Decide today, measure ten years from now
Some decisions produce results almost immediately.
Raise a tax, change a fee, build a road or modify a subsidy and, with more or less precision, within a few months you can start measuring what happened.
But other decisions are much more uncomfortable.
Those involving education, technology, artificial intelligence or childhood can take years, even decades, to reveal their full consequences.
And lately I have the feeling that many of the major debates we are having share exactly this problem.
Artificial intelligence: even those who know the most are worried
In March 2023, an open letter called for at least a six-month pause in the training of artificial intelligence systems more powerful than GPT-4.
It was signed by tens of thousands of researchers, entrepreneurs and specialists. They were not asking for an end to artificial intelligence, but for a temporary slowdown in the race towards increasingly powerful systems while better safety mechanisms were developed.
Soon afterwards, something even more striking happened.
Sam Altman, CEO of OpenAI; Demis Hassabis, CEO of Google DeepMind; Dario Amodei, CEO of Anthropic; Geoffrey Hinton; Yoshua Bengio and many others signed an extraordinarily short statement from the Center for AI Safety:
Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks.
These are hardly people disconnected from the technology.
Does that mean that the major American AI labs want to stop artificial intelligence?
Not exactly.
OpenAI, Anthropic and Google DeepMind continue to develop increasingly capable models. What is significant is that they have developed specific frameworks to evaluate severe risks before deploying certain systems: OpenAI’s Preparedness Framework, Anthropic’s Frontier Safety Roadmap and Google DeepMind’s Frontier Safety Framework.
It is a rather strange situation.
We are accelerating a technology at the same time as the people at the forefront of developing it acknowledge that some of its potential risks could be enormous.
So why not stop?
Because the potential benefits are also enormous, because companies are competing, because countries are competing and because nobody knows with certainty what will happen.
And that is where the first problem appears:
how do you make a drastic decision about something when it may take years to know whether that decision was necessary?
Education: when we discover the problems many years later
On 8 September 2026, the results of PISA 2025 were published.
They are not particularly reassuring.
According to the OECD’s country note for Spain, Spanish students performed below the OECD average in mathematics, reading and science. Compared with 2022, Spain declined in mathematics and reading and remained roughly stable in science. The 2025 results are also among the lowest Spain has recorded in all three subjects and are below the 2015 results.
Nor is this exclusively a Spanish problem.
In 2025 the OECD recorded its lowest average results in the history of PISA in reading, mathematics and science. Between 2015 and 2025, average reading performance across OECD countries fell by 28 points and mathematics by 22 points.
So it would be far too easy to reduce the discussion to:
“Spanish education is a disaster.”
I think the real question is much more interesting.
As a teacher, there is something I have always found strange: we constantly change curricula, education laws, assessment systems, methodologies and priorities, yet we expect almost immediate results.
A child starting primary school today will finish compulsory education roughly ten years from now.
The major educational decisions we make today should probably be judged then, not two school years from now.
PISA does not prove that Spain’s results are caused by a lack of political consensus or by changes of government either. That would mean turning an intuition into causation.
But there are good reasons to ask whether the way we make decisions is designed for problems that move this slowly.
The OECD itself repeatedly argues that education reform needs long-term strategic vision together with continuous monitoring and evaluation. Its Education Policy Outlook 2024 says reforms should be grounded in long-term strategic plans while also including mechanisms that allow policies to be measured, corrected and adapted as new evidence appears.
Perhaps, then, the question should not be:
what new education law do we need?
But rather:
what education system do we want fifteen years from now, and what are we prepared to maintain for long enough to find out whether it actually works?
That requires something particularly difficult in politics: accepting that another government may eventually receive the credit —or have to deal with the failure— of a decision made today.
Social media: we want to delay it until 16, but give children smartphones at 11
Here the contradiction becomes even clearer.
In November 2025, the European Parliament approved a non-legislative report that proposes a harmonised minimum age of 16 for access to social media, video-sharing platforms and AI companions, while allowing access between 13 and 16 with parental consent.
The distinction matters: there is currently no general EU-wide ban on social media for under-16s. This is a political proposal from the European Parliament and part of a broader debate around child protection and age verification.
Spain is moving in a similar direction. As of September 2026, the Organic Law project on protecting minors in digital environments is still going through parliament, and the government is proposing to raise the minimum age for registering on social networks to 16.
So far, this may sound like a fairly normal legislative discussion.
Until we look at when children actually receive their first phone.
The study Infancia, adolescencia y bienestar digital, based on a consultation with almost 100,000 children and adolescents, places the average age of receiving a first mobile phone in Spain at 10.8 years old.
According to UNICEF Spain, 41% of children already have their own phone by age 10. At age 12, the figure is 76%. In secondary school, it reaches 92.8%.
In other words:
as a society, we have placed access to the smartphone roughly five years before the age we are now considering for access to certain social networks.
And there is another part of this story we tend to forget.
Adults.
The 2026 Social Media Study from IAB Spain found that 86% of Spanish internet users aged 12 to 75 use social media, that the average user visits 7.2 different platforms and spends roughly one hour a day on them.
Perhaps we should ask ourselves what we are actually teaching.
Because we can introduce parental controls, verify ages or prohibit certain services.
But then a child can sit down for dinner opposite two adults who look at their phones every three minutes.
Digital education probably does not begin when we give a child a phone.
It begins much earlier, when they watch what we do with ours.
The problem with decisions whose results are too far away
AI.
Education.
Social media.
They look like three separate debates, but I think they share something important.
We have to make decisions today about problems whose full effects we may not understand for many years.
And the further away the evaluation point is, the harder it becomes to take significant action.
Imagine that we genuinely delay access to certain social networks until the age of 16.
How would we evaluate the policy?
Academic grades after two years?
Attention problems?
Anxiety?
Hours of sleep?
Social relationships?
Education outcomes ten years later?
And how would we distinguish its effects from every other technological and social change that occurred during the same period?
Education has exactly the same problem.
A serious reform might require ten years before an entire generation has gone through it.
With artificial intelligence, the problem may be even greater: some precautions could appear completely unnecessary for years… precisely because they worked.
This creates an uncomfortable paradox.
Preventive measures have a huge political problem:
if they fail, everyone will see the failure; if they succeed, it may look as though they were never necessary in the first place.
Perhaps we need to think a little further ahead
I do not think the answer is to ban all technology.
Nor to stop artificial intelligence.
Nor to turn PISA into the only measure of our education system.
Nor to imagine that we can keep a teenager isolated from the internet until the age of sixteen.
Technology also allows us to learn, create, communicate and access opportunities that would have been unimaginable twenty years ago.
Perhaps the problem lies somewhere else.
We have institutions and decision-making cycles that are relatively short trying to deal with transformations that are extraordinarily long.
And that forces us to do something we are not particularly good at:
making decisions whose benefits we may never personally see.
An education policy should survive for long enough to be evaluated.
A technology policy should be capable of changing when new evidence appears.
Social-media regulation should measure its results and be changed if it does not work.
And every important measure should include, from the beginning, one extremely simple question:
when are we going to check whether this actually worked?
Because if the answer is “ten years from now”, perhaps we should start building mechanisms that allow us to maintain a course for those ten years.
Even if the government changes.
Even if the technology changes.
Even if nobody can personally take credit for the result.
Sometimes governing the future means exactly that.
Planting trees whose shade will probably be enjoyed by someone else.