Brazil asked 576 of its federal and state public bodies whether they had adopted AI. Among them, 94% of judicial bodies had, against 55% of the executive, the branch that runs the services. A separate sample of 3,253 city halls put municipal use in the previous year at 27%.
It also asked where the technology sits. Among all federal bodies, 65% have generative-AI initiatives on internal processes and 29% in services delivered to the citizen. State bodies, 56% and 28%. City halls, 31% and 17%. Internal work runs between 1.8 and 2.2 times public-facing work, and that is the whole range.
The obvious objection is that this is what always happens, and it is. Administrative technology has arrived through the back office since Hollerith’s tabulators counted the 1890 census, and the first government mainframes of the 1950s went onto payroll, tax and benefit records rather than onto anything the public dealt with directly. But that history predicts the direction, not how narrow the range is. A federal ministry and a town hall of ten thousand differ in nearly everything that decides what they can buy, and land within four tenths of the same ratio.
A survey counts presence rather than effect, so none of this says whether a queue moved. Few governments ask themselves the question at all, and the same survey run twice is what would show whether the range holds.
Assorted links
-
Twenty countries that build none of the leading AI models have written rules for them anyway, and four in five of the bodies given the job predate it, most without power to act on what they find.
-
France’s inspectorates reckon unofficial chatbot use could reach as much as 40% of public agents, an estimate drawn from a study of local authorities, and none of it will appear in a procurement record.
-
The interim director of PoliceAI, the national policing centre launched in June, has asked forces in England and Wales to pause generative AI on criminal justice work, because it writes things that read as evidence and are not true.