THE GREAT ATTENTION EXPERIMENT: 11 - THE EXPERIMENT CAN STILL CHANGE DIRECTION
Tomorrow morning, before millions of people have climbed out of bed, the experiment will begin again.
Phones will be reached for.
Notifications will be opened.
Feeds will be refreshed.
Recommendations will be calculated.
Advertising opportunities will be auctioned.
Children will enter digital environments selected partly according to what held their attention yesterday.
Adults will open inboxes already containing other people's priorities.
Artificial intelligence will write, summarise, recommend, persuade, answer, generate and predict.
None of this will feel like an experiment.
That may be the most important thing we have discovered.
The experiment has become ordinary.
It sits inside communication.
Education.
Work.
Shopping.
Entertainment.
Friendship.
Parenting.
Politics.
Health.
Identity.
The systems examined throughout this series are no longer unusual technologies we consciously decide to use.
They have become part of the environment in which decisions are made.
That makes the situation serious.
It does not make it hopeless.
The attention economy was built through human decisions.
Advertising models were chosen.
Metrics were selected.
Notifications were designed.
Feeds were ranked.
Stopping points were removed.
Children's profiles were created.
Workplace expectations were normalised.
Regulations were delayed.
Protections were introduced.
Every part of the system that appears inevitable was shaped by people.
That means it can be shaped again.
The experiment can still change direction.
But it will not change because people become slightly more disciplined.
It will not change because parents confiscate more phones.
It will not change because schools deliver one online-safety assembly.
It will not change because companies add another wellbeing screen people can dismiss.
It will not change because governments ban whatever technology frightens them most this year.
And it will not change through nostalgic demands that society return to a childhood that no longer exists.
The direction will change only when we stop treating attention as an individual problem occurring inside a neutral environment.
The environment is not neutral.
The people using it are not powerless.
Both truths must shape what happens next.
What the Experiment Revealed
This series began by asking how technologies created to improve human life evolved into an economic system increasingly dependent upon capturing human attention.
The first answer was historical.
Nobody began with a complete plan to build the attention economy.
The World Wide Web connected information.
Search engines helped people find it.
Social networks maintained relationships.
Video platforms allowed people to teach and learn.
Smartphones placed communication, maps, cameras, music, banking and libraries into a pocket.
Those technologies created enormous value.
Then the services required funding.
Advertising supplied it.
Digital measurement made advertising more precise.
The click connected exposure to behaviour.
The cookie connected one behaviour to another.
The profile converted behaviour into prediction.
Automated auctions placed a price upon the opportunity to influence what someone did next.
The business model changed what engagement meant.
It was no longer merely evidence that people found a product useful.
It became commercially valuable in its own right.
Human psychology supplied the mechanism.
Curiosity.
Novelty.
Social approval.
Uncertainty.
Reward learning.
Ease.
Habit.
The technology industry did not invent these human tendencies.
It learned how to design around them.
Recommendation systems accelerated the process.
They did not need to understand people deeply.
They needed to predict which item would produce a response.
The recommendation shaped behaviour.
The behaviour trained the next recommendation.
The machine learned from millions of people while each individual experienced only one version of the feed.
Children entered the system before they could understand it fully.
Their attention created advertising value.
Their preferences generated data.
Their social influence affected family spending.
Their activity strengthened platforms even when they bought nothing.
Their future commercial value could extend across decades.
Adults did not escape.
We carried the same architecture into work.
Notifications became professional obligations.
Availability became evidence of commitment.
Messages entered evenings, holidays and bedrooms.
The device stopped asking only whether something interesting was waiting.
It asked whether somebody needed us.
Finally, the companies began seeing warning signs.
Internal research.
User complaints.
Employee concerns.
Evidence of social comparison, harmful recommendations, sleep disruption, compulsive use and risks concentrated among vulnerable groups.
The evidence was never simple.
The platforms also provided connection, creativity, knowledge, support and belonging.
But complete ignorance was no longer a credible defence.
Once the risk became visible, delay became a decision.
That is where the series has brought us.
Not to the conclusion that technology is evil.
Not to the conclusion that every company acted maliciously.
Not to the conclusion that every user has been manipulated into helplessness.
To a more difficult conclusion.
Useful technologies were placed inside economic systems that rewarded the capture of additional attention.
Those systems became extremely effective.
Human beings adapted to them.
Companies adapted them further.
The benefits and harms became entangled.
We cannot keep one and remove the other with a single policy, ban or setting.
The experiment can change direction.
The work is deciding what direction should replace it.
The Objective Cannot Be Less Technology
“Less technology” sounds like a direction.
It is not precise enough.
A video call with a distant family member is technology.
So is a personalised feed designed to continue indefinitely.
A digital map is technology.
So is an advert selected through a profile of a child's behaviour.
A school learning platform is technology.
So is a game using social pressure to sell virtual items.
A smartphone used as a camera is technology.
So is the same device delivering work messages throughout the night.
The number of hours spent near a screen tells us something.
It does not tell us everything.
Two hours creating music is not psychologically identical to two hours consuming short-form video.
Thirty minutes speaking to a grandparent is not identical to thirty minutes moving through outrage.
A child using a device to photograph the woods is not having the same experience as a child receiving an endless stream of appearance-focused recommendations.
The objective cannot be to reduce every digital experience until one arbitrary number looks healthier.
The objective must be to increase agency.
Can the person understand what the system is doing?
Can they choose rather than merely continue?
Can they stop without fear, penalty or social exclusion?
Can they reject personalisation?
Can they see why something was recommended?
Can they leave without losing essential parts of their life?
Can they use the benefit without accepting unnecessary surveillance?
Can they direct attention according to what matters to them?
That is a better test than screen time alone.
Attention Creates Agency
I spent most of my life believing that progress came primarily from work ethic.
Turn up.
Work hard.
Remain reliable.
Finish what you start.
Those qualities matter.
But I eventually realised that many of the most important changes in my life had come from something else.
Agency.
Work ethic asks how hard you are prepared to work inside the system.
Agency asks whether the system itself makes sense.
Is there a better way?
What assumption are we making?
Who benefits?
Who pays the price?
Does this task deserve to exist?
What happens if we stop accepting the default?
Agency is not recklessness.
It is not ignoring authority.
It is not refusing every system because somebody else created it.
It is taking responsibility for improving the outcome.
But agency requires attention.
People without enough attention drift between demands.
People with attention notice patterns.
They see the bottleneck.
Question the instruction.
Recognise the incentive.
Observe what keeps happening.
Understand who is gaining from the arrangement.
You cannot improve a system you never notice.
Attention makes agency possible.
That principle applies far beyond personal productivity.
Children need enough attention to notice when a recommendation system is narrowing their world.
Parents need enough attention to observe patterns rather than react only to screen-time totals.
Employees need enough attention to question communication cultures rather than become more efficient at surviving them.
Regulators need enough attention to examine product architecture, not simply individual pieces of content.
Companies need enough attention to recognise the human outcomes hidden behind engagement statistics.
Society cannot reclaim agency while remaining permanently distracted by the next immediate controversy.
Individual Agency Still Matters
Structural responsibility does not remove personal responsibility.
The fact that a product was designed persuasively does not mean every action taken inside it was forced.
The fact that a workplace rewards availability does not mean every boundary is impossible.
The fact that recommendation systems shape what appears does not mean people have no influence over what they consume.
Individuals can change their environment.
Turn off non-essential notifications.
Remove devices from bedrooms.
Choose when messages are checked.
Use chronological or non-profiled feeds where available.
Introduce stopping points.
Delete histories that no longer reflect genuine interests.
Pay for services that align revenue more closely with user value.
Leave groups that manufacture urgency.
Decide which applications deserve a place on the first screen.
Create periods in which nobody else can spend their attention automatically.
These decisions are not trivial.
Defaults shape behaviour because most behaviour occurs without repeated conscious negotiation.
Changing a default changes what happens when attention is tired.
But personal action must not become the industry's escape route.
An individual can protect their own bedroom.
They cannot audit a global recommender system.
A parent can explain advertising to one child.
They cannot see every commercial pathway passing through a personalised feed.
An employee can block two hours in a calendar.
They cannot single-handedly change a company that treats delayed response as disloyalty.
Agency is real.
Power remains unequal.
A fair system does not require every individual to defeat professional behavioural design through permanent self-control.
The First Responsibility Belongs to the Designers
The people building digital systems decide which behaviour is easiest.
They decide whether the next item appears automatically.
Whether notifications arrive at night.
Whether social approval is quantified publicly.
Whether personalised advertising is enabled by default.
Whether a child can understand why something was recommended.
Whether a person encounters a natural ending.
Whether leaving requires one click or twelve.
Design is not decoration.
Design allocates power.
A protective feature hidden inside a menu is not equivalent to a protective default.
A reminder that appears after an hour of use is not equivalent to a system that asks deliberately whether another hour should begin.
A parental dashboard is not equivalent to a child account designed from the beginning around the child's best interests.
A company cannot claim neutrality after choosing the path of least resistance.
The strongest design principle should be simple:
The easiest action should not automatically be the action that creates the most revenue.
Sometimes those actions will align.
A useful recommendation benefits both user and company.
A simple payment process can remove pointless frustration.
A well-timed reminder can help someone meet a goal they chose consciously.
The concern arises when ease is used to bypass reflection.
The transaction proceeds before cost is felt.
The next video begins before choice occurs.
The child shares information before understanding where it travels.
The user accepts surveillance because the alternative requires navigating confusion.
The product retains attention through the absence of a stopping point rather than the presence of continuing value.
That is where design moves from service towards capture.
Safety Must Be Built Into the Objective
For years, companies treated safety primarily as a layer added around the core product.
The main system optimised engagement.
Safety teams removed the worst outcomes.
Content was ranked.
Moderators handled violations.
The feed continued.
The user reported problems.
The company responded.
That model is changing.
Regulators now increasingly require companies to assess how product features themselves create risk.
Under the UK's Online Safety Act regime, services likely to be accessed by children must assess risks, implement protections, keep written records and review the assessment before making significant changes.
Ofcom explicitly identifies features such as recommender systems and group messaging as risk factors and expects providers to monitor whether safety measures actually reduce harm.
This matters because it moves responsibility earlier.
The question is no longer only:
How quickly did the company remove harmful content after it appeared?
It becomes:
Why did the product make that content likely to reach the child in the first place?
A company should not be able to run the largest behavioural experiment in history and examine safety only after deployment.
Risk must be considered before the experiment reaches children.
Regulation Has Already Changed the Direction
Regulation is often criticised for moving too slowly.
The criticism is justified.
Technology develops faster than legislation.
Poor rules can protect established companies from smaller competitors.
Age-assurance requirements can create privacy risks.
Automated content moderation can remove lawful expression.
National rules can fragment a global internet.
A government frightened by one moral panic can impose restrictions producing larger harms elsewhere.
Regulation is not automatically wise because it is regulation.
It is still capable of changing incentives.
The European Union's Digital Services Act bans profiling-based targeted advertising to minors on online platforms.
It requires large platforms to provide options not based upon personalised recommendations.
It prohibits dark patterns and imposes obligations concerning systemic risk, transparency and the protection of minors.
The European Commission's 2025 guidance on protecting minors goes further.
It recommends private-by-default accounts, greater control over recommendations and the disabling by default of features associated with excessive use, including autoplay, read receipts, streaks and some push notifications.
It also calls for safeguards around AI chatbots and warns against commercial practices exploiting children's limited commercial literacy.
The UK's Information Commissioner's Office states that profiling children for targeted advertising should be off by default for most services.
Its regulatory work has already resulted in some platforms switching off personalised advertising for under-eighteens or changing their practices.
These rules are imperfect.
Enforcement will determine their value.
Companies may comply formally while preserving the substance of the old system.
Users may be placed into confusing alternatives.
Protection may vary across borders.
But the regulations demonstrate something the industry once resisted acknowledging.
Defaults, recommendation systems, advertising models and persuasive features are legitimate subjects of public policy.
They are not merely private design decisions.
Regulation Must Target Systems, Not Headlines
Public policy often responds to the most visible event.
A dangerous challenge.
A shocking video.
A child accessing pornography.
A creator promoting a harmful product.
These events matter.
They are also outputs of wider systems.
Regulation that focuses only upon individual content risks becoming an endless game of removal.
One item disappears.
Another takes its place.
The ranking system remains unchanged.
The commercial incentive remains unchanged.
The data collection remains unchanged.
The most important regulatory questions concern architecture.
What does the system optimise?
Which signals carry weight?
How does it respond to vulnerability?
Can independent researchers inspect outcomes?
Are significant product changes risk-assessed before deployment?
Are children placed into the safest reasonable setting automatically?
Can users choose a non-profiled alternative?
Does the company measure whether protection works?
What commercial effect did the safer design produce?
Ofcom stated in March 2026 that algorithms were children's main pathway to online harm and that it was issuing legally binding information requests to assess major recommender systems.
It also warned platforms that significant changes should be risk-assessed before being tested upon children.
That is the correct direction.
Not because every algorithm is dangerous.
Because distribution determines scale.
Content matters.
The machinery deciding who sees it matters more.
Transparency Must Become Useful
Companies now publish transparency reports.
They describe removed content.
Account actions.
Safety tools.
Government requests.
Advertising rules.
These reports provide valuable information.
They often tell us little about the decisions that shape ordinary exposure.
How frequently did a vulnerable child encounter harmful recommendations before the system intervened?
Did a new safety feature reduce the risk or merely reduce reports?
How much engagement was lost after nighttime notifications were disabled?
How many users regretted the time they spent?
Which groups experienced negative outcomes hidden by the average?
What did the company know before the feature launched?
Which safer alternative was tested?
Why was it rejected?
Meaningful transparency cannot require companies to reveal every line of code.
Algorithms change too quickly for source-code publication alone to explain much.
Trade secrets and security risks exist.
Personal information must be protected.
Independent scrutiny can still be created.
Secure researcher access.
Standardised risk reports.
Audits conducted by people with technical expertise.
Publication of methods and outcomes.
Records showing which executives received significant safety findings.
Clear explanations of recommendation objectives.
Evidence that safety features produced measurable improvement.
Transparency should make accountability possible.
It should not merely generate more documents than anyone can realistically examine.
Companies Must Measure What Happens After Engagement
The attention economy became powerful because engagement was easy to measure.
A click happens now.
A view happens now.
A purchase happens now.
The wider consequence may occur later.
Sleep is lost at night.
A conversation is interrupted.
Body dissatisfaction develops gradually.
A worker completes less meaningful work.
A child becomes less comfortable with boredom.
A family relationship changes.
These outcomes are harder to attribute.
Difficulty cannot become an excuse for ignoring them.
Companies already ask users whether recommendations were relevant.
Whether adverts were useful.
Whether content was interesting.
They can also ask:
Was this time well spent?
Do you wish you had stopped earlier?
Did this session help you do what you intended?
Would you like fewer recommendations on this subject?
Are you seeing this because it matters to you or because you could not stop watching it?
A company should not be judged only by how accurately it predicts behaviour.
It should be judged by whether the predicted behaviour later aligns with the person's considered judgement.
Regret is not a perfect measure.
People regret useful effort.
They enjoy things that harm them.
Self-report contains bias.
But an engagement system that never asks whether engagement was worthwhile is measuring only the outcome most beneficial to itself.
The Financial Model Must Be Allowed to Change
The advertising-funded internet cannot be replaced overnight.
Nor would universal subscription solve everything.
Paid products can still manipulate behaviour.
They can still collect data.
They still benefit from retention.
Subscriptions would exclude people unable to afford multiple services.
Publicly funded alternatives raise questions about political influence.
No model removes every incentive.
Advertising itself is not the enemy.
Contextual advertising can fund useful services without requiring extensive behavioural profiles.
Subscriptions can reduce dependence upon additional impressions.
Public-interest funding can support information that markets undervalue.
Transaction models can align revenue with completed exchanges.
Co-operatives and alternative ownership structures can give users more influence.
Hybrid systems can distribute dependency across several sources.
The important principle is not that every company must choose one approved model.
It is that commercial success should become less dependent upon increasing the amount of human attention captured.
A business should be able to earn more by becoming more useful.
Not merely by becoming harder to leave.
Children Need Protection and Preparation
The argument around children is often divided into two camps.
One wants stronger bans, higher age limits and less access.
The other argues that technology is part of modern life and children must learn to use it.
Both contain truth.
Children should not be exposed to every commercial and psychological pressure merely because the technology exists.
They also cannot be prepared for adulthood by being kept outside the digital world until the day protection disappears.
A child who has never learned how recommendation systems, adverts, social comparison or synthetic media work may enter adolescence with access and no understanding.
Protection without preparation delays vulnerability.
Preparation without protection abandons the child to systems more sophisticated than they are.
The answer is mentored exposure.
Access introduced gradually.
Rules explained.
Protections adjusted according to development.
The child given opportunities to make manageable mistakes while adults remain close enough to help interpret them.
The objective is not obedience.
It is understanding.
In our house, my daughters' access increases when the rules have been explained and they can tell me why those rules exist.
Notifications are restricted.
Communication happens on their terms rather than every message demanding an immediate response.
We discuss why online content may not reflect ordinary reality.
I know the conversation has worked when they explain the principle back in their own words, sometimes through a metaphor better than the one I used.
That is parenting.
Not policing.
Understanding matters more than compliance.
Parents Should Not Be Required to Become Full-Time Regulators
Parents still need to parent.
No regulation removes that responsibility.
Families decide which devices enter the home.
Where they are used.
Which conversations happen.
What behaviour adults model.
When access increases.
How purchases are discussed.
Whether bedrooms remain connected overnight.
But the standard advice given to parents is often insulting.
Set limits.
Be consistent.
Talk to your child.
Model healthy behaviour.
All sensible.
None acknowledges the scale of the system opposing them.
A tired parent is making decisions after work.
The platform employs engineers, researchers, designers and data scientists professionally focused upon the product.
The parent sees fragments.
The company sees patterns across millions of children.
The parent discovers a feature after behaviour changes.
The company tested it before launch.
The parent must preserve trust while enforcing a boundary.
The product can make the boundary feel like social exclusion.
Parents need realistic tools.
Age-appropriate defaults.
Clear information about recommendations.
Spending controls expressed in real currency.
Notifications showing meaningful changes rather than every action.
Simple routes for resetting profiles.
Transparent reporting.
Products designed so that safety does not depend upon a technically confident parent discovering the correct menu.
Family responsibility and corporate responsibility are not alternatives.
The more powerful participant should carry the greater design burden.
Schools Must Teach the System
Online-safety education often begins with danger.
Do not speak to strangers.
Protect personal information.
Report bullying.
Question suspicious links.
These lessons remain necessary.
They do not explain the environment fully.
Children also need to understand:
Why feeds are personalised.
How advertising-funded services make money.
Why popularity metrics affect judgement.
How influencers are paid.
What virtual currency represents.
Why a recommendation is not a neutral window into the world.
How emotional content can receive wider distribution.
What information a digital action may generate.
How synthetic images, voices and videos are created.
Why an AI answer can sound certain while being wrong.
How to pause before sharing.
How to recognise when a system is directing attention rather than supporting an intention they chose.
The UK Government's current guidance places online safety across relationships education, health education, citizenship and computing.
In July 2026, the Government also opened a national conversation on children's online lives, stating that children should be equipped with the confidence, skills and support needed to thrive through rapid technological change.
That direction needs to become practical.
Children should not only learn how to operate technology.
They should learn how technology operates upon them.
Digital Literacy Must Include Commercial Literacy
A child may know how to search efficiently and remain commercially illiterate.
They may recognise an advert but not understand why it appeared.
They may know a creator was paid and still assume the recommendation is impartial.
They may understand that a feed is personalised without realising their own behaviour is training it.
Digital literacy must therefore include economics.
Who paid?
Who benefits?
What behaviour generates value?
Why is this service free?
Which outcome does the business model reward?
What information does the company need to improve that outcome?
The objective is not to make children cynical about every product.
It is to make the transaction visible.
Advertising becomes less manipulative when its purpose can be recognised.
Recommendation becomes less mysterious when its signals can be understood.
Choice becomes more meaningful when the surrounding incentive is visible.
AI Literacy Is Attention Literacy
Artificial intelligence will intensify many of the issues examined in this series.
It can generate content faster than humans can evaluate it.
Personalise language, images and recommendations.
Create synthetic creators and companions.
Respond continuously.
Adapt tone according to the person.
Reduce the cost of producing persuasive material.
Automate experimentation.
Create an effectively unlimited supply of material designed around increasingly specific interests.
That does not mean AI will inevitably damage attention.
AI can also filter noise.
Summarise long documents.
Reduce repetitive administrative work.
Support people with disabilities.
Translate information.
Help teachers adapt explanations.
Allow individuals to move from passive consumption towards creation.
Protect attention by removing work that never deserved human concentration.
The direction depends upon what is optimised.
An AI assistant designed to help someone complete a chosen task can strengthen agency.
An AI companion rewarded primarily for keeping the person talking may reproduce the attention economy in a more intimate form.
The European Commission's guidance on protecting minors now specifically calls for safeguards around AI chatbots integrated into online platforms.
The EU AI Act also requires providers and organisations using AI systems to support the development of AI literacy among staff and others operating those systems.
The Commission's current guidance says this should include understanding what AI is, how it is used, which risks apply and how outputs should be interpreted in context.
That obligation is significant.
We have an opportunity to build literacy while the technology is still developing rather than decades afterwards.
Schools Cannot Teach AI as a Better Search Box
Children need to learn that an AI system does not possess truth merely because it speaks fluently.
It predicts and generates.
It can be useful and wrong.
Creative and derivative.
Confident and uncertain.
It can compress information while removing context.
Produce an answer without revealing the judgement required to evaluate it.
The educational question cannot be only:
How do we help pupils use AI?
It must also be:
Which parts of thinking should remain theirs?
If AI writes the first sentence, chooses the argument, summarises every source and corrects every difficulty, the pupil may complete the task without developing the attention required to understand it.
Prohibiting AI entirely would be equally inadequate.
Children will use these systems in adult life.
They need guided practice.
When should AI assist?
When should it remain closed?
How should outputs be checked?
What information should not be entered?
Who owns generated work?
What bias may be hidden?
What happens to intellectual confidence when an answer is always available immediately?
UNESCO's AI competency frameworks for students and teachers emphasise human-centred judgement, ethics, technical understanding and responsible use rather than tool operation alone.
That is the right educational frame.
AI literacy is not simply learning better prompts.
It is learning when not to surrender the thinking.
Businesses Must Stop Treating Attention as Free
The attention economy does not end at the consumer platform.
Every unnecessary meeting spends attention.
Every broad email creates work.
Every after-hours message enters somebody's recovery.
Every unclear decision generates another round of communication.
Every tool added to solve communication can create another channel requiring monitoring.
Businesses account carefully for money.
They rarely account for collective attention with the same seriousness.
A one-hour meeting involving twelve people is not a one-hour cost.
It consumes twelve hours before preparation, recovery and follow-up are considered.
A message that takes thirty seconds to send may interrupt ten people completing more valuable work.
The sender receives convenience.
The cost is distributed.
Organisations should treat attention as a finite operational resource.
Define genuine urgency.
Create response expectations.
Protect periods for concentrated work.
Review recurring meetings.
Allow asynchronous communication to remain asynchronous.
Measure outcomes rather than visible responsiveness.
Do not praise the person who is permanently available while quietly making them responsible for everyone else's lack of planning.
The objective is not to make businesses silent.
It is to make interruption deliberate.
AI Could Improve Work or Flood It
Artificial intelligence may remove administrative load.
Summarise meetings.
Draft routine communication.
Organise information.
Automate repetitive reporting.
That could return attention to people.
It may also make communication almost costless to produce.
If every employee can generate five polished reports, twenty personalised emails and endless meeting summaries, the organisation may create more information than anyone can absorb.
Production becomes easier.
Attention does not expand.
AI may therefore increase the importance of restraint.
The question cannot be:
How much more content can we produce?
It must be:
Which information deserves to exist?
A machine capable of generating an answer instantly makes human judgement more valuable, not less.
Somebody must decide whether the answer was needed.
Whether the meeting should happen.
Whether the report should be read.
Whether automation removed effort or merely multiplied output.
Businesses adopting AI now should not measure success only through the amount generated or time saved by the sender.
They should measure the attention imposed upon everyone downstream.
Adults Must Model the Future They Want Children to Enter
Children notice contradiction.
A parent who tells them to stop checking while answering work messages at dinner teaches a stronger lesson than the words.
A teacher who warns about distraction while every lesson is interrupted by digital systems demonstrates the same conflict.
A leader who speaks about wellbeing and emails throughout the night defines the real culture.
Adults do not need to model perfection.
That would be dishonest.
They can model awareness.
“I checked that automatically. I didn't need to.”
“This message can wait.”
“I am putting the phone away because this conversation matters more.”
“I use this application too much when I am tired.”
“I got distracted. I am starting again.”
“Daddy got that wrong.”
Children do not require adults who never struggle.
They need adults who recognise the struggle and make visible choices.
That is how agency is taught.
We Need Better Questions
The attention debate is full of questions with easy emotional answers.
Should children have phones?
Is social media harmful?
Are algorithms addictive?
Should schools ban devices?
How much screen time is too much?
The more useful questions are harder.
Which child?
Using which product?
For what purpose?
Under which defaults?
At what developmental stage?
Receiving which recommendations?
With what support?
Funded through which model?
Producing which benefit?
Creating which risk?
Compared with what alternative?
Who has the power to change it?
Who gains if nothing changes?
Those questions resist moral panic.
They also resist commercial reassurance.
They force the discussion towards systems, evidence and context.
That is where meaningful change begins.
Success Will Not Mean the End of Distraction
Human beings will continue wasting time.
Avoiding difficult tasks.
Seeking novelty.
Comparing themselves with others.
Buying things they later regret.
Interrupting conversations.
Choosing entertainment over intention.
No regulation will remove human inconsistency.
No design code will create perfect self-control.
No school curriculum will make children immune to persuasion.
The objective is not a future in which nobody becomes distracted.
It is a future in which distraction is not industrially amplified without meaningful limits.
A future in which the system does not know more about a child's vulnerability than the child understands about the system's intention.
A future in which adults can use connected tools without work claiming every available space.
A future in which AI serves chosen goals rather than becoming another machine rewarded for prolonging engagement.
A future in which people remain imperfect but possess a fairer opportunity to choose.
How Would We Know the Direction Had Changed?
The experiment would be changing direction if companies celebrated users accomplishing their goals and leaving satisfied.
If safety teams could delay launches without fighting revenue targets alone.
If independent researchers could test platform claims using protected data.
If children entered private, non-profiled and age-appropriate environments automatically.
If commercial content looked commercial.
If virtual spending felt like real spending.
If feeds offered understandable choices between personalisation, chronology and explicit interests.
If schools taught children why systems behave as they do.
If parents were supported rather than blamed.
If workplaces rewarded contribution rather than permanent responsiveness.
If AI reduced noise instead of generating more of it.
If people could stop using a product without losing access to essential relationships, memories or services.
If time well spent mattered alongside time spent.
None of these changes requires the end of technology.
They require a different definition of success.
Hope Must Be Based on Evidence
Hope is not the belief that companies will suddenly abandon profit.
Nor is it faith that governments will regulate wisely.
Hope comes from recognising that the direction has already changed in places.
Personalised advertising to minors has been restricted.
Non-profiled feed options have become legal requirements for the largest European platforms.
Children's risk assessments are now required before significant product changes in the UK.
Private-by-default teen accounts have become more common.
Nighttime notifications have been limited.
Autoplay and other engagement features are now being questioned openly by regulators.
AI literacy has become a formal obligation within the European regulatory framework.
Governments are consulting children directly about the online world they want.
These measures may prove insufficient.
Some will fail.
Some will create unintended consequences.
Some companies will comply reluctantly.
Some regulators will be outmanoeuvred.
But the changes establish that the old direction was never inevitable.
Pressure changes behaviour.
Evidence changes law.
Law changes defaults.
Defaults change experience.
The system responds when the cost of continuing becomes greater than the cost of changing.
No Single Group Can Solve It
Companies cannot regulate themselves entirely.
Governments cannot design every product.
Schools cannot replace parents.
Parents cannot audit algorithms.
Children cannot carry the responsibility for defeating systems built by adults.
Individuals cannot repair workplace culture alone.
Researchers cannot reach conclusions without data.
Regulators cannot understand consequences without researchers.
Businesses cannot adopt AI safely without literacy.
The problem is distributed.
The response must be distributed too.
That can feel unsatisfying.
People prefer one villain.
One solution.
One law.
One rule for phones.
One age at which everything becomes safe.
The attention economy did not emerge from one decision.
It will not be redirected by one.
Change will come through accumulated pressure.
Better defaults.
Stronger literacy.
Different metrics.
Clearer duties.
More responsible parenting.
Less careless work design.
Independent scrutiny.
Commercial models less dependent upon endless engagement.
Millions of deliberate individual choices.
No single intervention will appear revolutionary.
The existing system was built from small decisions too.
We Do Not Need Permission to Begin
Regulation matters.
Corporate reform matters.
Education matters.
People do not need to wait for all of them.
Every family can make attention visible.
Every school can explain one recommendation system more honestly.
Every manager can remove one unnecessary meeting.
Every business can define one boundary around response.
Every product team can challenge one engagement metric.
Every adult can create one part of the day that no platform, employer or advertiser can access automatically.
Small changes are sometimes dismissed because they do not solve the whole problem.
They do not need to.
Agency is not the belief that one person controls everything.
It is the refusal to surrender responsibility for the part they can influence.
The Bells Will Continue Ringing
Technology will continue developing.
Messages will arrive.
Work will require attention.
Children will seek entertainment.
Companies will advertise.
Creators will compete for audiences.
Artificial intelligence will produce more information than any human being could absorb.
The bells will not stop.
Silence cannot become the default condition of modern life.
Nor should it.
Some bells announce people we love.
Important information.
Emergency.
Opportunity.
Beauty.
Humour.
Learning.
Connection.
The objective is not to smash every bell.
It is to remember that ringing does not create importance.
A signal can be immediate and meaningless.
Something important can remain quiet.
The greatest danger was never that technology could make noise.
It was that people might forget they were allowed to decide when to listen.
Final Thought
The attention economy was not built because humanity chose distraction over meaning.
It was built because useful technologies required money.
Advertising rewarded attention.
Measurement made attention visible.
Psychology made it predictable.
Algorithms made the process personal.
Scale made small effects enormous.
Children entered before they understood.
Adults built obligation around permanent connection.
Companies learned enough to recognise risk.
Governments began responding only after the system had become embedded in ordinary life.
That is the experiment.
Not one application.
Not one company.
Not one generation.
An entire society discovering what happens when limited human attention becomes measurable, tradeable and continuously contested.
The experiment produced extraordinary benefits.
It also revealed a flaw in the direction.
The systems became better at predicting what would hold attention than society became at deciding what deserved it.
That direction can change.
Companies can design for considered value rather than automatic continuation.
Governments can regulate architecture rather than chase symptoms.
Schools can teach systems rather than issue warnings.
Parents can mentor rather than merely restrict.
Businesses can protect attention rather than consume it carelessly.
Individuals can question defaults and rebuild agency.
Artificial intelligence can reduce unnecessary demand rather than multiply it.
None of this will happen perfectly.
None of it will happen all at once.
It does not need to.
The first version of the attention economy was not planned in one room.
The next version does not need to be either.
It requires enough people, in enough places, asking a better question.
Not:
How do we make people spend less time using technology?
But:
How do we build technology that leaves people more capable of deciding how their lives should be spent?
The experiment can still change direction.
But this time, we cannot leave the direction to accident.
Sources & Research Gaps
Principal Sources
Adam Fox, Agency and Work Ethic Biography
Adam Fox's agency reference document distinguishes work ethic from agency.
It defines work ethic as working hard inside a system and agency as taking responsibility for improving the system itself.
It also establishes the connection between attention and agency:
- people without attention drift;
- people with attention notice;
- agency requires attention;
- systems cannot be improved if their patterns are never noticed.
The document informs the central philosophical framework of this essay.
Adam Fox, Parenting and Lived-Experience Reference
The parenting material describes an approach based upon controlled, explained exposure rather than either total prohibition or unrestricted access.
It records:
- limited and age-appropriate device access;
- notifications disabled for a child's messaging;
- communication without pressure to respond immediately;
- access increasing only after rules are understood;
- discussion rather than accusation;
- treating understanding as more important than compliance.
This source supports the essay's argument for mentored exposure and parenting through explanation rather than policing.
Ofcom, Protection of Children Duties Under the Online Safety Act
Ofcom's guidance states that relevant services likely to be accessed by children must:
- conduct children's risk assessments;
- implement protections;
- retain records;
- monitor the effectiveness of safety measures;
- review risks regularly;
- reassess before significant changes to design or operation.
Ofcom identifies recommender systems and group messaging among features capable of increasing children's exposure to harmful content.
Ofcom, Online Safety Industry Bulletin, March 2026
Ofcom stated that recommendation algorithms were children's main route to harmful online material.
It announced legally binding information requests to major platforms and emphasised that significant product changes must be assessed before being deployed to children.
European Union, Digital Services Act
The Digital Services Act introduces:
- a prohibition on profiling-based targeted advertising to minors;
- non-profiled feed options on very large platforms;
- restrictions on dark patterns;
- advertising transparency;
- duties concerning systemic risk and the protection of minors.
The European Commission reported in May 2026 that platforms had altered feeds, advertising systems and user controls in response to the Act.
European Commission, Guidelines on the Protection of Minors
The Commission's 2025 guidance recommends:
- private-by-default accounts;
- safer recommender systems;
- prioritising explicit child feedback over inferred behavioural signals;
- disabling by default features associated with excessive use;
- controls over contact and content sharing;
- safeguards around AI chatbots;
- measures preventing exploitation of children's limited commercial literacy.
The guidelines provide a reference for assessing compliance with the Digital Services Act but are not themselves an automatic guarantee of compliance.
Information Commissioner's Office, Children's Code
The ICO states that profiling children for behavioural advertising should be switched off by default for most services.
Its 2025 progress update records changes by several platforms following regulatory engagement, including commitments to disable personalised advertising for users under eighteen.
UK Government, Teaching Online Safety in Schools
Current government guidance places online safety across existing curriculum subjects including relationships education, health education, citizenship and computing.
The guidance establishes a foundation for digital safety but does not by itself guarantee comprehensive teaching about commercial models, recommendation systems, attention or artificial intelligence.
UK Government, Growing Up in the Online World
The national conversation launched in July 2026 states an ambition to make children's online lives safe and healthy while giving them the confidence, skills and support required to thrive amid rapid technological change.
The consultation explicitly seeks contributions from children, parents and experts.
UNESCO, AI Competency Frameworks
UNESCO's frameworks for students and teachers treat AI literacy as a combination of technical understanding, ethics, human-centred judgement and responsible participation.
They are intended to help education systems prepare pupils and educators to act as responsible users and co-creators rather than passive consumers of AI.
European Commission, AI Literacy Under the AI Act
Article 4 of the EU AI Act requires providers and organisations deploying AI systems to support the development of AI literacy among relevant staff and other people using those systems on their behalf.
The Commission's guidance states that literacy should be adapted to context, technical knowledge and risk.
It should include an understanding of how AI is used, its opportunities, its limitations, how outputs should be interpreted and which legal or ethical risks require mitigation.
Enforcement of the relevant rules began in August 2026 through national market-surveillance authorities.
Research Gaps and Limitations
This essay is a synthesis and conclusion.
It does not provide a complete evaluation of every possible intervention.
There is no single accepted scientific definition of healthy digital agency.
The proposed questions concerning stopping, understanding, regret, choice and alignment with personal values are a philosophical framework rather than a validated diagnostic instrument.
Screen time remains a useful measure in some contexts.
The argument that screen time alone is insufficient should not be interpreted as evidence that duration never matters.
Digital experiences differ substantially in purpose, content, intensity, social context and individual effect.
The effectiveness of current regulations remains uncertain.
Many duties are new.
Enforcement is still developing.
Companies may alter products in ways that satisfy formal requirements without producing meaningful reductions in harm.
Age assurance may protect children while creating risks involving privacy, exclusion, surveillance or incorrect age classification.
Non-profiled feeds may increase user choice but could also reduce relevance, discovery or access to minority content.
Private-by-default accounts may reduce unwanted contact while limiting some forms of beneficial connection.
Protective defaults should therefore be evaluated rather than assumed effective.
Independent researcher access creates significant privacy and security challenges.
Poorly designed access systems could expose sensitive user information or reveal tools that allow harmful actors to manipulate platforms.
Contextual advertising may reduce behavioural surveillance but may generate less revenue for some publishers or services.
It is not yet clear whether the largest free digital platforms could operate at their current scale through contextual advertising alone.
Subscriptions can align revenue more closely with user value but may increase economic exclusion and create fragmented access to essential information or communication.
Media, commercial and AI literacy programmes vary substantially.
There is limited longitudinal evidence establishing which educational approaches produce lasting changes in behaviour or resilience.
Understanding persuasive design does not make a person immune to it.
Knowledge must be accompanied by practical control, supportive environments and realistic alternatives.
Parental mediation differs according to time, income, technical confidence, household structure, disability, culture and children's individual needs.
No one parenting model will suit every family.
Workplace-attention reforms may conflict with customer expectations, emergency responsibilities, global time zones and operational needs.
Protected focus should not be imposed where rapid response is genuinely necessary.
Artificial intelligence is changing quickly.
Predictions concerning personalised persuasion, synthetic companions and automated content remain partly inferential.
The same capabilities may strengthen or weaken attention according to business model, design objective, user context and regulation.
AI literacy requirements are new.
Their practical interpretation and enforcement will vary among organisations and countries.
Further research is needed into:
- reliable measures of time well spent and post-session regret;
- whether non-profiled feeds improve agency or wellbeing;
- the effectiveness of private-by-default teenage accounts;
- the long-term effects of removing visible social metrics;
- which forms of protective friction users accept without abandoning useful services;
- the commercial viability of advertising models requiring less behavioural data;
- how companies balance long-term user satisfaction against immediate engagement;
- whether risk assessments conducted before product changes alter real design decisions;
- how regulators can audit complex recommendation systems;
- how to provide researcher access without compromising privacy;
- which forms of media literacy transfer into everyday behaviour;
- how AI-generated and highly personalised content will affect attention;
- the effects of AI companions upon children and vulnerable adults;
- whether AI reduces or increases total workplace communication;
- how organisational attention costs can be measured accurately;
- whether regulatory changes produce consistent global protections or fragmented regional experiences;
- how children themselves define a safe, valuable and enjoyable online environment.
The essay does not argue that all persuasive design should be prohibited, that all behavioural advertising is equally harmful or that technology companies should abandon commercial growth. It argues that commercial growth should no longer receive automatic priority where it conflicts with human agency, child development, privacy, safety or the ability to direct attention deliberately.