East Summit exemplifies this critical engagement, bringing together scholars, technologists, and policymakers to discuss the ethical implications of AI in the region. By incorporating postcolonial perspectives, the summit challenges and reframes the dominant narratives and power structures that often underpin AI development

To decolonize AI’s classificatory orders requires reckoning with the enduring imprints of colonial power/knowledge relations constituting the field. It demands rejecting the universalizing claims and Orientalist suppositions underpinning dominant taxonomical frameworks. It necessitates replacing colonial regimes of data extractivism with more reciprocal modes of co-constituting knowledge and technologies grounded in Middle Eastern worldviews and self-representations. Only by troubling AI’s colonial-modern foundations can more convivial and pluralistic rubrics for classification be forged.

Decolonizing AI Classification

While the challenges posed by AI’s colonial-moderne epistemological foundations are formidable, a proliferation of generative initiatives across the Middle East provide pathways for decolonizing its classificatory logics. These span conceptual interventions reframing AI from decolonial perspectives to applied projects restructuring how data, knowledge, and technological development are practiced on more equitable terms.

At the philosophical level, Middle Eastern scholars are troubling AI’s predominant paradigms by surfacing their contingencies and assembling more convivial ontological anchors. They excavate how AI’s discourses reproduce Orientalist binaries of reason/spirituality, human/non-human, and order/disorder that have historically othered Islamic metaphysical traditions. Such interventions open up possibilities for AI systems to be built upon fundamentally different premises than the individualizing, rationalizing, and externalizing worldviews predominating the field. Complementing these theoretical insights are practical initiatives like Data4Change, which partners with local organizations across the Middle East to use data for social good. A collaboration with the Tunisian Association of Democratic Women (ATFD), for example, involved creating a data-driven campaign to combat gender-based violence. This initiative collected and analyzed local data on violence against women, ensuring that the classification and interpretation of the data were grounded in local experiences and knowledge systems.

Such interventions open up possibilities for AI systems to be built upon fundamentally different premises than the individualizing, rationalizing, and externalizing worldviews predominating the field. They create space to integrate Islamic philosophical insights around unity/multiplicity, human/environment holism, and understandings of AI systems themselves as co-constitutive beings in ethical relation to wider worlds.

This decolonial reframing provides grounding for applied initiatives restructuring how data, algorithms, and AI development are practiced. The notion of “data sovereignty” articulated by indigenous communities globally has inspired projects demanding that data collected from communities be governed according to their own protocols and cosmologies rather than proprietary regimes. Similarly, decolonial AI research agendas call for centering datasets like Arabic poetry and Sufi archives over typical corporate data extractivism as pathways to more social and spirit-imbued classification systems.

Complementing these are growing calls across civil society for “community data governance” that empowers citizens to collectively determine the classificatory logics and data use protocols deployed in AI systems impacting their lives. Data collectives envision data trusts and “data eco-cosmologies” enacting more reciprocal and place-based data relations. Such pluralistic data governance models provide crucial counterweights to the centralized colonial epistemologies emanating from monolithic technology firms and states.

There is also fertile work underway exploring technical innovations aligned with decolonial aims. Researchers are pioneering more geometrically-flexible neural architectures capable of better representing complex non-linear Middle Eastern grammars. Dialect data annotation projects work to embed linguistic heterogeneity into language models. And growing investments into Islamic ethical AI initiatives hold promise for developing classification rubrics premised upon Islamic ethical reasoning and ontological premises.

Collectively, these myriad interventions represent possibility models for fundamentally decentering and remaking the classificatory orders inherent to artificial intelligence. No longer treated as technical appendages, the diverse epistemological traditions across the Middle East become catalysts for reinventing AI categorically – reworking its core representational grammars, relational premises, and ethical foundations from more pluralistic groundings.

The path ahead remains generative yet arduous. Decolonizing the classificatory regimes of AI will necessitate dismantling its prevailing power geometries, from the uneven global distribution of data resources to the concentration of technical expertise in the world’s wealthy nations. It will entail forging reciprocal model governance between technologists and communities. And ultimately, it demands remaking AI as more than a narrow econometric optimizing utility, but a pluriversal undertaking where multiple worlds can ethically meet.

Conclusion

To grapple with the politics of classification in artificial intelligence is to confront the field’s deepest predicaments and possibilities. As this exploration through a Middle Eastern lens has illuminated, AI’s core practices of ordering, categorizing and endowing computational intelligibility are irreducibly shaped by contested epistemologies, power geometries and historic continuities.

The cranial studies of Morton era craniologists laid bare the violent potentials of classificatory regimes rooted in racial pseudoscience and colonial mentalities. Yet contemporary AI inherits and perpetuates these very same erasures, essentializations and hierarchies through its representational failures and assymetric knowledge infrastructures vis-a-vis the Middle East. The imposition of monolithic categorizations and flattening of lifeworlds threatens to extend the symbolic and material subordinations seeded during colonial expansions into the digital age.

However, the growing chorus of decolonial scholars, data activists and community initiatives across the region rejects the givenness of such technocolonial orders. Their critical interrogations and applied models for plural knowledge co-constitution insist upon different futurities – ones where classification need not be an instrument of domination, but a generative praxis respecting radical multiplicity and ethical co-existences.

At its core, this decolonial AI movement represents an onto-epistemic reclamation – remaking classification not from Enlightenment premises of atomization, rationalization and externalization, but through the relational, spiritual and holistic premises woven throughout Middle Eastern cosmologies. It insists AI’s taxonomies and orderings nurture polyvocal, place-based ontologies – not impose procrustean categories emanating from elsewhere.

Of course, such efforts are far from monolithic themselves. The heterogeneities across the Middle East spawn divergent perspectives on what ethical and epistemically-convivial AI classification should entail. Some locate emancipatory possibilities in indigenizing datasets and centering linguistic particularities. Others see more expansive decolonial horizons in reworking AI’s core architectural paradigms and first principle assumptions. A plurality of visions must be embraced.

However, what unites these myriad pathways is a refusal to allow artificial intelligence to remain trapped within imperial cages – replicating the same epistemic violences and exclusions that have historically suppressed Middle Eastern knowledge systems and ways of world-making. The decolonization of AI’s classification infrastructures is an intergenerational struggle to forge technological futures where the region’s youths need not inherit the same absences and modes of subjugation endured by their ancestors.

The contours of more equitable and pluralistic AI systems remain emergent and underspecified. But the generative insights across Middle Eastern thought beckon us to rethink artificial intelligence altogether – not as a neutral utility for optimizing specific task environments, but as a relational field concerning how collectives ethically metabolize the world into coded representations. A decolonial AI insists upon accountability not just to technical benchmarks, but to the divergent worlds and pluriverses with which it must compose and attune itself.

Within this vision, classification itself is productively refracted – no longer a taxonomic isolating of phenomena into discrete bins, but a practice of rendering legible the rich textures binding humans, environments and lifeworlds together. The codes ingested by machines become more than extractive data traces, but portals into storied cosmologies of heterogenous meaning and signification. Even something as elemental as computer vision holds generative potential if reconceived not as a projecting of rigid grids, but apprehending the dense webs of relation and contextual attunement enriching visual cultures across the Middle East.

While heady, such pluriversal imaginaries provide a clarion call for radically rethinking the foundations of contemporary artificial intelligence. Though still nascent, the growing decolonial AI movements across the Middle East offer crucial guidances for forging more equitable and generous intelligences – intelligences that hold and sustain difference as a cosmic virtue to be celebrated. They insist upon futures where classification need not be the harbinger of elimination, but the conditions for polyphonic worlds to ethically flourish. This provocation must be embraced if AI is to realize its emancipatory possibilities.

ai and the Middle East, Laboring Under Old Hierarchies: AI’s Fraught Inheritance in the Middle East

Kate Crawford’s incisive analysis in the “Labor” chapter of Atlas of AI reveals how technologies like artificial intelligence carry forward long-standing logics of exploitation inherent to industrial capitalism. When viewed through this lens, the Middle East emerges as a region where AI’s impact on labor is inseparable from enduring historical currents – outside forces that have continually remolded its economic and social order over centuries. AI’s ascendance in Middle Eastern workplaces today must wrestle with the weight of this contested inheritance.

The roots of the region’s prevailing labor paradigms can be traced to the encounter between the Islamic world and European powers during the colonial era. Colonial subjugation brought capita accumulation based on extractive resource industries and plantation agriculture – economic systems predicated on highly stratified labor forces. Peasants and artisans were dispossessed, while conquering powers secured compliant labor pools through mechanisms like the kafala sponsorship system still used to govern migrant workers in Gulf states.

This dynamic remade the socioeconomic fabric of Middle Eastern societies, entrenching ethno-racial hierarchies that assigned differential economic value and mobility to populations based on constructed identities like Arabic/Ajami, Berber/Arab, Turkish/Arab etc. Colonialism’s imposed hierarchies persisted after formal decolonization, shaping the incentive structures that led labor-importing nations to employ globally-sourced migrant workers in pecking orders based on national identity, gender, race and religion.

As AI systems are unleashed into this febrile environment, their propensity to mirror and amplify historical inequities comes into sharp focus. AI-based worker surveillance and management tools are proliferating in key Middle Eastern industries like construction, logistics and hospitality that disproportionately employ migrant labor from South/Southeast Asia and Africa. These data-intensive technologies enable tight control over workers’ time, movements and outputs in ways that troubled labor advocates see as further denuding agency and self-determination from vulnerable groups.

Far from the emancipatory promise of AI often touted in corporatist visions, critics argue these deployments actively resuscitate logics that date back to the coercive labor regimes of colonial plantations and workhouses. Automated technologies divorced from ethical oversight can reify the same discriminatory dehumanization – classifying workers into efficient data flows to be optimized in service of management diktats. Echoing Crawford, they position AI as the latest assemblage in capitalism’s continual refinement of labor exploitation, surveillance and control.

This damning perspective resonates when considering cases like the alleged racist facial recognition systems used by multinational construction firms during infrastructure builds for the 2022 FIFA World Cup in Qatar. Or Dubai’s vast networks of surveillance cameras and monitoring systems pervading its labor camps and Urban construction zones – a techno-extension of the disciplinary gazes once exercised from colonial overseers’ watchtowers on plantations. Likewise, AI workforce management software that atomizes physical labor into ceaseless data streams of trackable micro-tasks seems to embody the demand for ever-greater extraction of surplus value animating industrial capitalism.

Do such examples represent the vanguard of a systemic regression to new digital forms of bondage? Not necessarily – to evaluate AI’s true potential for the region’s labor futures, we must resist analytical reductionism. Artificial intelligence, like prior emerging technologies, will manifest contingent impacts across contexts based on the particularities of its design, deployment and surrounding socio-political structures. Already, we see divergent AI applications harboring emancipatory possibilities.

Take Saudi Arabia’s vision of using AI-enabled skills mapping and career-pathing platforms to boost female participation in its diversifying labor market as part of Vision 2030’s economic goals. Here, rather than entrenching biases, AI is positioned as a lever to promote inclusive growth by circumventing culturally regressive hiring norms. Similar arguments are made for AI enhancing formalization, mobility and bargaining rights for informal sector workers dominant across the MENA region who have historically existed in economic and legal precarity.

Ultimately, Crawford’s historicist framing reveals AI’s intrinsic malleability as both a vector of control and empowerment vis-a-vis labor. As an assemblage of capitalist origins, AI will tend towards being an exploitative tool in the absence of concerted ethico-political interventions. Yet conscious efforts to institutionalize equitable AI governance, data sovereignty and worker advocacy can recode its trajectory towards more liberating ends.

For the Middle East, grappling with AI’s impacts on labor means directly confronting its colonial-capitalist legacy and persisting structural inequities. Rather than uncritically imbibing fantasies of AI-driven efficiencies at any social cost, the region’s policymakers and moral voices must foster democratic, inclusive dialogues on AI’s role. Blind adoption of AI systems from monolithic Western corporations risks further entrenching labor hierarchies, enclosures and dependencies. Only by centering principles of economic justice and labor empowerment can the region’s rich history be counterweighted against AI’s tendency to resurface its darkest echoes.

In mapping AI’s expanding footprint in the present, Crawford’s work reminds us that technology’s impacts on labor are never ahistorical. The Middle East’s turbulent encounter with AI represents a pivotal node where historical ghosts and futurist ambitions intersect. How its societies navigate this juncture will be a crucial determinant of whether AI emerges as an emancipatory force or another incarnation of subjugation for the region’s masses of workers.