The Moment AI Changed

For most businesses, artificial intelligence arrived long before ChatGPT.

It was deciding which products appeared in recommendations, detecting suspicious transactions, ranking search results, optimising advertising and forecasting demand.

Most of us just didn’t experience it as AI.

It lived somewhere inside the software.

Then, almost overnight, we could talk to it.

That distinction might turn out to be much more important than it first appears.

When OpenAI released ChatGPT publicly in late 2022, artificial intelligence stopped being something most people encountered indirectly. Suddenly, millions of people could sit in front of a blank text box and interact with a remarkably capable AI system using nothing more complicated than ordinary language.

The temptation is to think ChatGPT itself was the breakthrough.

I think the bigger story is what it revealed.

Artificial intelligence was becoming a general-purpose capability that ordinary people and ordinary businesses could actually use.

And once that happens, the question changes.

It is no longer simply what AI can do.

It becomes what happens when almost everyone can access it.

AI didn’t suddenly appear#

The excitement around generative AI can make it feel as though artificial intelligence materialised in 2022.

It didn’t.

Businesses had already been extracting value from machine learning for years. Recommendation engines, fraud detection, predictive maintenance, computer vision, demand forecasting and optimisation systems were established applications long before anyone asked ChatGPT to write an email.

But there was an important limitation.

Most AI was something businesses built into systems, rather than something most people could interact with directly.

Using it meaningfully often required engineers, data scientists, specialist models, structured data and a clearly defined problem.

That naturally limited where it appeared.

Generative AI changed the relationship.

Instead of building a specialist interface for every task, people could increasingly describe what they wanted using the interface humans already understand best: language.

Write this.

Explain that.

Analyse this.

Summarise these documents.

Give me ten alternatives.

Find the problem in this code.

Turn these notes into a proposal.

The technical systems underneath remained enormously complicated.

The experience of using them suddenly wasn’t.

Language changed the interface#

There is something easy to underestimate about a blank text box.

Almost everybody already knows how to use one.

Traditional software requires us to learn the structure created by the people who built it. There are menus, fields, buttons, commands and workflows. The user has to understand, at least partially, how the software expects a task to be performed.

Generative AI begins to invert that relationship.

Increasingly, we can describe the outcome we want and allow the machine to determine some of the steps required to produce it.

That doesn’t mean conventional software disappears. Nor does it mean today’s AI systems can reliably perform everything we ask of them.

They can’t.

But it dramatically reduces the distance between having a technological capability and being able to access it.

And that matters commercially.

A technology can possess extraordinary capabilities and still have limited economic impact if deploying those capabilities is difficult, expensive or requires rare expertise.

Accessibility changes the size of the opportunity.

By April 2023, McKinsey found that 79% of respondents to its global AI survey had already experienced generative AI either at work or outside it. Twenty-two per cent were regularly using it in their own work. One-third said their organisations were already regularly using generative AI in at least one business function.

This was a technology that had barely entered mainstream consciousness.

The breakthrough wasn’t simply that AI became more capable. It was that sophisticated capability became dramatically easier for ordinary people to access.

The capability was changing too#

Accessibility alone wouldn’t matter very much if the technology behind the interface wasn’t improving.

It was.

When OpenAI released GPT-4 in March 2023, it demonstrated a significant improvement over GPT-3.5. GPT-4 could accept both images and text as inputs, and OpenAI reported human-level performance across several professional and academic benchmarks.

One example was particularly striking. On a simulated bar examination, GPT-4 performed around the top 10% of test takers. GPT-3.5 had performed around the bottom 10%.

Benchmarks need treating carefully.

Passing an examination does not make a machine a lawyer, and performance in a controlled test doesn’t automatically translate into reliable performance inside a business.

AI systems still make mistakes. They can confidently produce inaccurate information. They struggle with tasks humans find straightforward. Reliability, privacy, security and governance all present genuine problems.

But the direction matters.

The systems weren’t simply becoming easier to use.

They were becoming more capable at the same time.

That combination is what makes this moment interesting.

Capability is increasing.

Access is widening.

The number of potential applications is expanding.

And businesses can experiment with all three simultaneously.

Almost every business can experiment#

This is where I think the implications become particularly significant for smaller and established businesses.

Previous waves of advanced technology often came with substantial barriers to experimentation.

You needed infrastructure.

You needed specialist people.

You needed capital.

Sometimes you needed all three before you could even discover whether the technology was useful.

The barrier to experimenting with generative AI can be almost absurdly low by comparison.

A CEO can use it.

A salesperson can use it.

A developer can use it.

Someone in finance, marketing, customer service or operations can use it.

They don’t need to understand neural networks.

They don’t need to train a model.

They don’t even necessarily need their company to have an AI strategy before they can discover something useful.

That creates an unusual dynamic.

Experimentation can begin almost anywhere in an organisation.

Enterprise deployment is considerably more complicated than opening a browser and typing into ChatGPT. Security, privacy, proprietary data, integration, reliability, governance and workflow design all matter.

But experimentation comes first.

And the number of people capable of experimenting has exploded.

McKinsey’s 2023 survey showed how quickly this was already reaching company leadership. Nearly a quarter of surveyed C-suite executives said they personally used generative AI for work, while 28% of respondents at organisations already using AI said generative AI was on their board’s agenda.

That is an extraordinary transition in a remarkably short period.

This is bigger than ChatGPT#

There is a danger that we mistake the interface through which most people first encountered generative AI for the technological shift itself.

ChatGPT matters enormously.

But AI is not a chatbot.

The underlying capabilities can be embedded inside existing products, software, workflows and entirely new businesses.

They can analyse text.

They can generate it.

They can write software.

They can work with images.

And increasingly, different forms of information can be processed by the same underlying systems.

This is why I think businesses should resist categorising generative AI as simply another piece of software to purchase.

Microsoft Excel is a product.

Salesforce is a product.

ChatGPT is a product.

Artificial intelligence is a capability.

And capabilities can appear everywhere.

Electricity wasn’t important because somebody invented a better lamp. Its economic significance came from the enormous range of things that became possible once electrical power could be applied throughout the economy.

The internet wasn’t important because of one website.

Cloud computing wasn’t important because of one application.

It is too early to know whether AI ultimately belongs in precisely the same category of general-purpose technology. There are still significant technical limitations, unanswered economic questions and plenty of hype.

But businesses should at least consider the possibility.

Because if AI continues becoming simultaneously more capable, more accessible and easier to apply, its eventual impact won’t be defined by whichever chatbot currently has the most users.

It will be defined by everything people build with the underlying capability.

The economics are what matter#

Technology industries naturally focus on capability.

Businesses eventually have to focus on economics.

Can this help us produce something that previously required considerably more time?

Can we understand information we previously couldn’t process economically?

Can someone perform work that previously required specialist knowledge?

Can we improve a decision?

Can we serve a customer better?

Can we remove a constraint?

Can we create something entirely new?

Those are much more interesting questions than whether a business “uses AI”.

And they point towards an important distinction.

The opportunity isn’t valuable because the technology is impressive. It is valuable when the technology changes the economics of doing something useful.

That could mean reducing the cost of an existing process.

But it could also mean increasing revenue, making better decisions, improving customer experiences, allowing people to perform work they couldn’t previously perform, or making an entirely new product economically possible.

We are still very early in discovering where those opportunities exist.

And that’s precisely why businesses should be paying attention.

The interface isn’t the revolution#

There is a tendency to recognise technological shifts more clearly in retrospect.

At the beginning, we focus on the products.

Later, we understand the underlying change.

The early web looked like websites.

Mobile looked like smartphones and apps.

Cloud computing initially looked like somebody else’s servers.

Generative AI currently looks a lot like chatbots.

Perhaps that is ultimately what this becomes: an important but bounded class of software tools.

But I don’t think businesses should make that assumption yet.

Something unusual has happened.

AI systems have become capable enough to perform a growing range of useful intellectual tasks, accessible enough that ordinary people can interact with them directly, and flexible enough that businesses can begin applying the same underlying capability to very different problems.

Those things are happening at the same time.

That is why this moment matters.

The interesting part of the AI story may not be that machines learned to talk to us. It may be that intelligence just became something businesses can build with.