Case Studies
Real-world AI deployments examined through anthropological critique
Platform Capital and the Physical AI Pipeline: Krafton's Deep-Tech Expansion in India
Ludo Robotics Ludi Vision-Language Physical AI System and Krafton CPC (Co-Playable Character) Engine
Gaming conglomerate Krafton announced a $250M expansion in India specifically targeting physical AI, robotics, and deep-tech ecosystems beyond interactive entertainment. By deploying its embodied AI stack (such as Ludo Robotics' Ludi 4B-parameter vision-language-action agent) and gaming reinforcement learning pipelines, Krafton leverages India's digital infrastructure, high mobile user density, and local engineering talent. Viewed through decolonial and feminist perspectives, this represents a classic pattern of transnational platform capitalism: treating Global South ecosystems as testbeds and training grounds for social-intelligence models and embodied robotics while core intellectual property, governance, and capital value remain centralized in South Korea.
The Global AI Governance Gap and Infrastructure Exclusion
The UN-Proposed Global AI Infrastructure and Safety Governance Framework
Following the 2026 UN Independent International Scientific Panel's assessment, this case study critiques the emerging 'Global AI Governance Framework.' Despite being labeled as global, the framework reflects a significant exclusion of Global South nations from infrastructure and decision-making processes. This creates a decolonial tension where the Global South is positioned as a site of risk assessment and data extraction rather than as a co-creator of AI safety standards, effectively maintaining a northern-centric hegemony over the definition of 'ethical AI.'
Border Imperialism and the Digital Panopticon: ICE’s AI Surveillance Arsenal
Immigration and Customs Enforcement (ICE) High-Tech Surveillance Suite
This case study examines the expansion of ICE's technological capabilities through an integrated 'arsenal' of AI-driven surveillance tools. These systems facilitate the monitoring, tracking, and management of immigrant populations using biometric data, automated data processing, and predictive analytics. From a decolonial and algorithmic justice perspective, this deployment represents a shift toward 'digital borders' and 'e-carceration,' where algorithmic systems are weaponized to enforce state power over non-citizen bodies, often operating with minimal transparency or legal oversight and reinforcing racialized patterns of policing.
Amazon Rekognition & Police Surveillance
Amazon Rekognition (facial recognition)
Amazon's cloud-based facial recognition service was marketed to and purchased by law-enforcement agencies across the US. Independent audits (MIT Media Lab, ACLU) found error rates of up to 35% for darker-skinned women vs. <1% for lighter-skinned men. The ACLU's 2018 test matched 28 members of Congress to criminal mugshots.
AI Diagnostic Tools in Global Health (Chest X-ray AI)
AI-powered chest X-ray readers (e.g., Qure.ai, Google DeepMind CXR)
AI diagnostic tools for tuberculosis and pneumonia are increasingly deployed in low-resource settings as a 'leapfrog' solution to physician shortages. Yet training datasets are overwhelmingly from North American and European hospitals, producing models that underperform on the body types, equipment quality, and TB prevalence patterns of the populations they are meant to serve — a structural mismatch framed by decolonial scholars as 'digital health colonialism.'
Uber / Lyft Algorithmic Wage Management
Dynamic pricing & dispatch algorithms (Uber, Lyft)
Gig platforms use opaque algorithmic management — surge pricing, acceptance-rate monitoring, and automated deactivation — to coordinate millions of workers classified as 'independent contractors.' Workers bear all risk (no benefits, no job security) while algorithms exert fine-grained control once associated only with employment, a tension scholars call 'algorithmic despotism.'
ChatGPT in Secondary Education
ChatGPT (OpenAI GPT-4)
The rapid adoption of ChatGPT by students has destabilised assessment, academic integrity, and pedagogical philosophy in schools worldwide. Districts have flip-flopped between banning and embracing the tool. Beneath the moral panic lie deeper questions about what education is for — skill development, critical thinking, credentialling — and who benefits when AI proxies student cognition.
Predictive Policing — PredPol / Geolitica
PredPol (now Geolitica) predictive policing software
PredPol uses historical crime data to generate 500×500-foot 'hot spot' boxes where officers should patrol. Because its training data reflects biased over-policing of Black and brown neighbourhoods, the algorithm reproduces and intensifies those patterns — a feedback loop documented by the RAND Corporation and Upturn. Los Angeles and Santa Cruz have banned it after community pressure.
Aadhaar Biometric ID System
Aadhaar — iris + fingerprint biometric identity infrastructure
Aadhaar is the world's largest biometric database (1.4 billion enrolees), linking identity to welfare entitlements, banking, and mobile services. While celebrated as a leapfrog financial-inclusion infrastructure, it has also enabled exclusion: failed biometric reads have led to denial of food rations, pensions, and healthcare, with the burden falling disproportionately on manual labourers, the elderly, and rural communities whose fingerprints degrade.
HireVue Automated Video Interviewing
HireVue AI-powered video interview screening
HireVue's platform analyses facial expressions, word choice, and tone in recorded job interviews to rank candidates before any human sees them. Used by over 100 Fortune 500 companies, it screens millions of applicants annually. Critics — including the EEOC and AI Now Institute — argue the system encodes interviewer bias into training data and discriminates by disability, neurodivergence, race, and language.
Stable Diffusion & Stolen Creative Labour
Stable Diffusion / Midjourney / DALL·E (image generation models)
Generative image models are trained on billions of images scraped from the web without consent or compensation from the artists who created them. The Glaze project (University of Chicago) and class-action lawsuits by artists against Stability AI, DeviantArt, and Midjourney frame this as large-scale intellectual property theft and an existential threat to visual creative labour.
COMPAS Recidivism Algorithm
COMPAS (Correctional Offender Management Profiling for Alternative Sanctions)
COMPAS is a risk-assessment algorithm used by US courts to predict the likelihood of reoffending. A 2016 ProPublica investigation found it misclassified Black defendants as high-risk at nearly twice the rate of white defendants, while white defendants were more often mislabelled low-risk. The system encodes historical policing patterns, making past racial disparities self-fulfilling.
Content Moderation & Arabic/African Language Bias
Meta / YouTube automated content moderation AI
Automated content moderation systems over-remove Arabic, Amharic, Tigrinya, and other non-English content at dramatically higher rates than English, often suppressing human rights documentation, political speech, and journalism from conflict zones. The Facebook Papers revealed that Meta allocated <1% of its integrity budget to non-English languages in 2021 while those regions accounted for the majority of harmful content spread.
Amazon Mechanical Turk and Ghost Labor
Amazon Mechanical Turk / Crowd-sourced data annotation platforms
Behind every AI system lies human labor — the annotators, moderators, and click-workers who label data, verify outputs, and handle tasks too nuanced for automation. Amazon's Mechanical Turk platform epitomizes this 'ghost work': underpaid, precarious labor performed by workers in the Global South who are invisible to end users. The name itself — referencing an 18th-century chess-playing automaton that secretly hid a human operator — reveals the persistent cultural desire to erase human labor from machines.
GPT and the Homogenization of Knowledge
GPT-4 / ChatGPT (OpenAI)
Large language models trained predominantly on English-language internet text risk homogenizing knowledge production — privileging Western academic conventions, flattening cultural nuance, and creating a feedback loop where AI-generated text trains future AI. In educational contexts, the technology challenges notions of authorship, originality, and what counts as 'knowing.' Its deployment in the Global South raises questions about linguistic imperialism and epistemic dependency.
Clearview AI and Mass Surveillance
Clearview AI facial recognition
Clearview AI scraped billions of photos from social media to build a facial recognition database used by police departments worldwide, often without public knowledge or consent. The system raises fundamental questions about consent, public space, and the transformation of the human face into a data point. Multiple countries have fined or banned the company, yet its technology persists in law enforcement use.
COMPAS and Algorithmic Sentencing
COMPAS (Correctional Offender Management Profiling for Alternative Sanctions)
COMPAS is a risk assessment tool used across US courts to predict recidivism. ProPublica's 2016 investigation revealed the system was significantly more likely to falsely label Black defendants as future criminals, while white defendants were more likely to be incorrectly flagged as low-risk. The case raises profound questions about how historical bias becomes encoded in predictive systems and the delegation of judicial authority to algorithms.
