Kenya’s State Department for ICT and the Digital Economy put a 226-page draft AI policy out for comment in July. One row of its implementation matrix promises a “Fair Pay Reference Framework published and publicly available” for data annotation, content moderation and AI quality evaluation, “calibrated to international rates for equivalent roles”. It gives no rate, and does not say whose.
Statutory duty-of-care standards for these workers sit in the register of provisions needing legislation. The clauses scheduled now say the government “shall support”, “provide guidelines”, “promote”. What arrives first is a publication.
The draft’s problem statement lists these risks: “exposure to harmful content, insecure working conditions, and limited labour protections.” Publishing a rate for this data work would give a union, a court or a buyer’s audit something to compare it against: what it pays abroad. Say a Kenyan annotator earns a few dollars an hour for work paid several times more in San Francisco.
Kenya can raise a sectoral wage or licence the firms on its soil. It cannot reach the price, which sits in a contract signed abroad. Publishing the rate needs nobody else’s signature. That number would also state how large the labour-cost gap is. A union could cite it as a floor to organise around. A buyer could cite the same number as a reason to keep outsourcing there.
Assorted links
-
Vietnam’s science ministry told a conference on 14 August that AI is expected to support 60% of executive and administrative decisions across state agencies by 2030, a target it sets alongside 10,000 AI specialists and basic AI literacy for 10 million people.
-
Three researchers collected 84 potential cases across 11 jurisdictions of agents driving demand surges into public services, and report that the friction measures fastest to deploy, fees among them, often trade off equitable access.
-
A paper reading 500 model cards on Hugging Face argues they do not adequately inform a downstream user about a model’s heritage, the alignment work behind it, or how it has been observed to behave, and puts that burden on the licence and the acceptable-use policy instead.