Finland's plan to let AI 'replace' public workers gets the question backwards, argues researcher
Finland’s top civil servant on public finances says AI will effectively replace public sector staff by 2031. An AI researcher who studies the technology’s use in Finnish organisations says that framing starts from the wrong question entirely.
Text by Martti Asikainen, 23.7.2026 | Photo by Adobe Stock Photos
“AI will essentially replace humans.” That is how Juha Majanen, permanent secretary at Finland’s Ministry of Finance, described his expectation for the country’s public sector in an interview with the newspaper Helsingin Sanomat in June, as reported by Yle.
Majanen said the state, municipalities and the regional wellbeing services counties that run healthcare and social care would move to a shared AI platform by 2031, with the Ministry estimating the shift could lift public sector productivity by at least 20%. Some job losses would come through natural attrition as retiring staff go unreplaced, but Majanen acknowledged that many employees would lose their jobs directly as a result of the change.
A senior researcher at the Finnish AI Region and Haaga-Helia University of Applied Sciences, Dr Umair Ali Khan, has directly challenged that framing. Writing in an analysis published by FAIR, Khan argues that Majanen’s plan begins with the wrong question.
The issue, Khan writes, is not how many staff AI can replace, but what public problem the technology is meant to solve — a question that should be answered before deciding whether automation, better tools, redesigned processes or additional training is the right response.
Reversing the order
According to Khan, starting from a headcount target reverses the proper sequence: it lets staff reduction become the objective before the feasibility, cost, service quality, legal responsibility and impact on citizens have been properly assessed.
He argues that when a senior official states plainly that AI will replace people, it shapes how the reform is understood well before any of that groundwork has been done — employees may see it as a threat rather than an improvement effort, and private companies may start treating layoffs as proof their own AI investment is working, without having established whether it actually improves anything.
Khan notes that the timing sharpens the stakes of that framing. Statistics Finland recorded 376,000 unemployed people in May, 68,000 more than a year earlier, with the unemployment rate, unadjusted for seasonal factors, at 12.7% — the highest recorded since 1998.
Where Majanen frames the shift largely in terms of a productivity percentage, Khan argues that adoption and value are not the same thing, citing Deloitte and McKinsey research showing that AI use in Nordic organisations frequently produces isolated efficiency gains that never scale into broader outcomes, a pattern he says his own earlier analysis of Finnish companies found as well.
Applied to government, Khan writes, a public body that treats staff reductions as its main measure of success may later discover that the expertise it removed is harder to replace than the software it bought.
Where plan is on firmer ground — and where it isn't
Khan is not arguing against public-sector AI altogether. He identifies bounded, repetitive tasks — directing citizens to the right service, transcribing meetings, extracting information from applications, flagging missing paperwork — as reasonable starting points, since mistakes there are usually caught before they cause harm.
These are, in effect, the kind of tasks Majanen himself cited, such as reviewing lengthy documents currently handled by specialists.
The risk, in Khan’s account, rises sharply once AI moves beyond that kind of administrative support into decisions about rights, income, healthcare, housing or access to essential services, precisely the areas where wellbeing services counties, one of the three bodies named in Majanen’s plan, operate.
Khan notes that the EU’s AI Act already classifies systems used to assess access to essential public benefits as high-risk. He is also sceptical of human oversight as a fix on its own: a reviewer can only catch a system’s mistakes if they have the time and authority to genuinely challenge its output, he argues; otherwise the review ends up taking almost as long as doing the task without AI at all.
An unfinished plan
Khan describes the government’s proposal, in its current form, as “a high-level ambition rather than a detailed implementation programme,” pointing out that it is not yet clear which services will use AI, which decisions will stay with human staff, how performance will be measured, or how citizens could challenge a wrong decision.
The Ministry of Finance has said it will publish three separate impact scenarios this autumn, which may begin to answer some of those questions.
This article is based on an analysis piece by Dr Umair Ali Khan published by the Finnish AI Region, together with Yle’s report on Juha Majanen’s interview with Helsingin Sanomat. Helsingin Sanomat’s original interview was not directly consulted; quotes from Majanen are as translated and reported by Yle.
No comment was sought directly from the Ministry of Finance for this article.
