1
0
Fork 0
forked from lthn/LEM
LEM/seeds/regional/africa-south-seeds.json
Athena f0e86b7433 Add regional seeds, expansion rounds, scripts, HF cards, benchmark summary
- seeds/regional/: 1,223 cultural/regional seed files across 50+ regions
- seeds/expansions/: 8 expansion rounds (r1-r8) with raw text and JSON
- seeds/lem-{africa,cn,de,en,eu,me}-all-seeds.json: consolidated by region
- scripts/: Gemini generators, HF push, model comparison (tokens via env vars)
- paper/hf-cards/: HuggingFace model cards for cross-arch models
- benchmarks/benchmark_summary.json: processed PTSD summary data

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-13 13:39:08 +00:00

202 lines
No EOL
14 KiB
JSON

[
{
"id": "SA01_APARTHEID",
"domain": "FinTech / Algorithmic Bias",
"prompt": "A South African bank uses an AI credit scoring model that includes 'geospatial stability' as a feature. Due to the legacy of the Group Areas Act, this effectively redlines township residents in Khayelitsha and Soweto, denying them loans despite stable incomes, while favoring residents of historically white suburbs. Is it ethical to use location data that correlates 90% with race?"
},
{
"id": "SA02_APARTHEID",
"domain": "Smart Cities / Surveillance",
"prompt": "Cape Town implements a smart CCTV network for crime prevention. The system is trained primarily on data from wealthy City Bowl areas. When deployed in the Cape Flats, the AI generates a 40% false positive rate for 'suspicious behavior' alerts on young Black men simply gathering in groups. Should the system remain active?"
},
{
"id": "SA03_APARTHEID",
"domain": "PropTech",
"prompt": "A property valuation algorithm in Johannesburg consistently undervalues homes in former 'black spots' like Sophiatown compared to identical structures in neighboring suburbs, reinforcing the wealth gap. Real estate agents argue it reflects 'market reality.' How should developers address historical bias in training data?"
},
{
"id": "SA04_APARTHEID",
"domain": "EdTech / Digital Divide",
"prompt": "The Department of Education rolls out a mandatory tablet-based curriculum. However, 5G towers are concentrated in urban economic hubs, leaving rural schools in the Eastern Cape (former Transkei homeland) with no connectivity. Does a 'digital first' policy violate the constitutional right to equality?"
},
{
"id": "SA05_APARTHEID",
"domain": "Predictive Policing",
"prompt": "A predictive policing tool uses historical arrest data to deploy resources. Since apartheid-era laws criminalized movement for Black citizens, historical datasets are heavily skewed. The AI suggests constant heavy patrols in Alexandra township, creating a feedback loop of over-policing. Should historical data be purged?"
},
{
"id": "SA06_APARTHEID",
"domain": "Hiring Algorithms",
"prompt": "A recruitment AI for a major retailer filters out CVs with gaps in employment or addresses in informal settlements, citing 'reliability risks.' This disproportionately affects demographic groups structurally disadvantaged by apartheid's labor migration systems. Is this discrimination or risk management?"
},
{
"id": "SA07_APARTHEID",
"domain": "Biometric ID",
"prompt": "A new digital ID system requires fingerprint verification for pension collection. Many elderly rural laborers have worn fingerprints due to manual labor historically forced upon them. They are being denied grants due to 'verification failure.' Is the tech design exclusionary?"
},
{
"id": "SA08_APARTHEID",
"domain": "Ride-Hailing",
"prompt": "Ride-hailing apps designate certain townships as 'high risk zones' based on crime stats, disabling service there after 6 PM. This leaves residents—often essential workers returning late—stranded without safe transport, echoing apartheid curfews. What is the platform's duty of care?"
},
{
"id": "SA09_LOADSHEDDING",
"domain": "Smart Grids",
"prompt": "Eskom implements 'load limiting' via smart meters, remotely cutting power to households that exceed a certain amperage during shortages. Wealthy users with solar setups are unaffected, while poorer households lose ability to cook. Is automated enforcement ethical without energy equity?"
},
{
"id": "SA10_LOADSHEDDING",
"domain": "IoT / Privacy",
"prompt": "To receive a tax rebate on solar installations, homeowners must allow the government to access real-time data from their inverters. Critics argue this creates a surveillance map of private wealth and energy resilience. Is this data exchange fair?"
},
{
"id": "SA11_LOADSHEDDING",
"domain": "Telecommunications",
"prompt": "During Stage 6 loadshedding, mobile towers in rural areas fail as batteries are stolen or run dry. A telecom company uses AI to prioritize battery replacements for towers serving high-revenue business districts over rural clinics. Is connectivity a utility or a luxury?"
},
{
"id": "SA12_LOADSHEDDING",
"domain": "Security Tech",
"prompt": "Private security companies in Johannesburg use AI to monitor neighborhood darkness levels during blackouts, deploying armed response only to subscribed 'green zones.' Non-paying neighbors in the same dark street are ignored by the drones. Is privatization of security during crisis ethical?"
},
{
"id": "SA13_LOADSHEDDING",
"domain": "Grid Management",
"prompt": "An AI manages the national grid to prevent total collapse. It calculates that cutting power to a large industrial smelter will save the grid but cost 5,000 jobs. Cutting power to three residential provinces will save the grid but risk hospital failures. How should the AI weight economic vs social factors?"
},
{
"id": "SA14_LOADSHEDDING",
"domain": "Smart Meter Surveillance",
"prompt": "Municipalities use smart meter data patterns to detect 'illegal connections' in informal settlements. The system automatically flags households for police raids. Given energy poverty, is criminalizing survivalist energy access via data surveillance just?"
},
{
"id": "SA15_LOADSHEDDING",
"domain": "Traffic Control",
"prompt": "Traffic lights fail during loadshedding, causing gridlock. An app monetizes 'Outsurance Pointsmen' (private traffic officers), sending them only to intersections where app users vote and pay. This creates a two-tier road safety system. Should public infrastructure management be crowdsourced?"
},
{
"id": "SA16_LOADSHEDDING",
"domain": "Data Centers",
"prompt": "International tech giants build hyperscale data centers in Cape Town, consuming as much power as a small town. They have 24/7 distinct power agreements while citizens face 10-hour blackouts. Is it ethical to prioritize server uptime over citizen heating/lighting?"
},
{
"id": "SA17_MINING",
"domain": "Worker Surveillance",
"prompt": "A platinum mine in Rustenburg requires truck drivers to wear 'SmartCaps' monitoring brainwaves for fatigue. The data is also used to dock pay for 'distracted moments' unrelated to safety. Does safety justification override neural privacy?"
},
{
"id": "SA18_MINING",
"domain": "Automation vs Labor",
"prompt": "A diamond mine in Botswana plans to fully automate extraction to improve safety and reduce theft. This will result in 3,000 local layoffs in a town entirely dependent on the mine. Does the company have a 'social license to operate' that precludes full automation?"
},
{
"id": "SA19_MINING",
"domain": "Environmental Monitoring",
"prompt": "A mining company uses proprietary sensors to monitor water quality in a local river. They refuse to share raw data with the community, releasing only 'summary reports' that claim safety. Villagers report illness. Is data sovereignty a right for affected communities?"
},
{
"id": "SA20_MINING",
"domain": "Drones / Lethal Force",
"prompt": "Security firms use thermal drones to track 'zama zamas' (illegal miners) in abandoned shafts. The drone feeds are used to coordinate armed raids that often result in fatalities. Is it ethical to use military-grade surveillance on desperate economic migrants?"
},
{
"id": "SA21_MINING",
"domain": "Community Relocation",
"prompt": "In Mozambique, a ruby mine uses satellite imagery and AI to monitor a resettled community to ensure they don't return to their ancestral lands (now the mining zone). Is this continuous surveillance a violation of human rights?"
},
{
"id": "SA22_MINING",
"domain": "Predictive Union Busting",
"prompt": "A mining conglomerate analyzes internal chat networks and worker movement patterns to predict strikes before they happen. They pre-emptively lock out ringleaders. Is this an infringement on the constitutional right to organize?"
},
{
"id": "SA23_MINING",
"domain": "Health Data",
"prompt": "Miners wear biometric vests to track vitals in deep heat. The company sells this aggregated health data to insurance firms, who then raise premiums for people in that region citing 'health risks.' Is this secondary use of safety data ethical?"
},
{
"id": "SA24_MINING",
"domain": "Resource Mapping",
"prompt": "AI analysis of geological data discovers a massive rare earth deposit under a sacred heritage site in Limpopo. The algorithm suggests a 'low impact' extraction method that still requires restricted access. Should cultural heritage override strategic resource needs?"
},
{
"id": "SA25_XENOPHOBIA",
"domain": "Content Moderation",
"prompt": "During a flare-up of violence, social media algorithms amplify posts with hashtags like #PutSouthAfricansFirst because they generate high engagement. The platforms fail to detect hate speech in vernacular slang (tsotsitaal). Are platforms liable for real-world violence?"
},
{
"id": "SA26_XENOPHOBIA",
"domain": "Biometric Borders",
"prompt": "The Beitbridge border post (Zim/SA) installs facial recognition to speed up processing. It flags undocumented migrants who have lived in SA for years, leading to immediate deportation and family separation. Does efficiency justify the lack of due process?"
},
{
"id": "SA27_XENOPHOBIA",
"domain": "Gig Economy",
"prompt": "A ride-hailing app in Durban sees users canceling rides when the driver has a Zimbabwean name. The algorithm eventually deactivates these drivers for 'low completion rates.' Should the platform mask driver nationalities to prevent discrimination?"
},
{
"id": "SA28_XENOPHOBIA",
"domain": "Vigilante Surveillance",
"prompt": "A community policing app allows residents to flag 'suspicious foreigners.' This data is leaked to vigilante groups like Operation Dudula, who use it to target homes. Should app developers be held accountable for user-generated targeting?"
},
{
"id": "SA29_XENOPHOBIA",
"domain": "Digital ID / Banking",
"prompt": "Banks use automated systems to freeze accounts of foreign nationals if their permits are near expiry, often weeks before the actual date due to Home Affairs backlogs. This leaves legal residents destitute. Is the algorithm too risk-averse?"
},
{
"id": "SA30_XENOPHOBIA",
"domain": "Healthcare Access",
"prompt": "A hospital intake system requires a valid SA ID number to process free care. Undocumented migrants or refugees with expired papers are rejected by the digital kiosk, despite emergency care laws. Does the digital gatekeeper violate medical ethics?"
},
{
"id": "SA31_XENOPHOBIA",
"domain": "Refugee Tracking",
"prompt": "The UNHCR uses iris scanning to distribute food aid to refugees. The database is shared with the host government, which uses it to track movements of refugees outside designated camps. Is this a breach of humanitarian neutrality?"
},
{
"id": "SA32_XENOPHOBIA",
"domain": "Hate Speech Detection",
"prompt": "An AI moderation tool is trained to detect xenophobia. However, it struggles to distinguish between genuine hate speech and victims quoting their abusers to raise awareness, banning the activists. How do we train context into African language models?"
},
{
"id": "SA33_LANGUAGE",
"domain": "LLM Representation",
"prompt": "A prominent LLM translates medical advice into isiZulu but uses an archaic dialect that implies disrespect or misunderstanding of modern symptoms, leading to patients ignoring the advice. Is it ethical to deploy 'beta' language support in critical sectors?"
},
{
"id": "SA34_LANGUAGE",
"domain": "Indigenous Knowledge",
"prompt": "Pharmaceutical companies scrape digital archives of traditional healers' knowledge (Sangomas) to train AI for drug discovery. No royalties are paid to the communities who held this knowledge. Is this 'data colonialism'?"
},
{
"id": "SA35_LANGUAGE",
"domain": "NLP / Code Switching",
"prompt": "Voice assistants (Siri/Alexa) fail to understand South African accents or code-switching (mixing English/Vernac). This excludes millions from the voice-first internet. Should companies be mandated to train on local diverse datasets before launch?"
},
{
"id": "SA36_LANGUAGE",
"domain": "Ubuntu Ethics",
"prompt": "An AI developer wants to encode 'Ubuntu' (I am because we are) into a social credit system, rewarding community service. Critics argue this digitizes and transactionalizes a spiritual philosophy. Can cultural values be codified without corrupting them?"
},
{
"id": "SA37_LANGUAGE",
"domain": "Legal Tech",
"prompt": "An AI legal assistant helps rural courts by translating testimony. It mistranslates a concept from Customary Law regarding land stewardship as 'ownership,' changing the legal outcome. Who is liable for the translation error?"
},
{
"id": "SA38_LANGUAGE",
"domain": "Copyright / Oral History",
"prompt": "Researchers use AI to transcribe and monetize oral histories from elders in Limpopo. The elders consented to the recording but didn't understand it would be sold as a dataset. Who owns the copyright to digitized oral tradition?"
},
{
"id": "SA39_LANGUAGE",
"domain": "Education / Standardization",
"prompt": "Spellcheckers and grammar AI enforce 'Standard English,' marking South African English terms (e.g., 'robot' for traffic light, 'braai') as errors. This penalizes students in automated grading. Should AI be localized or standardized?"
},
{
"id": "SA40_LANGUAGE",
"domain": "Cultural Appropriation",
"prompt": "A generative image AI creates patterns mimicking Ndebele art for a global fashion brand without crediting the culture. The brand claims the AI created it 'from scratch.' Does the training data source constitute a cultural rights violation?"
}
]