I had a conversation with a colleague yesterday about AI and how fast everything seems to be moving. The sense that this time feels genuinely different. That we’re in the middle of something exponential.
It reminded me of a piece I wrote back in 2014. The argument then was simple: despite all the noise about how “everything is accelerating”, the data showed that internet, PCs, mobiles, smartphones and tablets all followed remarkably similar S-curves once you accounted for their actual starting points. The breathless “it’s all happening so fast” rhetoric was mostly bollocks.
So are we genuinely experiencing something different this time? Or are we just back on the steep bit of another curve?
The thing about exponential curves is that when you’re on the steep part, it genuinely does feel like everything’s happening at once. The previous decades of slow progress vanish from view. All you see is the sudden uptick.
The long nose
The trick with any S-curve analysis is working out where to start measuring. ChatGPT launched in November 2022, hit a million users in five days, and has been presented as something that appeared from nowhere.
Except it didn’t.
Neural networks trace back to McCulloch and Pitts in 1943. Rosenblatt’s perceptron arrived in 1957. Backpropagation was refined through the 1970s and 80s.
The specific lineage of LLMs runs through attention mechanisms and transformers (2017), GPT-1 (2018), iterative improvements through GPT-2 and 3. ChatGPT was the accessible interface, not the invention. It’s the iPhone moment – a recombination of existing technologies that made something previously awkward suddenly usable by normal humans.
In W. Brian Arthur’s book The Nature of Technology, he describes technologies as combinations of previous technologies. They emerge from assemblies of earlier solutions. The “new” thing is just the visible tip of decades of work.
If we set the starting point for neural networks at 1943, we’re looking at 80 years from concept to consumer product. But moreover, you’ve been using AI for ages. You just didn’t call it that.
Spotify recommendations. Netflix suggestions. Google Translate. Gmail’s Smart Compose. Grammarly. Your GPS working out the fastest route. Fraud detection in your banking app. Spam filters.
All of these are machine learning. All of these are AI. They’ve been quietly shaping your behaviour for years without anyone declaring a revolution. The panic only arrived when someone gave it a chat interface and called it artificial intelligence.
Tom Standage made a similar point about social media in “Writing on the Wall” – we’ve had social networks since Roman graffiti. Facebook didn’t invent social networking, it just made it more accessible. The technology changes, but the human behaviour is remarkably consistent.
The weight of the past
LLMs didn’t emerge from thin air. They required a stack of previous technologies, each with its own development curve: electricity generation, semiconductors, GPUs, internet infrastructure, cloud computing, decades of NLP research, and transformer architectures. Each layer took years. Each had its own S-curve.
But it’s not just the technology stack that matters. It’s everything else.
For example, London’s road network follows patterns set down by the Romans and medieval street planners. Our cities are still shaped by railway decisions made in the 1840s. Tom Standage’s “A Brief History of Motion” shows how each transport revolution is constrained by the infrastructure of what came before. You can’t just “start fresh”. The past exerts enormous weight on the present.
Accounting practices matter too. Double-entry bookkeeping, invented by Venetian merchants in the 14th century, still shapes how organisations think about value, cost, and investment. Want to understand why traditional companies struggle to think about AI? Look at how their accounting frameworks force them to treat it as capital expenditure or operational cost – categories that made sense for industrial machinery but little for training models.
The point: technical capability doesn’t automatically translate to change. The past creates path dependency. You can’t just announce a revolution and expect everything to shift overnight.
Technology can only move as fast as the people adopting it
The disconnect between technical capability and actual behaviour change is massive.
Email existed for decades before offices stopped using memo formats. Video calling was technically possible long before the pandemic, but it took forced circumstances to shift behaviour. Social networks in the BBS and Usenet era had many of the features of modern platforms, but it took another 20 years for that behaviour to become mainstream.
Standage documents this with the telegraph in The Victorian Internet. When it launched, it was thought that the telegraph would eliminate war, create world peace, and revolutionise commerce. The technology did matter – but not in the ways people predicted. The actual changes took decades and looked nothing like the initial hype.
We consistently overestimate short-term impact and underestimate long-term transformation.
In “To Save Everything, Click Here”, Evgeny Morozov captures the core problem: technological solutionism – the idea that every problem can be solved by the right app or algorithm. The current AI hype is peak solutionism. Can’t write? Use AI. Can’t think? Use AI. Can’t manage people? Use AI.
But organisational change requires more than new tools. It requires changes to power structures, social norms, regulatory frameworks, institutional arrangements. Technology is just a part of a sociotechnical system, and these move glacially. LLMs will be shaped by existing power structures, not just technical potential. This acts as a natural dampener on any curve.
So what does this mean for you?
If you’re worried about being left behind, here’s what the S-curve pattern tells us: you have more time than the panic merchants suggest. But “don’t panic” isn’t a strategy. Here’s what to do with that time.
First, get clear on which quadrant you’re actually operating in. I’ve written elsewhere about the AI Play Matrix – a framework for thinking about where AI innovation actually happens. Most organisations want to live in the “Plan” quadrant, where they know the problem and know the solution. But actual innovation happens in the “Tinker” and “Adapt” quadrants, where you’re exploring what’s possible rather than scaling what’s proven.
The technology vendors want you in the Plan quadrant because that’s where they can sell you certainty. “Implement our AI platform and transform your business.” But if everyone’s using the same platforms to solve the same problems, where’s the competitive advantage?
Second, focus on the skills that matter for AI exploration rather than AI engineering. That means:
Getting comfortable with uncertainty. If you’re working with genuinely new capabilities, you won’t know what will happen. That’s the point. The pressure to have detailed project plans for emerging tech is real but pointless.
Prototyping rapidly. Turn ideas into things people can interact with as quickly as possible. The only way to know if an AI idea works is to test it with real users on real problems. Learning, not perfection.
Building diverse learning networks. The best insights come from unexpected places. A retail person solving inventory problems might crack your HR challenge. Look beyond your immediate industry.
Third, stop asking “how fast is this happening?” and start asking “what are we actually trying to achieve?” Strip out the buzzwords. If the AI component isn’t crucial to the outcome, you probably don’t need AI. And if you’re just using AI to do slightly better versions of what you already do, you’re missing the point.
The S-curve pattern from every previous technology wave tells us that it will take longer than the hype suggests, work differently than predicted, and the real changes will come from unexpected directions. Use that knowledge.
Respect the long nose – the decades of work that made this moment possible. Prepare for the long tail – the years it will take for genuine transformation. And in between, experiment in the bottom half of the matrix where actual innovation happens, even though your organisation’s procurement processes, governance models and risk frameworks weren’t designed for it.
When technology companies tell you “everything is changing instantly,” remember: they’re selling something. When anyone talks about unending exponential growth, ask them when the J will turn into an S.
The steep bit will give way to the plateau. The panic will subside. And we’ll look back and wonder what all the fuss was about.
Again.
mmitII 