Before screens, interfaces were physical.

Doors. Switches. Control panels. Everyday objects that communicated their purpose through their shape, position, movement and response.

A door handle suggests pulling. A switch suggests a change of state. A dial offers a way to adjust something, often with immediate feedback.

These interactions might seem obvious, but they're the product of a relationship between human cognition and the designed world.

How do we know what something does? How do we understand what actions are available to us? And how do we recognise whether an action has achieved what we intended?

These are questions we now associate with user experience design. But they existed long before websites, applications or even personal computers.

And understanding that history tells us something important about the future of digital design.

UX existed before the screen

It's tempting to think of UX/UI as a relatively modern discipline, emerging alongside the internet and the growth of digital products.

The terminology might be relatively recent, but many of its underlying principles are not.

Ergonomics, human factors, cognitive psychology and industrial design had already spent decades examining how people interact with objects, environments and complex systems.

Consider the cockpit of an aircraft.

Its controls need to be accessible, distinguishable and understandable, often under considerable pressure. Their positioning and behaviour aren't simply aesthetic decisions. They influence how effectively a pilot can interpret information and take action.

The same principles apply to factory machinery, household appliances and public transport systems.

Every designed interaction creates an opportunity for understanding, confusion or error.

This relationship between people and designed objects became central to the work of cognitive scientist Don Norman.

In The Design of Everyday Things, Norman explores how design can communicate possible actions, provide meaningful feedback and help people form accurate conceptual models of how something works.

One of his most influential contributions concerns the relationship between what an object allows someone to do and how those possibilities are communicated.

A door might be perfectly functional, but if its design encourages someone to push when they need to pull, there's a mismatch between the intended interaction and the person's understanding.

The problem isn't necessarily the person.

It's the communication between the person and the object.

And that distinction would become fundamental to digital interface design.

When computers borrowed the physical world

Early computers were powerful, but interacting with them demanded a level of technical understanding that most people didn't possess.

Command-line interfaces required users to learn specific instructions and syntax. The computer operated according to its own language, and people had to adapt to it.

Graphical user interfaces began changing that relationship.

Research at Xerox PARC, followed by commercially influential systems such as the Apple Lisa and Macintosh, helped establish a visual approach to computing built around recognisable objects and interactions.

The desktop.

The folder.

The document.

The window.

The wastebasket.

These weren't arbitrary visual choices. They provided familiar reference points for understanding an unfamiliar environment.

People already understood that documents could be stored in folders, folders could be organised, and unwanted material could be discarded.

The graphical interface translated those concepts into a digital context.

Rather than requiring people to understand the underlying computational process, the interface presented an abstraction they could recognise and manipulate.

This is where the first major transformation becomes interesting.

Computers didn't simply introduce a new way of interacting with information. They assimilated the visual language of the physical world to make that interaction possible.

A digital folder wasn't a physical folder, but the metaphor helped people understand its purpose.

A graphical button wasn't a mechanical switch, but it communicated a similar expectation: perform an action and receive a response.

In many ways, the success of early graphical interfaces depended on making the unfamiliar feel familiar.

Assimilation was only the beginning

The desktop metaphor helped people understand computers, but digital systems were never going to remain confined to the physical objects they imitated.

Consider a filing cabinet.

It occupies space. Its storage capacity is finite. Retrieving information requires physically locating a document, and duplicating that document takes additional effort.

A digital folder initially resembles that same organisational structure.

But its capabilities are fundamentally different.

Its contents can be searched in milliseconds, duplicated almost instantly, shared across continents and reorganised without moving a single physical object.

The metaphor remains, but the constraints that gave the original object its form begin to disappear.

The same transformation occurred across countless interactions.

Physical letters became emails. Printed maps became interactive navigation systems. Record collections became searchable music libraries.

In each case, digitisation initially borrowed something recognisable from the physical world before developing capabilities that the original object couldn't offer.

I think of this as a process of assimilation followed by something approaching annihilation.

Not the literal destruction of physical objects, or the elimination of physical experience, but the progressive removal of constraints that once defined how interactions worked.

A digital interface no longer needs to respect the dimensions of a filing cabinet, the capacity of a bookshelf or the geographical limitations of a printed map.

It can create entirely new forms of interaction.

And once those constraints begin to disappear, the possibilities for design expand considerably.

From designing objects to designing behaviour

Digitisation didn't just change what designers could create.

It changed the scale at which design could operate.

A physical product might take months to manufacture, distribute and modify. Once produced, changing its behaviour can be expensive or impractical.

A digital interface can be updated overnight.

Its changes can reach millions of people almost immediately.

And crucially, those interactions can be measured.

Where do people click? How long do they spend completing a task? Where do they abandon a process? Which variation produces the desired outcome?

Digital products introduced an extraordinary capacity to observe, test and optimise the relationship between people and designed systems.

This helped accelerate the development and commercial importance of UX/UI.

Researchers and practitioners such as Jakob Nielsen contributed influential methods for evaluating usability. His ten usability heuristics, refined in the 1990s, articulated principles including system visibility, consistency, error prevention and user control.

These principles gave designers practical ways to identify problems and evaluate interfaces.

But digital measurement also introduced another dimension.

An interface could now be judged not only by whether someone understood it, but by whether it produced a measurable commercial outcome.

Conversions. Engagement. Retention. Revenue.

These metrics can reveal meaningful problems and opportunities. A confusing checkout process, for example, can frustrate customers while simultaneously costing a business sales.

Improving the experience benefits both.

But the relationship isn't always so straightforward.

An interface can encourage someone to spend more time on a platform without improving the quality of that time.

A subscription journey can be optimised for retention while making cancellation unnecessarily difficult.

A recommendation system can increase engagement by repeatedly presenting content that provokes an emotional response.

In each case, the interface may be performing successfully against its chosen metrics.

Whether it's providing a better human experience is a different question.

Digitisation gave designers the ability to influence behaviour at enormous scale. It didn't automatically tell us which behaviours were worth encouraging.

This is where UX becomes more than a question of usability.

It becomes a question of responsibility.

The medium changes the experience

Marshall McLuhan's work offers an interesting perspective on this transformation.

In Understanding Media: The Extensions of Man, McLuhan explored how communication technologies influence human perception, behaviour and society.

His argument wasn't simply that new media allow us to communicate existing information more efficiently.

It was that the characteristics of a medium can reshape the way we experience the world.

Consider how digital interfaces have changed everyday expectations.

We expect information to be searchable.

We expect communication to be immediate.

We expect services to be accessible at almost any hour.

We expect personalised recommendations, instant feedback and increasingly seamless interactions.

These expectations aren't solely the result of individual interface decisions. They're the cumulative effect of technologies, business models and social behaviours developing together.

But interface design plays a significant role in making those expectations feel natural.

The way a product presents information influences what we notice.

The way it structures choices influences how we make decisions.

And the way it provides feedback influences what we learn to expect from technology.

Over time, digital conventions become familiar enough that we rarely question them.

Navigation menus, search fields, notifications, infinite scrolling and gesture-based interactions have become part of a shared language.

In a relatively short period, we've developed an entirely new collection of learned behaviours.

Which makes the next transformation particularly interesting.

We've spent decades learning how to speak to computers

For much of computing history, successful interaction has depended on people understanding the language of the interface.

We learned where to click, what icons represented, how menus were organised and which sequences of actions would produce the outcomes we wanted.

Even as interfaces became more intuitive, the underlying relationship remained relatively consistent.

The person provides an instruction.

The system processes it.

The interface communicates the result.

Designers helped translate between human intentions and computational operations.

And over decades, we became increasingly sophisticated at doing so.

We established conventions, interaction patterns, accessibility standards and design systems that make digital products more predictable and understandable.

But artificial intelligence is beginning to challenge some of those conventions.

Instead of navigating through several screens to complete a task, we can increasingly describe the outcome we want.

Instead of manually searching through documents, we can ask a system to identify, compare and synthesise relevant information.

Instead of performing every individual action, we can delegate parts of a process to software capable of planning and executing steps on our behalf.

This doesn't mean buttons, forms and conventional interfaces are about to disappear.

For many tasks, explicit controls remain faster, clearer and more reliable than conversation.

But AI introduces a different interaction model alongside them.

We're moving from telling computers exactly what to do towards telling them what we want to achieve.

And that distinction has significant implications for design.

A new kind of unfamiliarity

There's an interesting parallel between the arrival of graphical interfaces and the emergence of AI-driven products.

Early graphical interfaces borrowed physical metaphors because people needed help understanding unfamiliar computers.

Today, we face another period of unfamiliarity.

Only this time, the challenge isn't simply understanding how to operate the interface.

It's understanding the capabilities and limitations of the system behind it.

A traditional interface generally presents a defined collection of possible actions.

A button performs a particular function. A form requests particular information. A navigation menu provides a relatively predictable set of destinations.

AI-mediated interfaces can be less deterministic.

A natural-language request may produce different outputs depending on its interpretation, available context, underlying model and connected tools.

A confident response may still contain errors.

A system may complete a task successfully without making every intermediate decision visible to the person using it.

And when an AI system can take actions on someone's behalf, the consequences of misunderstanding an instruction become more significant.

This creates a difficult design problem.

An interface can appear remarkably simple while concealing considerable complexity.

A single text field might replace an entire collection of menus and controls.

But removing visible complexity doesn't necessarily remove the underlying complexity.

Sometimes, it makes that complexity harder to recognise.

What happens to UX/UI when the interface starts making decisions?

Don Norman's principles remain relevant here.

People still need meaningful feedback.

They still need understandable conceptual models.

They still need to recognise what actions are possible and what consequences those actions might produce.

Nielsen's principles around system visibility, consistency, error prevention and user control remain equally important.

But AI introduces new questions about how these principles should be applied.

What does useful feedback look like when a system is generating an answer rather than retrieving a predetermined result?

How should an interface communicate uncertainty without overwhelming someone with technical information?

When should an automated system request confirmation?

How can people inspect, correct or reverse actions taken on their behalf?

And perhaps most importantly, how do we help people develop appropriate trust in systems whose behaviour they cannot fully predict?

These aren't simply technical challenges.

They're questions about human understanding, psychology and agency.

A well-designed AI product shouldn't require someone to understand every mathematical operation occurring beneath its interface.

That would be neither practical nor necessary.

But people should understand enough about what the system is doing, what it can reliably accomplish and where human judgement remains necessary.

The responsibility of design is shifting towards making those boundaries visible.

The next evolution of digital design

I don't think the future of UX/UI is simply about making interfaces more conversational.

Nor do I think it means abandoning the visual and interaction principles we've spent decades developing.

The more interesting shift is in what designers are being asked to make understandable.

Historically, much of interface design focused on helping people operate systems.

With increasingly capable AI, designers must also help people evaluate, supervise and collaborate with systems.

That could mean designing clearer explanations of automated decisions.

Creating interaction patterns that distinguish suggestions from confirmed actions.

Making uncertainty and limitations understandable.

Building meaningful opportunities for correction, intervention and human approval.

Or designing experiences where traditional controls and natural-language interactions complement one another rather than compete.

The role of UI may evolve, but it won't become irrelevant.

If anything, the need for thoughtful interaction design becomes greater when a system can act with more independence.

A beautifully designed interface that conceals uncertainty or removes meaningful control isn't necessarily a good experience.

And an AI product that can complete extraordinary tasks but leaves people unable to understand or challenge its behaviour presents a serious design problem.

The goal shouldn't be to make technology feel intelligent at any cost.

It should be to make increasingly intelligent technology genuinely useful, understandable and accountable to the people using it.

We didn't invent UX. And we haven't finished defining it.

The history of UX/UI isn't a neat progression from physical objects to digital screens and eventually artificial intelligence.

Human factors, industrial design, psychology, information science and computing have developed through overlapping histories, each contributing to how we understand human interaction.

But there's a common thread running through them.

The relationship between what people intend to do, what technology allows them to do and how they understand the outcome.

Physical objects taught us the importance of clear interactions.

Graphical interfaces borrowed those lessons to make computing accessible.

Digitisation expanded the possibilities of interaction, removing constraints while introducing new forms of measurement, influence and responsibility.

Now AI is challenging the conventions we've built around digital systems themselves.

And perhaps the most interesting part is that we've arrived at a familiar problem.

Once again, people are encountering technology whose capabilities aren't immediately obvious.

Once again, designers are being asked to bridge the gap between human understanding and technological complexity.

But this time, the systems we're designing may be capable of interpreting intentions, generating information and making decisions on our behalf.

The challenge is no longer just making technology easier to use.

It's preserving human understanding, trust and control as technology becomes increasingly capable of acting for us.

Because the history of UX/UI has never really been about screens, buttons or visual interfaces.

It's been about the relationship between humans and the technologies we create.

And that relationship is still being designed.