*Lawrence Peter “Yogi” Berra, Baseball Coach
To our readers outside of the UK, you may (or may not) have noticed that we have a new Prime Minister. Whilst this week’s article is a plea to Mr. Burnham - please do not see this as a purely parochial UK view. The call for a strategic vision that covers both sides of the human-AI equation surely has universal applicability.
Dear Prime Minister,
We hope you feel that your first week in office has been positive. There has certainly been a flurry of policy activity. Bus fares got a little cheaper, there were a few quid off the ever-rising cost of energy, and pubs, clubs and music venues got a thousand pounds off their annual rates bill. And all of this, you tell us, is fully funded.
At the risk of adding to your already crowded in-tray, we wanted to throw out a few concerns that should be troubling your new Prime Ministerial AI Taskforce.
The taskforce has been asked to place AI at the center of government, transform public services and unlock growth and prosperity across the country. It also inherits an AI Opportunities Action Plan built around infrastructure, adoption and homegrown capability, alongside a commitment to equip ten million UK workers with AI skills by 2030.
Your new AI minister, Kanishka Narayan, told the FT that AI was going to be “front and centre”, the “way to reindustrialise the country” and the way for Britain to “get its mojo back” whilst at the same time rejecting “tech boosterism” (whilst seemingly indulging in some ‘Burnham boosterism’). So there is no shortage of soundbites and some signs of activity. There is nothing inherently wrong with any of this. Britain needs investment, infrastructure, capable technology companies and people who understand how to use the tools now arriving in their workplaces. But we are still not convinced there is a realistic strategy that ties it all together. It brings to mind the Scottish proverb … “If wishes were horses, beggars would ride” (as everything now has to have a Mancunian angle, you could have “If wishes would bide, beggars would ride” which is apparently a local variation found in historic Greater Manchester and Lancashire dialect)
A country can spread AI rapidly without becoming noticeably better at using it. It can train millions of people to prompt a model without preparing them to question its output. It can automate parts of public services without improving the experience of the citizen. It can increase the amount of work produced while weakening the judgement, expertise and accountability on which good work depends.
Others will advise you on compute, data centers and our venture capital infrastructure. Our interest is in the other side of the equation: what happens to human capability as AI becomes embedded in education, work and government? This does seem to be a missing link. From that perspective, there are four issues we would put before the taskforce.
1. Education must prepare people to think with AI—and without it
The commitment to provide ten million workers with AI skills is ambitious, measurable and politically attractive. But it risks reducing a profound educational challenge to a very large training target. Functional AI literacy matters. People should understand what generative AI can do, how to interact with it, where its information comes from and why its answers cannot always be trusted. But teaching people how to use AI is the easy part.
The harder task is ensuring that they retain enough knowledge, curiosity and intellectual independence to judge what it gives them. A pupil who can generate an elegant essay in seconds has not necessarily learned how to construct an argument. A student who can summarize a book without reading it has acquired an output, not an understanding. An employee who can produce a polished report may still be unable to recognize that its central assumption is wrong.
The danger is not simply cheating. It is that we begin to confuse the successful production of an answer with the development of the person producing it. Education policy therefore needs to ask a more fundamental question than whether schools and universities permit or prohibit AI: Which forms of effort remain essential to learning, even when technology can remove them?
We have spent several decades trying to make education more efficient, measurable and accessible. AI appears to offer another leap forward. It can provide personalized explanations, generate practice materials, translate complex ideas and give learners immediate support. Used intelligently, it could widen access to forms of tuition previously available only to the fortunate.
But learning is not always improved by making it frictionless. Working through a difficult text, struggling to articulate an idea and discovering why an apparently persuasive argument does not hold together are not unfortunate delays on the way to the answer. They are often the means by which understanding develops.
Guidance on the acceptable use of AI is fine as far as it goes, but we really need a serious review of curriculum, assessment and teacher development for an AI-mediated world. That review should not attempt to preserve every existing examination or essay by building increasingly elaborate systems for detecting AI use. It should ask what we now need to assess. There will need to be more emphasis on oral defense, supervised application, project work and the reasoning behind an answer. Students should sometimes be allowed to use AI and then required to explain what they accepted, what they rejected and why. At other times, they should have to demonstrate what they can do without it.
The point is not to create an artificial contest between humans and machines. It is to ensure that the machine does not conceal the absence of human understanding.
2. Do not allow AI to dismantle the route from novice to expert
The same problem continues when people leave education and enter work. Much of the discussion about AI and employment concentrates on how many jobs may disappear. That is understandable, but it misses a subtler problem. AI may change not only the number of jobs available but the process through which people learn to become good at them.
Many of the tasks now described as ideal for automation are performed by people near the beginning of their careers. Graduate consultants conduct research and prepare initial analyses. Trainee lawyers review documents. Junior accountants work through accounts. New civil servants prepare briefings. Young marketers study customers, competitors and markets.
Some of this work is repetitive. Some of it will be done faster and better with AI. But these tasks are not only producing something for the employer. They are also producing the future expert. People develop judgement by encountering the detail of a domain: the awkward exception, the misleading pattern, the customer who does not behave as expected, the apparently minor clause that changes the meaning of everything around it.
The expert sees more because, over time, they have learned what deserves attention. If AI increasingly performs the work before the novice encounters it, we may remove the experience from which that judgement develops. The immediate output will still appear. What becomes less certain is where the next generation of experienced professionals will come from. The answer is not to preserve junior drudgery as a form of character building. It is to redesign professional development as deliberately as we redesign the work itself.
Government has leverage here. Skills England, apprenticeship bodies, universities and professional associations should be asked to review how AI is changing the path from novice to expert in major occupations. Employers benefiting from publicly supported AI programs should be expected to explain how entry-level roles will change and how people will acquire the experience that automation removes. This could include supervised casework, simulation, rotation across different forms of practice, structured review of AI output and greater exposure to the messy situations that rarely fit the textbook answer.
The central question should be: If AI removes the bottom rung of the career ladder, what are we putting in its place? That question matters economically, because Britain will continue to need experienced professionals capable of making difficult decisions.
It also matters socially. The familiar bargain offered to young people was that education and effort would create a route into work, progression and some degree of security. That bargain was already becoming less convincing. AI did not create expensive housing, insecure employment or weak wage growth. But it may remove or narrow more of the entry points through which educated people expected to build a career.
And we are beginning to see the unrest. The students booing optimistic accounts of AI at American commencement ceremonies were not necessarily rejecting the technology; many will use it every day. They were reacting to the apparent absurdity of being congratulated on completing an expensive education and then told that the careers for which they had prepared may require far fewer of them. Of course there is anger …! Gen Z is pushing back on a system that appears unable to deliver a fair route into a career or professional life. Seven in ten members of the UK public say they are worried about the economic impact of AI, while six in ten believe it will eliminate more jobs than it creates.
We have previously called those most exposed to this shift the newly disadvantaged: educated and capable people who are not excluded from technology but cannot see a credible route to sharing in its gains. Education must therefore do more than prepare people to operate AI. It must preserve their ability to build expertise, exercise agency and continue progressing once AI becomes part of almost every occupation. That deserves a place in AI policy. Not as an argument for resisting the technology, but as a reason to take the development of people as seriously as the development of systems.
3. Give the public a more honest account of the transition
People are repeatedly told that AI will transform almost everything while being offered very little clarity about what that transformation will mean for them. The public conversation tends to alternate between two implausibly confident positions. One promises a surge in productivity, better services and more rewarding work. The other predicts mass unemployment and social collapse.
The honest answer is that we do not yet know how the gains and disruption will be distributed. What government can control is whether people are treated as participants in the transition or as obstacles to it. Public concern about AI should not automatically be dismissed as technophobia or a deficit of understanding. People may have entirely rational concerns about whether their work will be degraded, monitored or removed; whether their children will have access to the same career opportunities; and whether the benefits will flow mainly to those who own the systems.
The answer is not a communications campaign explaining that AI is good for them. It is to give people some agency. Employees should be involved in decisions about how AI changes their work. Professional bodies should help define where judgement and accountability must remain human. Educators should be given the time and resources to redesign learning rather than being left to improvise institution by institution.
And government should be honest that transition involves choices. Productivity gains do not automatically become shorter working weeks, higher wages, better services or more secure careers. Policy and organizational decisions determine who captures them. This does not require another grand social compact with a memorable three-word slogan. It requires a consistent principle: public support for AI adoption should also support the people and capabilities through which its value will be realized.
4. Teach organizations how to adopt AI—not merely workers how to use it
AI skills policy tends to focus on the individual worker. Learn how to use the tool. Complete the course. Acquire the digital badge. But many of the most consequential choices will be made by managers and organizational leaders. They will decide which work should be automated, where human judgement remains necessary, how performance is measured and whether employees are given time to learn or simply expected to produce more. A workforce can be perfectly capable of using AI while the organization adopts it badly.
AI may be inserted into an existing process because a supplier has demonstrated that it can be. A nominal human approval stage may be retained so that the organization can claim there is still a person in the loop. Productivity may be measured through hours saved, without asking whether people now spend those hours correcting machine-generated work or dealing with problems displaced elsewhere.
The government should therefore complement its mass skills program with a serious management-capability program. This should not be another generic course on “leading in the age of AI”. It should help organizations work through practical questions:
What problem are we trying to solve?
What knowledge does the work require?
Where does judgement enter?
What will happen when the AI is wrong but sounds convincing?
Who has the authority to challenge it?
What capabilities may deteriorate if people stop performing parts of the work?
How will we know whether the change has genuinely improved performance?
This is where the distinction between cognitive offloading and cognitive surrender becomes useful. Good adoption allows AI to carry some of the cognitive burden while people remain engaged in framing, interpreting and deciding. Poor adoption removes the burden and the engagement together.
Government cannot prescribe the design of every workplace. It can, however, shape the support it funds and the expectations attached to it. Publicly backed AI adoption programs should place organizational redesign, workforce involvement and evaluation alongside technical implementation. The question should not be how many businesses have adopted AI. It should be how many have learned to use it intelligently.
Prime Minister, Britain does need more AI adoption. But adoption should be understood as the beginning of the task, not its completion. The policy levers are available:
Use education policy to protect understanding, not simply improve access to tools.
Use skills policy to preserve the route from novice to expert.
Use business support to develop management and organizational capability, rather than counting software licenses.
Use the public sector to demonstrate how AI can expand human capability without obscuring accountability.
And give people a credible reason to believe they will have a place in the future being built around them. The countries that prosper in the AI era will not necessarily be those that deploy the most technology most rapidly. Nor will lasting advantage come from teaching everybody the same collection of prompting techniques. It will come from the quality of the judgement, imagination and institutional capability wrapped around the technology.
That is the part of the AI strategy that no model provider can sell us and no foreign government can build on our behalf.
Yours sincerely,
The Polymath Mind