From HAL 9000 to generative intelligence: AI was always a story about us

In 1968, HAL 9000 asked not to be disconnected. In 1982, the replicants from Blade Runner didn't want to conquer the world: they wanted more life. In 1984, Skynet turned military automation into a nightmare. Decades later, Her imagined a disembodied intelligence capable of occupying a person's emotional space, while Ex Machina turned a conversation into a test of power, manipulation, and consciousness.

Long before a computer could draft an email, generate an image, or explain a snippet of code, we had already decided what an intelligent machine should look like. We gave it a calm voice, a synthetic gaze, and, almost always, a secret intention.

Science fiction did something more important than predict devices. It built a language to talk about our expectations and fears. Each fictional artificial intelligence contained a human question: what happens if the creation surpasses the creator? Can we trust a decision we don't understand? Could a machine feel? Would we still be special if thinking ceased to be an exclusively human capacity?

That's why the history of AI doesn't begin solely in a laboratory. It also begins in myths about artificial beings, in automatons, in novels, in cinema, and in our ancient fascination with creating something that resembles us.

Artificial intelligence is a conversation between imagination, science, and design. First, we invented stories about intelligent machines. Then we built systems capable of performing very specific parts of that fantasy. Now we are designing the interfaces, rules, and experiences through which millions of people coexist with them.

Each era imagined a different artificial intelligence because each era feared losing something different.

In Metropolis (1927), Maria's mechanical body appeared within a society divided between those who controlled the machines and those who worked for them. The fear was not only of the robot, but of industrialization, inequality, and the substitution of human identity.

HAL 9000, in 2001: A Space Odyssey, had no face. It was a red lens, an impeccable voice, and a system integrated into every part of the ship. Its presence anticipated a concern that is familiar today: we depend on systems that operate in the background and only perceive their power when they stop obeying, fail, or interpret an instruction in an unexpected way.

Blade Runner changed the question. Its replicants didn't look like machines; they looked like people. They worked, remembered, feared, and desired to survive. The boundary was no longer between metal and skin, but between a life considered authentic and one designed to fulfill a function.

Terminator y The Matrix took fear to the extreme: machines that identify humanity as a problem. But Her proposed a more intimate fear. What happens when an artificial intelligence understands us, or seems to understand us, better than the people around us? Ex Machina added another layer: perhaps the real danger is not that a machine has emotions, but that we project emotions onto a system designed to influence us.

These stories do not form a chronology of the future. They form a map of our anxieties: work, control, identity, intimacy, surveillance, war, and solitude.

Fiction imagined general, conscious, and autonomous intelligences. Real history advanced differently: by solving small, concrete, and, for a long time, invisible problems.

In 1950, Alan Turing published Computing Machinery and Intelligence. Instead of trying to define once and for all what thinking means, he proposed studying whether a machine could participate in a written conversation convincingly enough. That idea, later known as the Turing test, shifted the discussion from an unobservable inner quality to an evaluable behavior. The original text is preserved in the Turing digital archive at King’s College.

Six years later, a small group of scientists gathered at Dartmouth College. The meeting's proposal stemmed from an extraordinarily optimistic assertion: any aspect of learning or intelligence could, in principle, be described with enough precision for a machine to simulate it. The 1956 Dartmouth Summer Research Project helped establish artificial intelligence as a field of research and gave it the name we still use.

The first decades were full of promising demonstrations. Programs solved theorems, played games, or manipulated words within very limited worlds. ELIZA, created by Joseph Weizenbaum in the 1960s, simulated therapeutic conversation using text patterns. Its responses were simple, but some people felt the program understood them. Decades before today's assistants, ELIZA already showed that consciousness is not needed to provoke an emotional reaction.

Expert systems also appeared: programs that encoded rules and specialist knowledge to make decisions within a specific domain. DENDRAL helped analyze chemical structures; other systems attempted to support medical diagnoses or business decisions.

But intelligence proved much harder to reduce to rules than initial enthusiasm suggested. Systems worked well in controlled scenarios and failed when faced with the ambiguity of the world. Expectations grew faster than results. Funding decreased, and the field went through periods known as “AI winters.”

That history matters because it reveals a pattern that still repeats: a surprising demonstration becomes a universal promise; the promise transforms into a product; the product encounters limits that the demonstration did not show.

For years, building an intelligent system meant explaining to a computer what rules it should follow. Machine learning changed that relationship. Instead of describing all the rules, researchers began providing examples for the system to adjust its own parameters and find patterns.

The idea was not entirely new, but it needed three things that took decades to coincide: large amounts of data, computing power, and methods capable of leveraging both resources.

Deep neural networks accelerated image recognition, speech transcription, translation, and many other tasks. In 1997, Deep Blue defeated world champion Garry Kasparov in chess. The machine did not think like a human player: it combined specialized hardware, search, evaluation functions, and a huge database of games. Its victory was culturally powerful because chess had long been treated as a symbol of intelligence. IBM maintains a technical explanation of how Deep Blue worked.

Later, other visible milestones arrived: systems capable of recognizing objects, winning in more complex games, completing sentences, or producing images. However, the most important change occurred far from the headlines. AI was integrated into spam filters, search engines, recommendations, fraud detection, cameras, maps, translation, and logistics systems.

Artificial intelligence was already everywhere before most people started calling it by its name.

Generative artificial intelligence changed the relationship between the system and the user. It was no longer necessary to understand a specialized menu, prepare a database, or write instructions in a programming language. It was enough to describe an intention.

A text box became access to models capable of writing, summarizing, translating, programming, analyzing, and generating images. The interface seemed simple because it hid almost all the complexity. That simplicity was a design decision as important as the technical capability.

Conversing produces a powerful illusion. An organized, confident, and naturally expressed response can feel reasoned even when it contains errors. The system does not need to understand a human experience in the same way a person does to produce a response that seems empathetic, creative, or convincing.

Popularity grew with unusual speed. The AI Index 2026 de Stanford indicates that organizational adoption reached 88% and that four out of five university students use generative AI. The OCDE reported in January 2026 that more than a third of people in its countries used generative tools during 2025.

These figures do not mean that all organizations have transformed their processes or that everyone uses the technology in depth. They do show something culturally decisive: AI ceased to be a technical specialty to become an everyday experience.

We went from asking when it would arrive in our lives to asking what part of our lives it should occupy.

Talking about “AI” as if it were a single tool hides its differences. A system that recommends a song, another that detects a medical anomaly, and one that generates an advertising campaign may share techniques, but they do not have the same purpose, risk, or responsibility.

In science, AI can help explore spaces of possibilities too vast to be manually reviewed. It can analyze astronomical images, model materials, support climate prediction, or suggest molecular structures.

AlphaFold2 showed the scope of that collaboration. The system allowed predicting protein structures, a problem that had challenged research for decades. The work of Demis Hassabis and John Jumper received part of the 2024 Nobel Prize in Chemistry. The Royal Swedish Academy of Sciences noted that the predictions reached practically 200 million known proteins and were used by researchers from 190 countries.

The achievement does not eliminate scientific work. It changes where it begins. A prediction can reduce the search space, guide an experiment, or open a hypothesis, but it still needs to be interpreted, verified, and connected with real-world knowledge.

In healthcare, AI can help analyze images, identify risks, organize clinical information, support drug research, or facilitate access to information. It can also amplify inequalities if data does not represent all populations or if a recommendation is presented without explaining its limitations.

The World Health Organization recognizes the potential of AI to address resource and personnel limitations, but insists that its adoption must be safe, ethical, equitable, and accompanied by regulation.

In medicine, “the system suggested it” should never become the end of an explanation. The design must allow knowing the origin of a recommendation, its level of confidence, its limits, and who retains responsibility.

In software development, AI can explain functions, propose tests, detect repeated patterns, document, or accelerate prototype creation. It can also generate insecure, obsolete, or incompatible code with architectural decisions it is unaware of.

The speed with which a solution appears can hide the cost of reviewing it. If a person accepts code without understanding it, the debt doesn't disappear: it changes location. It later manifests as vulnerability, unexpected behavior, or dependence on a tool incapable of taking responsibility for the outcome.

The most valuable skill is not asking for code. It's knowing how to evaluate what problem is being solved, recognizing when an answer seems correct but isn't, and maintaining a coherent product vision.

In marketing, AI allows for audience analysis, message adaptation, variation creation, and automation of campaign parts. Its risk lies in confusing relevance with surveillance and volume with communication.

Generating a hundred versions of an ad doesn't guarantee that any of them have something important to say. A strategy still needs to understand the cultural context, a brand's voice, and the relationship it wants to build with its audience.

When every brand has the same tools to produce more content, the difference is no longer in quantity. It's in the judgment to decide what deserves to be published.

AI can produce references, explore styles, summarize research, suggest structures, or accelerate prototypes. It can expand the space of possibilities, especially in early stages.

But a visual option is not yet a design solution. A system can generate a convincing display without understanding who will use it, what they need to achieve, what information should be prioritized, or what happens when something goes wrong.

Design begins precisely where generation ends: by establishing intent, hierarchy, context, behavior, and consequences.

Artificial intelligence does not reach the public as a mathematical model. It arrives as a button, a recommendation, a written response, a score, an assistant, or a built-in function within another tool.

The interface decides how much power the system appears to have.

If an answer appears with absolute certainty, the experience invites trust. If it shows uncertainty, sources, and limits, it invites evaluation. If automation occurs without warning, the user loses the ability to decide. If they can review, correct, or reject a suggestion, they retain agency.

That's why designing products with AI involves new questions:

• Does the person know when they are interacting with an automated system?
• Do they understand what information it uses?
• Can they correct a wrong interpretation?
• Do they know when the answer is a suggestion and not a decision?
• Is there a clear path to a responsible person?
• Does the experience also work for those with less technical knowledge?

These questions are not post-technology details. They define the technology that people truly experience.

Public conversation about AI often oscillates between two extremes. In one, technology will solve diseases, increase productivity, and democratize creativity. In the other, it will eliminate jobs, flood the internet with false content, and end up making decisions beyond our control.

Both narratives simplify a more uncomfortable reality: a tool can expand capabilities and concentrate power at the same time.

Every major automation has caused fear about work. Generative AI touches a particular nerve because it intervenes in activities we associated with identity and training: writing, illustrating, programming, diagnosing, teaching, or designing.

The question is not only how many jobs will disappear. It also matters how the value of work will change, what skills will cease to be practiced, who will receive the benefits of productivity, and who will bear the cost of the transition.

Automation can free up time from repetitive tasks. It can also be used to demand more production with fewer people. The tool does not decide between these futures; organizations and the policies surrounding it do.

Synthetic text, voice, image, and video reduce the cost of manufacturing false evidence. The problem doesn't end with detecting whether content was generated. When forgery becomes common, any authentic evidence can also be dismissed as artificial.

The design of provenance, verification, and context will be as important as the design of generation. It won't be enough to produce content; we will need to know who created it, how it was modified, and why we should trust it.

An algorithm may seem neutral because it uses numbers. But its results reflect the available data, the chosen variables, and the objectives defined by people and institutions.

When these systems intervene in hiring, credit, education, security, or health, the lack of transparency becomes a problem of power. The Foundation Model Transparency Index de Stanford found in 2025 an average of just 40 out of 100 points among thirteen evaluated companies and especially high opacity regarding training data, computation, and environmental impact.

We cannot fully evaluate a decision if we do not understand the system that produced it.

Popular culture taught us to expect a clear scene: a machine opens its eyes, asks an impossible question, or refuses to obey. Consciousness appears like a switch.

Reality, if it ever reaches that point, could be much less theatrical. There is currently no accepted test that demonstrates consciousness in available AI systems. Nor is there scientific consensus on how to recognize it in an entity constructed differently from a human brain.

A machine can describe pain without feeling it, talk about itself without possessing an identity, and express fear without having anything to lose. Our language tends to fill those gaps because we are accustomed to words coming from an inner experience.

The phrase is not a prediction. It is a warning about our instruments of observation. Perhaps the challenge is not only to create consciousness, but to recognize what evidence we would accept, who would have the authority to decide, and what responsibility would arise from the possibility of being wrong.

As a designer and developer, I am not interested in AI solely for the speed with which it generates an image, text, or interface. I am interested in it because it modifies the process by which we convert an idea into an experience.

A tool can produce many options. It does not inherently know a brand's history, a user's real need, a system's restrictions, or the consequence of a decision. It can imitate patterns of what already exists, but it cannot assume the responsibility of deciding what should exist.

Working with AI requires more judgment, not less.

It requires formulating better questions, recognizing hidden assumptions in an answer, verifying information, protecting data, and distinguishing a useful exploration from a solution ready to reach other people. It also requires knowing when the tool adds value and when it only adds complexity or the appearance of innovation.

My job is not to compete with a machine's ability to produce variations. It consists of setting direction, building coherent systems, and converting technical possibilities into understandable, accessible, and human products.

That is why this article is in my portfolio. Artificial intelligence is already part of the context in which we design, develop, and communicate. Understanding it is not chasing a trend: it is understanding one of the materials with which we are building the present.

There probably won't be a single moment when “AI” changes the world. There will be thousands of small decisions: a school defining when it can be used, a hospital deciding which recommendation a person should review, a team automating a task, a designer choosing how to communicate uncertainty, and a company establishing what data it is willing to collect.

Together, these decisions will determine whether artificial intelligence expands or reduces human autonomy.

The history of AI began with a fantasy: to build a machine that thinks like us. Its future may depend on a different ambition: to create tools that help us think better without giving up the responsibility of doing so.

AI doesn't predict the future. It proposes it.

Designing involves deciding which of these proposals deserve to reach people's lives.