Person working in a laboratory

Europe’s future is already at work

Leave a reply

Imagine it’s 2035 and young Europeans feel the best is yet to come.

They can confidently build a business, make a scientific discovery, learn a profession, raise a family. They live in good-quality, affordable housing and receive good care when they need it. They have every reason to believe their children will have it better than they do.

They also feel safe in a Europe with the economic, military and diplomatic strength to defend its freedoms and preserve peace. Our security rests on our own capabilities as well as dependable allies.

I’d like to contribute to this future.

The current reality

Today, it can feel some distance away. Europe faces weak productivity growth, expensive energy, pressure on public finances and growing defence needs. Our political fragmentation makes difficult decisions harder.

But there is also a great deal to build on. We have world-class scientific capabilities, skilled and educated people, successful businesses, democratic institutions and experience of cooperating across borders. We have places where people live exceptionally well.

How can we create more prosperity and extend it to more people? Draghi has of course made good and ambitious proposals.

I believe learning has an important part to play, especially if we become better at putting people’s capabilities to use.

Think about a European manufacturer whose employees learn how to use AI to improve production. They reduce waste, improve quality and win new customers. The business can pay better wages and invest in its next product.

Or a healthcare team that develops a better way of organising treatment. Patients receive care sooner, staff spend less time on administration and the same budget goes further.

Or a researcher who learns how to turn a discovery into a business, finds an experienced team and raises the capital to grow it in Europe.

Small scale, large impact

Multiply those improvements across thousands of organisations and the economic consequences could be substantial. People become able to do more. Their organisations become more effective. The resulting prosperity gives us greater capacity to invest in the things we value.

There is considerable work to do. In the European Investment Bank’s 2025 survey, 79% of EU businesses identified the availability of skilled staff as an obstacle to investment. Skills are already influencing what businesses believe they can achieve.

Some of the response starts at school. Every child should become a confident reader, develop good numeracy skills, understand scientific evidence and discover and develop their personal talents. Strong vocational routes and universities should help young people develop those capabilities further.

Learning a living

Much of the opportunity also lies with people already in work.

A person of forty-five may have another twenty-five years of contribution ahead. They bring experience, relationships and judgement. Helping them master new technology, change profession or take on greater responsibility could benefit them and many others.

Europe could become exceptionally good at helping people develop throughout their lives.

I would like that to be a European ambition. A child discovering a talent. An apprentice mastering a profession. A researcher learning how to spin-out an innovative business. Someone of fifty preparing for the next stage of their working life.

If we became exceptionally good at this, what could it mean for our prosperity?

There is an economic reason to do this, and it fits a European humanist tradition that values education, human dignity and the development of each person’s potential. This is a European perspective on a prosperous future.

AI could make useful learning much more widely available. Employers could create better opportunities to practice and apply new skills. Universities and vocational institutions could make returning to education a normal part of working life.

Mind the gap

Leaders have a particular responsibility to support the development of their people.

The OECD finds that many highly skilled workers have limited opportunities to use their capabilities. There is potential within our organisations that better management could help release.

We also need to make it easier for capable organisations to succeed.

A company should be able to grow across Europe without repeatedly overcoming different national barriers. Investment should reach promising businesses. Energy and infrastructure should support their ambitions. Governments should be able to deliver important projects within a reasonable time and budget.

Better learning achieves more when people can apply it. Successful businesses create resources for education, public services and security. Effective institutions give people confidence to invest and build.

By 2035, that confidence should be widely shared. A young person should see a fulfilling future here. Someone halfway through their career should see possibilities ahead. An entrepreneur should be able to imagine building a global business from a European home.

I have spent much of my working life in education and leading organisations. I now work with boards and leaders, and I’m increasingly interested in how developing people can contribute to a more prosperous Europe.

Over the coming years, I want to explore the organisations making this happen: what they are doing, what we can learn from them. And most importantly, I want to help more of them succeed.

If you lead or invest in an organisation where developing people could help the business grow, I would like to hear from you.

Glowing staircase leading to a bright doorway among geometric crystal shapes

Helping CEOs and investors work out what needs to change

Leave a reply

Being close to a problem can make it hard to see, for management and for boards alike.

Management knows the business in detail, which is a strength and can also be a limitation. Boards see the numbers but not always what sits behind them. Investors often have a view but not the time to test it. The result is that the first explanation for a performance gap tends to stick, even when it is only part of the story.

I’ve been a CEO and Chairman in learning and edtech, and I have seen how these conversations work from both seats. I’m now open to taking on a new strategic advisory assignment, particularly in education, publishing and information services.

What the work involves

  1. Working out what is driving the performance gap. This means looking past the headline metrics to the commercial and organisational causes underneath them.
  2. Testing the business plan, including assumptions nobody has questioned for a while. Assumptions that were reasonable at the time of the deal or the last strategy review are worth checking again.
  3. Identifying the changes that matter most. Usually that is a short list, and it helps just as much to be clear about what should be left alone.

How I work

I work as an independent, directly with the CEO and the investor or Chairman, so everyone sees the same analysis. I start with a few conversations and the numbers, and I give an honest view of what I find, including where I think the plan or the diagnosis needs to change. The aim is a clear set of priorities the team can act on.

Who it suits

It tends to suit CEOs who want an experienced sparring partner, and investors who want an independent view before or after a deal. It can also help boards that want a fresh read on a question they have been discussing for some time.

If you run or back a business where that would help, send me a message, either here, on LinkedIn, or through www.executive-mentor.com. A first conversation is easy to arrange, and it is fine if the outcome is that no outside help is needed.

← Back

Thank you for your response. ✨

AI and content businesses: what does the customer actually value?

Leave a reply
Three healthcare professionals reviewing medical data on an interactive display

I work with CEOs and investors in content, data and workflow companies, mostly in education, science, healthcare and business information.

Some of these companies tend to think in products: a journal or a database, a textbook or a course, a subscription, an assessment. That is natural, because products are what they build, price and sell. But customers don’t want a product for its own sake. They want a job done.

A teacher wants pupils to reach their learning goals, to save preparation time, and to be confident the materials are high quality and cover the curriculum. A scientist wants to know what is already established, judge the evidence and decide what to investigate next. A hospital wants its staff to have the right information at the moment they decide with a patient about a treatment. A business wants to know what a new regulation means for it and what it has to do.

The work around the content

Whatever the field, professionals go through broadly similar steps. Here is a simplified version for three of them.

StepTeacherDoctorScientist
1. Understand the objectiveDefine what pupils should learnEstablish the patient’s needs and goalsDefine the research question
2. Assess the situationCheck prior knowledge and difficultiesTake history, examine and gather findingsReview existing evidence and available resources
3. Access relevant knowledgeConsult curriculum, materials and teaching approachesConsult clinical evidence and guidanceConsult literature, methods and datasets
4. Interpret in contextJudge what this class needsInterpret findings for this patientDevelop an explanation or hypothesis
5. Decide what to doChoose the lesson and supportAgree investigations or treatmentDesign the study or experiment
6. Do itTeach and set workDeliver careConduct the research
7. Evaluate what happenedAssess understanding and progressReview response and outcomesAnalyse results and uncertainty
8. AdjustReteach or change the approachRevise the assessment or care planRefine the hypothesis or method
9. ConcludeClose the topic and identify gapsClose an episode or arrange continuing careDraw and communicate conclusions

Content and data are often used at one or more of the steps above. What happens around that is interpretation: working out what the evidence means for this pupil, this patient or this experiment, and what to do next. Until now that has been done mostly by the professional, often in consultation with the client. Professionals are skilled and know the situation, but nobody can hold the whole body of evidence, including the latest insights, in their head, and most are short of time.

AI can now help with that interpretation, at several steps on the journey. I don’t suggest it replaces the professional’s judgement at steps 4 and 5 (in healthcare, within the limits regulation sets), but it could significantly support it.

At Sanoma I held the view that Sanoma Learning was not a content company but a workflow company. I believe education is among the least digitalised professional segments, and yet one where the publisher is most embedded in the professional’s core workflow, which gives it high potential for value creation. Our focus was on student learning outcomes, and time is one outcome teachers care about. Based on teacher feedback, we estimated that for every €100 a school spent on our solutions, teachers saved time worth roughly €1,000, because we supported their core workflows. That is about a 10x return on the school’s spending.

For much of the past, the value of the information sat mostly in the content itself. In the future, I expect it to sit more in how well a solution helps the customer reach the outcome they care about, in their own circumstances. In other words, the context and interpretation around the workflow will become increasingly important.

Past value was largely determined by inputs. Future value will be determined by outcomes.

Excellent content that knows nothing about the situation in which it is used is worth less than content that fits into the work and supports a good outcome.

Take a teacher with a pupil who keeps making the same mistake with fractions. The pupil’s answers are data. Working out what she has misunderstood takes interpretation. Choosing an explanation and exercises draws on content and didactics. Setting the work and reviewing progress has to fit into the teacher’s day. That is content, data, interpretation and workflow in one small task.

Where AI could create value

I see five places, and I would welcome challenge on the list.

  • How can we improve the product?
  • Which customers or services could become viable?
  • How can we improve commercial performance?
  • How should we distribute and license our content?
  • Where could we reduce development and delivery costs?

Creating value does not guarantee capturing it.

Customers may benefit from competition, regulation, new procurement models or spending caps, so they may capture much of the value created. Existing and new competitors may cut prices. The value may end up split across many players. The commercial task is to see where these changes could make a material difference to the performance of the business, and what to do about them.

Which platform?

One of the questions I find most important commercially is where the customer will do the work. Is it in your product, or on another platform with your content inside it?

You can build a good business either way, but they are different businesses. In the second, someone else holds the customer relationship and largely sets the terms on which you earn. The more the value moves towards outcomes and context, the more that question matters.

What I think

First, many content companies have a lot to build on: subject expertise, editorial judgement, IP, brands, distribution and customer relationships. In education, pedagogy and didactics are central. In science and healthcare, it is the quality of the evidence and how it is assessed.

Why does a customer choose you? A trusted brand may help. High-quality proprietary content, including data, may be hard to replace, especially if much of the new competition is AI-slop. Deep understanding and embedding in core customer workflows can be a powerful moat. A product used every day may be hard to leave. How much does each matter, and could AI change that?

Second, AI could cut the time and cost it takes to prepare an edition, adapt material for another market or update a product when requirements change. Who will keep those savings? Unless you can sustain a lower cost position than your competitors, it’s possible much of the financial benefit might end up with customers. Competitors will make the same savings, and customers will want to see them in the price. The companies that keep more will be the ones whose products become more valuable (i.e. provide better outcomes), not just cheaper to make.

Third:

Ideas of what you could do with AI are not scarce. The hard thing is to decide what you should do. And the really hard thing is to actually do it.

There is plenty to work with. The task is to choose where AI can improve the customer’s work and the economics of the business, and to invest accordingly.

Why I’m writing this

My career has taken me from scientific research (bioinformatics) to academic publishing, then to leading Sanoma Learning, and since then to board and advisory roles in education and information businesses. AI is part of many of my conversations with CEOs and investors.

In the posts that follow I will look at what customers pay for, who gets the money when costs fall, what protects a business and who holds the customer relationship, and which AI investments deserve funding. I will also look at how people develop expertise when AI does more of the work, what CEOs still need to understand themselves, and what investors should ask before and after a deal.

I expect some of my views to change along the way. I would be interested to hear where yours differ.

A question for you. If a customer asked what they get back for every euro they spend with you, what would you tell them? Would they agree?

← Back

Thank you for your response. ✨

One is too risky. 100 is La La Land.

Leave a reply
Optimistic CEO prioritizing core business initiatives

I work with private equity sponsors and CEOs of European knowledge and workflow businesses, mostly in education. Sometimes it’s during due diligence. Sometimes it’s post-deal, as a non-executive director, adviser or mentor. The businesses differ, but some of the same questions keep coming up.

1. Which initiatives will actually create value?

CEOs are usually optimistic entrepreneurs who see opportunities everywhere. When you get into a company, you sometimes find a great number of initiatives, but little detail behind them and no clear view of which will make a difference. Which ones could really grow revenue, profit or the value of the business? How big is each opportunity? How are we going to make it happen? Often, if you can structure the plan around something like 5–10 core initiatives, you can find a workable way to create value and maintain the focus needed to drive execution. One is probably too risky. 100 is La La Land.

2. Do we have the team for the next stage?

The people who built the company have often done something remarkable. But running a larger, international business under new ownership can ask something different of them. They may still be brilliant but no longer be playing the game at which they excel.

Making changes to the team is often one of the hardest things a CEO has to do, especially when the person is a co-founder or an old friend.

Dealing with the non-performing jerk is easy, and it has usually happened already. The hard case is the person with the great attitude and work ethic, the one you play tennis with. They were a superb HR director in a small family-owned business. Now they are out of their depth in a larger international company owned by private equity.

A wrong hire in the local team hurts, but you typically see it and fix it quickly. A wrong leader in a new market can cost so much time that the market opportunity is simply lost to the investor.

3. How will we know whether it’s working?

Some businesses report a great deal and still aren’t on the right track. Many years ago, I received more than 1,000 pages of pre-reading for a regular monthly meeting. I may have written 100 of them myself and that was in the the B.C. era (Before Claude). The number of slides created and read was not a very good KPI for that particular business. Lesson learned.

A former colleague once told me that he had worked at a company where he spent a day each month writing a management report but suspected the leadership didn’t actually read it. So one month he wrote: “This month I lost all my customers to our main competitor.” There was no reaction. The month after, he stopped writing the report. Nobody noticed that either.

A measure or report is useful if it leads someone to act differently, and if it tells them early enough to change course. If an international expansion is behind plan, is the problem the market, the proposition, the sales approach or the person leading it?

4. What happens when it isn’t working?

Sooner or later something isn’t. What matters is whether people can say so. Agree who owns each initiative. Make the goals as clear as possible, review the progress rigorously and honestly and unblock the things that are standing in the way. It’s not always pretty and clean work, it’s messy and unpredictable and can be frustrating. You need to show some grit. Decide when to invest more, when to change the approach and when to stop.

How I can help you depends on the situation. Sometimes it’s knowledge of a market or experience of a situation the team hasn’t faced before. Sometimes a capable CEO simply needs someone who will listen, challenge an assumption and help them work out what they think.

If that sounds useful, get in touch.

← Back

Thank you for your response. ✨

The Anxious Generation and AI

Leave a reply

The central claim in Jonathan Haidt’s The Anxious Generation is that in little more than a decade, childhood went from mostly play-based to mostly phone-based. Children spent less time together, unsupervised and in the physical world, and far more time in digital environments designed to hold their attention. Haidt links that shift to the sharp decline in young people’s mental health since the early 2010s.

A group of children joyfully playing soccer in a park during sunset.

The decline is well documented. Whether phones caused it is still argued, since some of the evidence is correlational. Even so, I find Haidt’s case persuasive. Given the critical nature of the harm, the evidence is strong enough to act on. The burden of proof should sit with the technology, not with children, just as it sits with the pharmaceutical company and not the patient. If we wait for certainty, we will run the experiment on another generation first.

Four Foundational Harms

Haidt names four foundational harms: social deprivation, sleep deprivation, attention fragmentation and addiction. They feed each other, and together they displace the experiences through which children develop into resilient adults.

The book makes the case that the great rewiring of childhood is causing an epidemic of mental illness. I suspect the same thesis holds, at least in part, for learning. Average PISA scores peaked around 2012 and have fallen ever since, and the slide began well before the pandemic. Other explanations compete, and I would not put all of it on phones. But the timing and much of the extensive evidence is very hard to ignore.

Social media is probably the main culprit behind the four harms in girls, with pornography and online gaming also having a large impact on boys. This is why I think Haidt is right to describe the change as “from play to phone”, rather than “from play to social media”. The phone is omnipresent, and there is typically only one of them, carrying several potentially harmful agents.

AI is on the Phone

That phone is of course now also carrying AI.

Will it turn out to be one more harmful agent, or a cognitive probiotic?

AI does have the potential to deprive children of embodied social contact, to disrupt sleep, to fragment attention and to be addictive. Time and research will tell how much.

My instinct is that a general-purpose homework assistant is probably less harmful to the four foundational harms than social media, pornography or gaming. Companion chatbots, built to be a friend at midnight, are another matter, and I would not make the same assumption.

Where AI seems more likely to matter is in the cognitive health and learning of a child. Here the early evidence suggests that it depends on how it is used. In one field experiment with nearly a thousand high school maths students, those given a ChatGPT-style tutor did far better on practice problems. Once the tool was taken away, they scored 17% lower than students who never had it. A version designed to safeguard learning largely avoided that damage. The students with the plain chatbot also felt they had learned just as much as the others.

Productive Struggle

There are two very different ways a child can use AI. A fifteen-year-old asked to write a poem for homework can ask an AI to explain how a poem is built, write her own first version, and ask for feedback that sends her back to rethink it. Or she can ask it to write the poem. Only the first involves the “productive cognitive struggle” in which learning happens. In the second, she might hand in something more polished and have learned a lot less.

Adults can usually tell the two apart. Children whose self-control is still forming, and who have already learned from other apps to take the frictionless option, will find it much harder. The maths study points that way, but I know of no evidence that tests it directly. It is, though, the pattern Haidt describes, applied to thinking instead of play.

I have spent many years in businesses serving education, and technology typically reaches schools on the strength of a promise, rather than proof. Products get adopted because teachers and students like them. Engagement is easy to measure. Learning is not.

The Engagement Trap in Learning

Social media was built to maximise engagement. One of the core metrics used to sell AI to schools will likely also be engagement. But engagement can mean very different things. High engagement with productive struggle is likely to boost learning. High engagement with what I call cognitive surrender, handing the thinking over to the machine, will do the opposite. Usually you get what you measure, so we need to measure the right thing: how much thinking the child does, not how long they spend in the tool.

What I liked most about Haidt’s thinking around solutions is that he does not leave this to individual parents. One family holding back while everyone else carries on can leave their child isolated. This is a collective-action problem, and schools are where norms can change for a whole group at once. His proposals are a sensible start: no smartphone before high school, no social media before sixteen, phone-free schools, and far more independence and free play.

Fifth Proposal

I would add a fifth. Schools should decide deliberately where AI belongs and what each use is for.

Does it help the child think, or save them from thinking?
We were too slow to ask what smartphones were displacing. We should not make the same mistake with AI.

A Tale of Three Poems

Suppose the fifteen-year-old’s own poem is this:

Roses are red.
Violets are blue.
Social media sucks.
Will AI too?

Four lines, a bit rough but unmistakably hers.

Ask AI to challenge her rather than rewrite it. What did social media take from her and what AI might take? She might write:

Roses are red.
My screen glows blue.
TikTok stole my sleep.
Will AI steal my thinking too?

The thinking is still hers. AI has helped her take it a step further without taking it over.

Ask AI to write it and you might get this:

Roses Are Red, Violets Are Blue: A Nuanced Exploration of the Digital Landscape ✨
Roses are red, a vibrant crimson bloom,
Violets are blue, dispelling the gloom,
Social media, it’s true, is a double-edged sword,
But will AI, too, be thoughtfully explored?

Longer, smoother, apparently more intelligent, and empty.

A young woman sits at a desk writing in a notebook, surrounded by stacks of books. Behind her, a split scene depicts various activities: one side shows her using a phone and feeling distracted, while the other features peaceful outdoor scenes like nature and sports, symbolizing imagination and exploration.

← Back

Thank you for your response. ✨

The first year in a new leadership role

Leave a reply

I remember the week I became CEO for the first time. It all happened very quickly and unexpectedly. I was an “expert” and suddenly had to lead an organisation. I was as unprepared as I could have been. Actually, it was the best time of my life, although perhaps the organisation felt differently about it at the time.

The knowledge, skills and behaviours that help you to get the job don’t necessarily help you do it. If you’ve built your career on being the person who knows the answer, solves the problem and steps in when things get difficult, it can be surprisingly hard to realise that the new job requires a different approach. I see this in my own experience of CEO and Chairman roles, and in clients going through the same thing now.

I usually see three shifts: leading yourself, leading your team, and leading the organisation.

They’re not mutually exclusive. How you handle yourself shapes how your team experiences you. How your team performs shapes what you’re able to do.

I’ve often found limited value in leadership development delivered as a stack of modules, such as communication, strategy and marketing. The real work and learning happens around live and important issues that you are actually dealing with.

With this in mind, I’m offering a year-long programme for leaders stepping into a bigger role. All sessions are 1:1. It’s paid, not pro bono. It has two parts.

The first is eight structured sessions that work through the transition itself:

  1. Purpose and priorities
  2. The transition
  3. Leading yourself
  4. Leading your team
  5. Trust and influence
  6. Letting go
  7. Leading the organisation
  8. The leader I want to become

The second is six individual coaching sessions with no set agenda. Your CFO resigns. Your board pushes back on a call you thought was obvious. A high performer tells you they’re leaving. That’s what these sessions are for. You bring whatever’s live, and we work on it together, with experimentation and reflection between sessions where it helps.

If you’re stepping into a bigger role and think this might be useful, DM me to learn more.

← Back

Thank you for your response. ✨

PISA 2025: Reading is collapsing. AI isn’t the villain.

1 Reply

I’ve just spent a few hours reading the PISA 2025 report, and the ongoing decline in performance continues to shock.

Reading scores across the OECD have fallen from 489 to 461 points since 2015, equivalent to more than a full year of learning. Science dropped from 489 to 482. Maths fell from 485 to 463. One in five 15-year-olds across the OECD now can’t reach a basic level of proficiency in any of the three subjects, up from one in six just three years ago. After nine or ten years of formal schooling, a fifth of learners can’t follow a simple text or reliably interpret a graph. What an enormous waste of opportunity (and $100k per child).

Reading has fallen furthest and fastest of the three. Between 2018 and 2025 it dropped in three out of four education systems with comparable data. It wasn’t disadvantaged students who fell hardest, but students from wealthier families. Something is happening to an entire generation, cutting across income lines.

Why reading is falling off a cliff

The report looks at this through four inter-connected dimensions:

Students. There isn’t a uniform decline in ability. There’s a decline in stamina. Students are still fine at locating a single fact in a short passage. Where they’re losing it is on long texts that demand evaluation, reflection, pulling information together across sources. The share of “hasty readers”, kids who blast through a question and get it wrong, grew from 7% in 2018 to 11% in 2025. Fluent, accurate readers dropped by seven percentage points over the same period.

Classrooms. Teachers are spending more time managing disruption and less time teaching. TALIS 2024 puts the number at 16% of class time spent on discipline in 2024, up from 13% in 2018. Bullying and verbal abuse in schools dipped during the pandemic and have since crept back up, and the countries where student-reported threats rose the fastest are the same ones where reading scores fell the hardest.

The system. Classrooms are more mixed than they used to be. There are more refugee students, more non-native speakers, more students with special needs. Diversity isn’t the problem on its own; PISA is explicit that it doesn’t explain the decline by itself. What’s compounding it is that a lot of systems have simultaneously pulled back on using performance data to guide teaching.

Culture. Reading enjoyment has been sliding for over a decade, actually among students and parents. Fewer books, fewer print magazines, less fiction, more short, functional, digital reading. The cohort sitting the 2025 test was born around 2009. They’ve never known a world without a smartphone in the house.

Where AI and screens actually fit

The headline finding on AI chatbots is that students who don’t use them for schoolwork tend to outperform students who do. Among students using AI generally “to help me learn,” it’s the moderate users rather than the heavy users or abstainers, who come out on top.

Mean score in science, by frequency of AI use for schoolwork

The same pattern shows up with digital devices more broadly: moderate use for learning is associated with better outcomes, heavy use or leisure use during class time is associated with meaningfully worse outcomes. Distraction kills attention kills outcomes. 28% of students across the OECD say their classmates are distracted by devices in most or every science lesson, and in roughly two-thirds of countries, that distraction correlates with lower scores.

Distraction kills attention kills outcomes

The report’s framing of the use of GenAI is worth remembering and quoting directly:

“The patterns are complex, and depend a lot on how GenAI is used, but the bottom line is simple: In the same way that we do not become fit by watching sports but by doing sports, learning does not occur through the consumption of content, but as a productive cognitive struggle of the mind with new material. Where technology enables or enhances that cognitive struggle, students will advance. Where technology short-circuits the productive struggle of learning, it will undercut students’ development. If we get this right, AI becomes a scaffold, not a crutch. A tool for thinking, not a substitute for it. And education remains what it has always been at its best: a human endeavour, powered by judgment, relationships and trust.”

Actually the skills that seem to be deteriorating the most are precisely the skills that require sustained cognitive effort.

This is very close to the argument I’ve been making recently in

AI in Education: Are We Automating Learning or Augmenting Learners?,

AI and Learning: There’s a Fraction Too Little Friction, and

Cognitive Surrender.

What would I do with my own school-age children?

If I had school-age children today, I’d protect their reading time fiercely. It would be non-negotiable, daily, and mostly on paper. Students who read fiction and longer texts perform better, and those who read primarily on digital devices or rarely, perform worse, even after accounting for background.

I wouldn’t ban AI. I’d insist it comes after the struggle, not instead of it. Write the draft first, then let AI critique it, not the other way round.

Every device would come with a genuine off switch during homework and during school hours, because the correlation between leisure screen use during lessons and weaker performance is brutally clear.

I’d talk to them about what they read, because the one thing that seems to move the needle across every layer of this is whether an adult remains engaged and interested.

But none of this guarantees anything. The issue at hand here isn’t resources. The kids losing ground fastest right now are disproportionately the wealthy ones with every advantage. The issue is attention. There’s probably not much I could do to fully insulate my own children from a culture built to claim attention. But I would make every effort to set clear boundaries around undisrupted reading and engaging with them on what they have read.

← Back

Thank you for your response. ✨

Surreal figure screaming on a pier beneath swirling stars and floating papers; text reads “Cognitive Surrender.”

Cognitive Surrender

2 Replies

Human beings have always used technology to boost their life chances. Initially, we used technology to support our physical needs, through tools and later farming. About 5,500 years ago we started writing, which was built on language and may have been one of the first technologies to support our cognitive capabilities. In more recent times we have given arithmetic to the calculator, navigation to GPS, and retrieval of information to the search engine. Each time, something was lost and something was gained, and mostly we haven’t mourned the loss. I don’t miss doing long division by hand.

AI think therefore AI am

AI is different. It reaches further into the layer of reasoning itself. It can generate the hypothesis, build the argument, critique its own plan, weigh the evidence, and decide what question to ask next. Calculators did arithmetic. AI does some of the thinking.

This can happen in several ways, and the distinction between them is meaningful:

  • Cognitive assistance. “Help me solve this problem.” The thinking is still mine; I’ve brought in a tool to sharpen it.
  • Cognitive delegation. “Work this out for me and give me the answer.” I’ve handed over the task, but I chose to, and I could take it back.
  • Cognitive dependence. “I don’t really want to think this through myself; I’ll ask AI.” The choosing has started to erode. It’s becoming the default, not the decision.
  • Cognitive surrender. “AI is better at this than I am, so I’ll defer to it.” A judgement, however implicit, that the machine’s answer is simply more trustworthy than my own thinking would be.
  • Cognitive substitution. “I can no longer do this, so will get AI to do it.” This isn’t really a choice anymore. The capacity itself has begun to disappear.

One of the stories we tell ourselves about AI is that it will make people less intelligent. The more likely story is that people gradually stop exercising certain cognitive muscles because they no longer experience any need to. Yet their output keeps rising.

Cognitive Wealth

Should that trouble us? Adam Smith might have recognised the argument. In The Wealth of Nations, he showed how specialisation and the right tools could dramatically increase productivity. A worker doing one narrow task, with the right machinery, could produce far more than a generalist.

If the AI produces a better answer, why mourn the fact that we didn’t work it out ourselves?

But Smith didn’t think rising output settled the matter. In the same book, he warned that a worker confined to a few operations had little occasion to exercise his understanding and could become “as stupid and ignorant as it is possible for a human creature to become.” Smith saw both things as true. Specialisation made us more productive, and it could also make us less capable.

There is an important difference with AI, though. Smith’s worker became narrow in one part of his working life. His broader capacity for judgement remained intact. With AI, what we’re potentially outsourcing isn’t one particular task. It’s the act of reasoning itself. Losing long division doesn’t cost me my ability to think through a difficult decision. Losing the habit of thinking difficult things through might.

Use It, Or Lose It

This is no longer only a philosophical worry. Early evidence points in this direction. A 2025 survey of 666 participants found heavier AI use correlated with lower critical thinking scores (Gerlich, 2025). An MIT Media Lab study using EEG compared essay-writers working unaided, with a search engine, or with an AI assistant, and found the AI-assisted group had the weakest neural connectivity and the lowest sense of ownership over their own work (Kosmyna et al., 2025). Both build on an older, well-established finding: that offloading mental effort onto external tools shapes what a mind stays practised at (Risko & Gilbert, 2016).

Yet, this does not mean decline is inevitable. The flaw in the historical parallel is that Smith’s factory workers were cogs in a physical system they did not control. AI is not a conveyor belt; it is a collaborative partner. The erosion of our critical thinking isn’t an inevitable consequence of the technology. It depends, at least in part, on how we choose to deploy it.

Imagine two people working the same hard problem with AI. The first treats the system as a cognitive replacement: they ask a question, accept the first 30-second response without friction, and paste the output. They have chosen cognitive substitution.

The second treats the system as a cognitive sparring partner. They use the AI to generate counterarguments to their own hypotheses, to stress-test their logic, and to explore blind spots they couldn’t see alone. They might solve the problem in a fraction of the time, but their brain has been actively engaged in an intellectual heavy-lifting session. They have chosen cognitive expansion.

AI has the profound potential to strip away the mechanical, administrative cognitive load we currently carry. By automating the grunt work, it can free up mental space for higher-order judgement, curiosity, creativity, and the fundamental human task of deciding what is actually worth doing in the first place.

The real dividing line of the AI era is not between the technology and us. It is the boundary between using AI to extend cognition versus using it instead of cognition. If we treat AI as an escape from the challenge of thinking, substitution is likely. But if we use it to push our reasoning further than we could go alone, the technology could become a powerful amplifier of human agency.

If AI makes me more capable in practice, while making me less capable without it, have I actually become more capable? The answer depends entirely on which of us is steering the machine.

Notes

  • Gerlich, M. (2025). AI tools in society: Impacts on cognitive offloading and the future of critical thinking. Societies, 15(1), 6.
  • Kosmyna, N., Hauptmann, E., Yuan, Y. T., et al. (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task. arXiv preprint 2506.08872.
  • Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688.

← Back

Thank you for your response. ✨

AI and Learning: There’s a Fraction too Little Friction

2 Replies

I’ve been thinking quite a lot recently about what happens to learning when we introduce generative AI.

My concern is that if AI makes it easier to complete a task, are we also making it easier to avoid the thinking that gets us there?

That’s why I was interested in a recent paper looking at human-AI collaboration at scale.

AI, Claudius

The researchers analysed almost 250,000 real-world conversations with Claude. Rather than looking at what AI could do in a controlled experiment, they looked at what people were actually doing with it.

There’s quite a lot of learning going on.

In 67% of learning and upskilling conversations, Claude engaged in some form of teaching. This wasn’t just school or university work. Software development, health and lifestyle, and business were among the most common areas.

A pie chart illustrating that 67% of teaching is done by AI, alongside a bar graph detailing teaching methods: Mixed methods (61.1%), Step-by-step (17.4%), Explanation (12.8%), Example-driven (7.7%), Socratic questioning (0.6%), and Analogy/metaphor (0.4%). The graph also categorizes teaching domains: Health & Lifestyle (18%), Software development (18%), Academic coursework (15%), Business & entrepreneurship (12%), Creative content (10%), Legal & financial (10%), AI & data (9%), Career development (5%), and Harmful & sensitive (2%).

The most common approach was combining conceptual explanation with step-by-step guidance.

So far, so good. But teaching and learning are not the same thing. I can ask AI to explain how something works. I can also ask it to do the work for me. The two interactions might look superficially similar, but the consequences for learning are very different.

Then I read that almost half of the conversations involved some form of friction. The AI misunderstood the question, made a mistake, couldn’t do what the user wanted, or perhaps the user’s instructions weren’t clear enough.

Normally, we think of this as something technology should get better at eliminating. However, in 78.7% of these cases, users tried to recover from the problem, and when people challenged the AI’s reasoning, those attempts were successful 81.5% of the time.

This is much more interesting than the headline that AI can teach.

Resistance is not futile

Use of AI in learning often looks like this: Ask a question. Get an answer. Move on.

There is a more productive version: Ask a question. Get an answer. Notice something doesn’t make sense. Challenge it. Work through the disagreement. Correct the answer.

The second interaction looks rather like learning. There is effort, uncertainty, feedback. There is the possibility of being wrong. The learner still has to think.

Are we using AI to make learning more effective, or to make tasks more efficient?

A student who uses AI to produce a better piece of homework may well get a better piece of homework. But that doesn’t tell us whether the student has become better at the underlying task.

On the other hand, an AI system that helps a student formulate a question, tests their understanding, gives them feedback and challenges their reasoning could potentially be a very powerful learning tool.

The difference is what the human is doing.

The answer isn’t to keep AI away from learning. Quite the opposite. The potential is enormous. But we should be careful about celebrating every reduction in effort.

Effort is the point. Productive effort drives learning.

If AI removes all of it, we may discover that we have superficially made learning “easier” but also thrown the baby out with the bathwater.

AI can teach.

But can we use AI to make learning easier without taking the learning out of it.

← Back

Thank you for your response. ✨

AI in Education: Are We Automating Learning or Augmenting Learners?

3 Replies

I was struck by a recent article in the Economist about whether AI is stopping children from learning.

The headline finding is extraordinary.

Recent research, based on 30 months of data from 26,811 Chinese secondary-school students, found that after adopting generative AI, students’ homework scores increased (on average by 18%), while the time they spent doing their homework fell (by 30%).

Graph illustrating the effect of AI use on homework scores in China from January 2023 to June 2025. It compares average homework scores of pupils who never used AI versus those who did, highlighting that pupils using AI scored higher.

Graph comparing homework completion times in minutes between students using AI and those who never used AI, showing faster completion times for AI users.

So far, so good. Except that their exam scores subsequently fell by 20%.

In other words, students were producing better homework, faster, but learning less.

Graph showing exam scores comparing students using AI and those who never used AI, with average scores highlighted.

The authors found that the problem was concentrated among students whose behaviour was consistent with outsourcing their homework to AI. Those who used AI but continued to spend roughly the same amount of time on their work experienced only small learning losses.

The Power of Productive Struggle

This made me think about productive struggle.

Learning is not simply about getting the right answer. It involves building and retrieving knowledge, making connections, trying something, getting it wrong, receiving feedback, revising your thinking and trying again.

That process requires effort and involves some friction or struggle. In the case of learning, in contrast to that of work, that effort is not a bug in the process but rather a key part of it.

This is consistent with the idea of “desirable difficulties” developed by Robert and Elizabeth Bjork: conditions that make learning feel harder or less efficient in the short term can, under the right circumstances, produce more durable learning.

The Tutoring Paradox

I keep coming back to Benjamin Bloom’s famous “2 sigma” research. Bloom found remarkably large gains from one-to-one tutoring compared with conventional classroom instruction. Shouldn’t using AI (tutors) result in similar learning gains?  

That would be the Holy Grail, but maybe we’re not there yet.  Tutoring isn’t simply a matter of giving each student the right answer. The tutor continually diagnoses where the learner is, asks questions, provides feedback, adjusts the level of challenge and keeps the learner engaged in the task. There is a human relationship involved too.

This is the question for AI in education.  How can AI provide the benefits of a tutor without removing the struggle that makes the learning happen?

Augmentation Works

Interestingly, a recent randomised experiment by Zara Contractor and Germán Reyes found that students using AI to learn an unfamiliar subject performed better on subsequent unaided knowledge tests, with the gains persisting a week later. But the strongest delayed gains came from students who used AI to augment their learning. For example, by asking it to explain concepts rather than using it to automate the production of their work. The authors found that AI users shifted time away from drafting and towards reading and searching for information.

To Be or Not To Be

We should ne distinguishing between AI that does the thinking for you and AI that makes you think better.

If we optimise education for productivity (higher homework scores in less time) we may accidentally optimise away something much more important: the learning itself.

← Back

Thank you for your response. ✨