# The Future Will Not Wait for Us

## Artificial intelligence is transforming humanity's very capacity to produce knowledge. Yet we keep discussing this revolution as if it were just another technological innovation. The price of this misunderstanding may be far greater than we imagine.

There is something deeply unsettling about the way we are watching the evolution of artificial intelligence. Not so much because of what the technology can already do, extraordinary as the advances are, but because of the apparent calm with which much of society continues to regard a transformation that could alter the foundations of our economy, of work, of science and of the very organisation of collective life.

What surprises me most is not the indifference of those who are far removed from technology. It is the difficulty I find in so many intelligent, well-informed people with significant responsibilities in grasping the true scale of what is happening. Experienced managers, university professors, economists, business leaders and policymakers who continue to discuss artificial intelligence essentially as a tool for automating tasks, raising productivity or cutting costs. They are not necessarily wrong, but I fear they are focusing on a relatively small part of a much deeper transformation.

It is remarkable how easily we manage to absorb something new into our daily lives without truly questioning its consequences. Three years ago, many of us had never used a generative artificial intelligence system. Today we talk to these tools, ask them for analyses, write programs, produce documents and solve problems that previously required specialist knowledge. All of this happened with a speed that, in other circumstances, would have been enough to provoke deep reflection about the future. Instead, we got used to it. The technology became part of the routine and, in doing so, lost some of its capacity to surprise us.

Perhaps this is one of our greatest weaknesses. The human capacity to adapt, normally an extraordinary advantage, can also make us dangerously indifferent to what ought to force us to stop and think. When a change becomes familiar too quickly, we run the risk of no longer noticing its importance.

On 17 October, Jack Clark, co-founder of Anthropic, delivered a speech in Bangkok to an audience of economists, bankers and policymakers that deserves far more attention than it is likely to receive. The occasion was the Per Jacobsson lecture, associated with the meetings of the International Monetary Fund, and the title he chose was *Inside the Event Horizon: Why Everyone Should Take AI Progress Seriously*.

Clark is not a distant observer of this industry. He began by studying and writing about data centres, joined OpenAI in 2016 and later took part in founding Anthropic, one of the companies responsible for developing the most advanced artificial intelligence models. He is someone who follows this evolution from the inside, who knows its capabilities and limitations and who, naturally, also has direct interests and responsibilities in how it develops. That does not make his claims beyond dispute, but it makes them important enough that they cannot be ignored.

The message he brought to that audience was disconcerting: in his view, the technological Singularity has already begun.

He was not announcing the arrival of conscious machines or an omnipotent artificial intelligence, as we have so often imagined in science fiction. He was arguing something far more concrete and, perhaps for that very reason, more disturbing. Artificial intelligence systems have begun to show the ability to take part in the discovery of new scientific knowledge, to solve problems that had remained open and to contribute to the development of technologies that may, in turn, accelerate the progress of artificial intelligence itself.

It is important to understand what is really at stake, because the difference between a machine that performs intellectual tasks and a machine that takes part in producing new knowledge is not merely a matter of degree. It is a difference that could change the speed at which humanity progresses and, with it, many of the assumptions on which we have built our institutions.

## When we are no longer the only ones discovering

For practically the whole of human history, scientific progress has depended on a condition we rarely questioned: the availability of human beings able to investigate, formulate hypotheses, run experiments and understand the results. Science developed through an extraordinarily demanding process in which generations of researchers added small and large discoveries to the knowledge accumulated by their predecessors. Human intelligence, the time needed to train it and the resources available for research have always been fundamental limits on the speed of that progress.

Artificial intelligence is beginning to call that reality into question.

In his speech, Jack Clark presented results from scientific evaluations that illustrate the speed of recent advances. On *Humanity's Last Exam*, a test designed to assess reasoning and knowledge on particularly difficult problems, the best systems went, according to the figures he presented, from a score of 8% in January 2025 to 67.7% in October 2026. On another evaluation, *FrontierMath*, devoted to solving complex mathematical problems, the progression he cited was from 2% to 94%.

These numbers should be read with caution. Evaluations have their own methodologies and limitations, and a high score does not mean that a system can replace a researcher or autonomously run a complete scientific programme. But it would be equally unrigorous to ignore what they reveal about how the models' capabilities are evolving.

There was, however, one episode that strikes me as especially significant. In a recent theoretical physics evaluation called *TheoPhysEval*, researchers gathered fifty unpublished problems, complex enough to challenge specialists in the field. According to the account Clark gave, one of the systems evaluated correctly solved 49 of those problems and, when some of its answers were initially marked as wrong, it was able to identify errors in the calculations of the very experts who had prepared the solutions.

It is worth thinking about what this episode means, even while acknowledging that the results need independent validation. For decades we grew used to the idea that machines were extraordinary instruments in the service of human knowledge. They could calculate faster, analyse enormous quantities of information and carry out operations that would be impossible to perform by hand. But it was still we who formulated the relevant questions, interpreted the results and set the limits of knowledge.

Now situations are beginning to arise in which artificial systems can help correct what human experts thought they knew. We are not just talking about retrieving existing information or repeating patterns learned during training. We are talking about reasoning and problem-solving abilities that, in certain circumstances, can contribute to producing original knowledge.

We do not know how far these capabilities will go, nor whether progress will continue at the pace we have observed. But I find it hard to look at these results and conclude that we are simply facing another piece of software.

The question becomes even more important when we realise that the same technology can be used in disciplines as different as mathematics, physics, biology, programming or research into new materials. Unlike many specialised scientific systems, large AI models are beginning to show a capacity for generalisation that allows them to apply knowledge and methods to problems of very different kinds.

It is precisely this characteristic that makes the change so hard to compare with any previous transformation.

The steam engine multiplied our physical strength, electricity allowed production to be reorganised and computers vastly increased our capacity for calculation. The Internet transformed the circulation of information and the way we communicate. None of these technologies, however, had the general ability to take part directly in the research needed to develop the next generation of technologies.

Artificial intelligence may be able to.

If increasingly capable systems can help improve other systems, investigate new algorithms, develop materials, speed up experiments and find solutions to problems we do not yet know how to solve, we will be facing a mechanism of progress with characteristics profoundly different from those of previous industrial revolutions.

This does not mean that growth can become infinite, nor that physical, economic or scientific limits will disappear. But it does mean that the very speed of innovation may come to be shaped by the technology we are developing.

That is the Singularity hypothesis. And even if Jack Clark is wrong to claim that it has already begun, the advances he describes justify far more attention than we are giving them.

## Intelligence may stop being the scarcest resource

One of the most interesting parts of the speech concerns the economic consequences of this transformation. For decades, economists have studied a growing difficulty in scientific research: the need to mobilise ever more resources and researchers to keep producing relevant discoveries. Finding new ideas has become, in many fields, progressively harder.

Training a highly specialised researcher can take ten or fifteen years. Building a competent scientific team requires sustained investment, solid institutions, accumulated experience and conditions that are not within every country's reach. Even the most developed economies struggle to gather the human resources needed in certain areas of research.

But what will happen if a scientist can work with hundreds or thousands of artificial agents able to analyse information, formulate hypotheses, develop models and propose experiments? And what if those agents can work in parallel, be replicated quickly and combine knowledge from different disciplines?

Clark invites us to imagine a world in which the roughly ten million human scientific researchers in existence, according to the estimate he uses, could coordinate a billion or even ten billion virtual researchers.

It is a hypothetical scenario, not a proven forecast. But the possibility deserves to be considered, because it profoundly changes the relationship between the number of people available to do research and the amount of intellectual work that can be done.

If this capability materialises, we may witness an extraordinary acceleration in drug discovery, in the development of new materials, in energy research, in agriculture and in many other areas essential to human well-being. Problems that today require decades of work could, in some cases, be tackled in much shorter periods.

We should not, however, confuse available intelligence with guaranteed results. A scientific hypothesis, however brilliant, does not automatically turn into a medicine, a new source of energy or an industrial process. It has to be tested, validated, manufactured and approved, and the conditions have to be created for a discovery to be used safely.

This is where one of the most interesting economic consequences of this revolution emerges. If the intellectual capacity needed to develop new ideas becomes far more abundant, scarcity may shift to the physical resources that allow those ideas to become reality. Laboratories, scientific equipment, industrial facilities, energy, robotics, raw materials and experimental capacity may become even more important.

Jack Clark recalls that he spent years warning governments of the need to build energy capacity, data centres and the supply chains associated with artificial intelligence. Many did not give those recommendations the importance they now recognise. He now believes the next phase will require equally significant investment in scientific infrastructure and robotic systems capable of carrying out experiments in the physical world.

This strikes me as a particularly relevant observation, because it helps us see that this transformation will not be decided only by whoever develops the best artificial intelligence models. It will also be determined by the ability of each country and each company to use that intelligence and turn it into concrete results.

A country can have access to the most advanced models and still lack the laboratories, energy, companies or institutions needed to draw significant benefits from them. It can consume technology developed elsewhere without ever truly taking part in creating the economic value it generates.

And it can discover, too late, that knowledge has stopped being its main constraint, but that it never built the conditions needed to turn it into wealth.

## The slowness of those still living in the previous world

This is where I find the most worrying dimension of this whole story.

The problem is not only the speed at which artificial intelligence is evolving. It is the enormous gap between that speed and the capacity of our institutions, companies and societies to adapt.

There is a form of intellectual isolation that seems to me particularly dangerous in times of deep change. It has nothing to do with a lack of education or intelligence, but with the difficulty of abandoning ways of thinking that worked well for a long time. The more experience one has accumulated in a given reality, the harder it can become to accept that some of its fundamental premises are ceasing to hold.

We find this difficulty in many discussions about artificial intelligence. We keep asking how many jobs will be automated, when perhaps we should also be asking how companies will be organised when a small human team can coordinate hundreds of digital agents. We keep debating whether students should use these tools for their schoolwork, when perhaps we should be reconsidering the very nature of learning and of the skills we want to develop. We keep measuring productivity in terms of the human resources available, without giving enough thought to what may happen when a significant share of the intellectual capacity needed for production can be multiplied digitally.

This does not mean the traditional questions no longer make sense. Employment, education and productivity remain fundamental issues. But we need to understand that the answers may change because the assumptions on which we built those questions are also changing.

It is easy to find examples of this difficulty in organisations. Many companies began their approach to AI with isolated experiments, writing tools, coding assistants or administrative automation. It is a natural first step and, in many cases, a useful one. The problem arises when using new tools is confused with actually transforming the organisation.

A company can raise its employees' productivity and go on working in exactly the same way, with the same processes, hierarchies and business models. Meanwhile, another can use the same technologies to completely reorganise how it develops products, delivers services and makes decisions. The difference between them will not lie only in the amount of technology used, but in the ability to understand that it has become possible to do things that were previously not economically viable.

The same problem faces governments. Public institutions were built to guarantee stability, predictability, legality and continuity. These are essential characteristics in a democratic society, but they can become obstacles when decision-making processes cannot keep up with very rapid structural change.

Reforming an education system takes years. Building energy infrastructure can take a decade. A solid scientific community cannot be built through a short-term funding programme. Changing how a public administration is organised requires persistence, competence and a capacity for execution that often outlasts political cycles.

And yet some of the capabilities of AI systems are evolving on a scale of months.

We do not know whether this pace will be sustained, but we do know that institutions cannot transform themselves as quickly. It is precisely this gap that should worry us, because the preparation needed to seize an opportunity may take far longer than the time left before that opportunity arrives.

Some decisions, by the time they become obvious to everyone, already come too late.

## Portugal cannot simply stand by and watch

I cannot think about this transformation without looking at Portugal and asking whether we are truly aware of what is at stake.

We have universities, researchers, technology companies and professionals capable of doing work of great quality. There are scientific and industrial fields in which we have accumulated relevant knowledge and experience. We are not starting from scratch, and it would be unfair to reduce the country to its difficulties.

But we also know all too well the distance that so often separates our intentions from our ability to deliver. We have a persistent tendency to multiply strategies, programmes, structures and announcements without ensuring that there are enough resources, responsibilities and implementation mechanisms to produce lasting results.

Technological development does not wait for us to finish our administrative procedures, for different bodies to reach agreement or for a strategy to pass through every stage of approval. And I fear we continue to underestimate the economic cost of that slowness.

The question is not whether Portugal should bet on artificial intelligence. That seems obvious to me. The truly important question is understanding what it means to bet on a technology that may simultaneously transform scientific research, industry, services, work and education.

It is not enough to teach people how to use AI tools, although that is necessary. It is not enough to support a few technology companies or to create more specialised courses. Nor is it enough to announce investments in data centres, as if the availability of computing capacity were, in itself, a guarantee of economic development.

We need to discuss which scientific and industrial capabilities we want to develop, in which areas we can create competitive advantages, what infrastructure will be needed and how we can more effectively connect the knowledge produced in universities with the needs of companies and society.

We need to understand how small and medium-sized enterprises, which make up such an important part of the Portuguese economy, can benefit from these capabilities without becoming entirely dependent on external suppliers. How we can help organisations with few resources reorganise processes, develop new products and become more competitive. And how we can prevent the gap between companies able to make use of AI and those that cannot from becoming a new source of economic inequality.

But perhaps the most important question is education. We continue to prepare children and young people for working lives spanning several decades, in a context where some of the skills valued today may change profoundly within a few years. We do not know exactly what the professions of the future will be, and it would not be serious to pretend we can predict them. But it seems clear to me that the ability to think critically, keep learning, frame problems, understand systems and exercise human judgement will become even more important.

The answer cannot simply be to bring artificial intelligence tools into classrooms. We have to rethink what we teach, how we teach and what it truly means to learn when a machine can produce convincing answers to problems that once required years of training.

I would like these discussions to take a far more prominent place in Portuguese public debate. Not because the country's other problems have ceased to exist, but because many of them may be profoundly affected by this transformation. Productivity, wages, the sustainability of public services, the competitiveness of companies and the ability to retain talent are not issues separate from artificial intelligence.

They are precisely some of the areas where the consequences may be most significant.

Portugal does not need to compete in everything, nor to entertain the illusion that it could reproduce the scientific and financial capacity of the largest economies. But it does need to understand where it can create value and to have the discipline to concentrate resources, set priorities and execute.

Because there is an enormous difference between taking part in a transformation and merely consuming the products it generates.

And that difference may determine a significant part of our future prosperity.

## Progress can also make us more vulnerable

It would be tempting to look at all these possibilities and conclude that we are on the verge of an age of unprecedented abundance. Perhaps we will manage to develop treatments for diseases that are incurable today, produce cheaper energy, create more efficient materials and find solutions to environmental and social problems that for decades seemed impossible to solve.

There are legitimate reasons for enthusiasm. History shows that science and technology have contributed decisively to improving the living conditions of a very significant part of humanity. Advances in medicine, food production, sanitation, energy and communications have profoundly transformed what it means to live in the contemporary world.

But technological progress has never, on its own, guaranteed a fair distribution of its benefits.

New capabilities can create wealth and, at the same time, concentrate economic power. They can raise productivity and leave workers without alternatives. They can make essential services more accessible and create new forms of dependence. They can bring countries closer together or widen the gap between those who master the technology and those who merely use it.

Artificial intelligence may intensify some of these tensions, precisely because it acts on intellectual activities that have long been associated with education, specialisation and job security.

And there is an additional dimension we must not ignore. The same capabilities that can accelerate medical research can be used to develop dangerous biological agents. The systems that help improve cybersecurity can also facilitate attacks. The possibility of using AI to develop ever more capable AI systems raises difficult questions about oversight, control and accountability.

Jack Clark acknowledges these risks and argues that humanity will have to learn to distinguish between what it can do and what it should do. It is a seemingly simple idea, but one with profound consequences in an industry where economic and strategic competition creates constant incentives to develop more capable systems.

We cannot assume that every technology should be developed and released as quickly as possible simply because the technical capacity to do so exists. But nor can we use the risks as a justification for indiscriminately blocking progress that could improve the lives of millions of people.

The challenge will be to find ways of promoting research and the responsible adoption of technology while maintaining rigorous oversight in the areas where mistakes or misuse could have serious consequences.

And here we return to the question of institutions. We need governments and regulators capable of understanding fast-moving technologies, assessing real risks and making informed decisions. We cannot leave these choices solely in the hands of the companies that develop the models, but nor can we expect structures that do not understand how the technology works to regulate it properly.

Artificial intelligence may vastly increase our ability to solve problems. That does not mean we are developing, at the same speed, the maturity needed to decide how to use that ability.

Perhaps this is one of the most unsettling contradictions of the revolution we are living through.

## The illusion that we still have plenty of time

There is a question that has stayed with me since I read Jack Clark's speech: what if we are underestimating not only the scale of this transformation, but above all how quickly its consequences may arrive?

I am not convinced that anyone can accurately predict how artificial intelligence will evolve over the next five or ten years. There are too many unknowns, from the technical limits of the models to the availability of energy, computing capacity, investment and scientific validation. The most ambitious forecasts may prove exaggerated, just as some expectations about earlier technologies did not materialise within the announced timeframes.

But it seems to me a mistake to use that uncertainty as an argument for continuing to postpone decisions.

When the potential consequences are profound and preparation takes many years, we cannot wait for certainties that we will probably only have once events have already happened. It would be like deciding to build energy infrastructure only when a lack of electricity was already preventing new industries from setting up.

Preparation does not begin when the future becomes obvious. It begins when there are strong enough reasons to recognise that this future is possible and that we will not be ready if we do not act.

Perhaps the technological Singularity, as Clark describes it, will not materialise in the coming years. Perhaps science will continue to face limitations that AI models cannot overcome. Perhaps the economic impact will be more gradual and uneven than some forecasts suggest.

But it is also possible that certain advances will happen much faster than we expect. And if that happens, the speed with which we can react will be decisive.

It is important to remember that great economic transformations do not distribute their benefits automatically. Countries, companies and communities that fail to adapt can lose relevance, even as humanity as a whole becomes richer and more technologically advanced.

History is full of examples of companies that dominated their markets and did not realise in time that the conditions of competition had changed. There are also countries and regions that lost important economic activities and took generations to find alternatives.

What sets artificial intelligence apart is the possibility that this transformation will reach an extraordinary number of sectors and activities at the same time, including some of the mechanisms we use to develop knowledge and create new technologies.

We are not talking merely about one industry being replaced by another.

We are talking about a potential change in the way innovation itself happens.

And that makes time a particularly precious resource.

## The greatest risk may lie within us

What disturbs me most about this story is not imagining machines able to solve problems that humans cannot. On the contrary, I see in that possibility an extraordinary opportunity to tackle difficulties that have accompanied humanity for centuries.

What disturbs me is the thought that we may be facing one of the greatest opportunities in our history and fail to seize it for lack of the collective capacity to understand, anticipate and decide.

I would like to see more intellectual curiosity and fewer hasty certainties. More willingness to question what we think we know and less need to reduce a complex reality to comfortable explanations. More attention to the economic and social consequences of this transformation and less concern with proving that a given tool still makes mistakes or cannot fully replace a professional.

Of course mistakes matter. The limitations of these systems must be studied and the risks need to be assessed. But a technology does not need to be perfect to profoundly transform a society. Nor does it need to replace every human capability to radically change the economic value of some of them.

There is a difference between acknowledging that we do not yet know what the future will look like and behaving as if it must resemble the past.

It is precisely that confusion that seems dangerous to me.

Our institutions were built in a world where knowledge advanced essentially at the speed of human work. Education systems, business organisations, political processes and many of our professional expectations rest, more or less explicitly, on that reality.

If we begin to have systems capable of multiplying a significant share of the available intellectual capacity, we will have to reconsider many of those structures. Not because human beings stop mattering, but because their role may change profoundly.

Oversight, responsibility, judgement, the choice of goals and the understanding of consequences may become even more relevant. And perhaps we will discover that some of the most important qualities for living in a technologically advanced society are not the ones we have traditionally associated with the ability to accumulate knowledge.

Artificial intelligence does not excuse us from thinking. On the contrary, it may force us to think better, about harder problems and with far greater consequences.

Jack Clark ended his speech by invoking Carl Sagan and the idea that humanity would need to develop more of its virtues and fewer of its weaknesses to reach a better future. To be more far-sighted, more capable, more prudent and more aware of its responsibilities.

Sagan imagined that evolution over a very long timescale. Clark believes we may have only a few years to make fundamental decisions about technologies of growing power.

I do not know whether that timeframe is realistic. But I do know that a society's ability to prepare for profound change is not built overnight.

And so I cannot help thinking about the responsibility we bear towards those who will live with the consequences of the decisions we are making, or avoiding making, right now.

I would like to live in a country capable of looking at this revolution with ambition, but also with maturity. One that recognised the importance of scientific research, education and innovation without turning every new technology into a miraculous promise. One that knew how to identify opportunities, take calculated risks and correct mistakes. And one that had the courage to recognise that some of the ways we organise ourselves, teach, work and decide may stop making sense much sooner than we would like.

I would not like to see Portugal resigned to following, a few years late, transformations that others had the foresight to anticipate. Nor would I like future generations to pay the price of our difficulty in letting go of certainties that no longer matched reality.

Because that price will not be measured only in lost economic growth. It may take the form of companies that were never born, jobs that disappeared without alternatives, researchers who had to seek elsewhere the conditions they needed to work, wasted industrial opportunities and a growing dependence on technologies and decisions made beyond our borders.

And, above all, it may take the form of a reduced collective capacity to choose the future we want to build.

We do not know whether we are truly at the beginning of the Singularity. But we do know that artificial intelligence is already transforming the frontier of technological and scientific capabilities, that this transformation may accelerate and that its consequences will reach far beyond the technology sector.

That should be enough to make us act with a seriousness very different from the one we have shown so far.

Perhaps we still have time to prepare our companies, our schools, our institutions and the country itself. I want to believe we do. But the time needed for that preparation may be much longer than the time left before we need it.

And there is something deeply sad about the possibility that one day we will look back and realise that all the signs were in front of us, that there were people explaining what was happening, that we had the knowledge and resources to begin changing, but that we remained too busy managing the same old problems to grasp the scale of what was coming.

History will not be particularly forgiving of that failure.

Perhaps the greatest danger of artificial intelligence is not that we create ever more intelligent machines. Perhaps it is that we keep making decisions with an understanding of the world that no longer keeps pace with reality.

And when we finally see the distance that has opened up between what it was possible to do and what we actually did, we may find that the price of our slowness can no longer be recovered.

Not because the future caught us by surprise.

But because, despite all the signs, we chose to keep looking the other way.

---

*Opinion essay, October 2026.*

*Inspired by the speech by Jack Clark, co-founder of Anthropic, "Inside the Event Horizon: Why Everyone Should Take AI Progress Seriously", delivered on 17 October 2026 and published in issue 477 of the Import AI newsletter. The scientific results mentioned are those presented by the speaker. The reflections on Portugal, the economic and social consequences and the capacity of institutions to respond are the interpretation of the author of this essay.*