Equity AI, bias and access
Every new tool lands on an uneven world. Someone writes the training data, someone labels it for a few dollars an hour, someone pays for the electricity, and someone else collects the gains. AI can narrow old gaps, because a good tutor, translator or second opinion becomes cheap enough to reach a village clinic. It can also widen them, because people with fast internet, widely spoken languages and existing skills tend to benefit first and most. The picture is mixed, and it depends on choices about design, pricing, language and labor that are still being made. The questions here are old ones: who pays, who gains, and who gets a say.
Wonder
Access for everyone
Every language and community is served, and owns what it shares.
200 languages
translated by one model released free for non-commercial use, many with few written resources. Measured
92%
reported accuracy of te reo Māori speech recognition built by a Māori-led broadcaster, which keeps the data under a licence requiring its permission for any use. Reported
16% fewer
diagnostic errors, and 13% fewer treatment errors, by clinicians randomly given an AI safety-net tool across about 40,000 visits in 15 Kenyan clinics. Unreviewed preprint, partly written by OpenAI staff. Measured
4% to 21% fewer
poor people were wrongly left out when Togo targeted COVID cash aid with machine learning on phone and satellite data, compared with targeting by geography. About half of the poorest were still missed. Estimate
It takes Communities holding their own data, tools and say.
Worry
Who bears the cost
27.5% vs 15.4%
of working-age people used generative AI in the Global North vs the Global South in early 2026, and the gap is widening. Estimate
27.7% vs 22.8%
is how often language models chose the death penalty in hypothetical trials for speakers of African American English vs Standard American English. Measured
3% vs 34%
Women in one image generator's pictures of a “judge” in 2023, against the real share of US judges. Measured
$1.32 to $2 an hour
reported take-home pay of Kenyan workers labeling toxic content for an AI developer. The contractor disputes it. Reported
Up to 15 times
more tokens can be needed for the same text in some languages than in others, so speakers of those languages can pay more and fit less into a model's working memory. That is the extreme case, not the typical gap. Measured
About 10% worse
is how struggling Kenyan small-business owners did with a GPT-4 business mentor on WhatsApp, while owners already doing well may have gained over 15%. Across all 640 owners there was no measurable average effect. Measured
2.2 billion
people are still offline. In high-income countries 94% of people use the internet, compared with 23% in low-income countries. Estimate
What is being done
The hidden workforce is organizing.
Kenyan data labelers who train AI systems formed the Data Labelers Association in early 2025, with 339 members joining in its first week. It seeks fair pay, mental-health support and collective bargaining agreements.
Worth knowing
AI is spreading faster than earlier technologies.
By late 2024 about 45% of Americans aged 18 to 64 were using generative AI, a faster uptake than the personal computer or the internet at the same stage, and the authors' survey put it at about 62% by spring 2026. Fast spread can close access gaps sooner, though this is one country.
An open question
Will cheap AI expertise reach the people earlier technologies left out, or mostly speed up those already ahead?
This page does not list everything. Last reviewed October 11, 2026.
- Measured counted or experimental data
- Survey what people report or believe
- Estimate a judgment or model, often a company's own
- Forecast a projection of the future
- Reported a company's, agency's or journalist's own account