Ah, models.
Artificial Intelligence is not the first generation of models I’ve worked with. Far from it.
I still remember the days before Excel transformed the spreadsheet world. Back then there was VisiCalc, the grandfather of modern spreadsheet software. By today’s standards it was primitive. The models we built were crude, narrow in scope and laughably limited by modern computing standards.
Yet those limitations turned out to be one of their greatest strengths.
The machines simply lacked the computational power to build castles in the air. Every additional calculation carried a cost. Every variable had to justify its existence. Every assumption had to earn its place because memory was precious, processing power scarce and complexity came with very real penalties.
The models remained close to reality because they had very little room to wander away from it.
Then computing power exploded.
What once took hours became seconds. Memory limitations disappeared. Processing power became effectively free. Suddenly models could incorporate hundreds, then thousands, then millions of variables. Complexity ceased to be something to avoid and instead became something to celebrate.
And that is where a subtle but profound problem emerged.
Distance.
Abstraction.
The model slowly drifted away from the thing it was supposed to describe.
The carefully selected handful of verified inputs became enormous mountains of data gathered from every conceivable source. Some of it was measured. Some estimated. Some inferred. Some statistically reconstructed. Some copied from other models. Some based on assumptions. Some built around policy objectives. Some reflecting little more than educated guesses.
And then there is what I like to call wishdata.
Wishdata is not simply bad data.
It is data that exists only because people want it to exist.
It reflects aspirations rather than observations. Political objectives rather than measurable reality. Institutional assumptions rather than empirical evidence. Forecasts treated as facts. Model outputs recycled as inputs for other models until speculation slowly acquires the appearance of certainty.
It survives not because reality continuously confirms it, but because people continuously feed it back into the system.
No model can correct that.
No amount of mathematical sophistication can rescue defective foundations.
The oldest rule in computing never disappeared.
Garbage in.
Garbage out.
Artificial Intelligence has not escaped this law.
It merely operates on a vastly larger scale.
People often speak about AI as though it possesses some mysterious ability to transcend the quality of its inputs. It does not. An AI model is only as reliable as the information it receives, the assumptions embedded within that information, and the discipline exercised by the people using it.
I work with AI every single day.
It has become an extraordinarily useful tool.
It helps me research.
It helps me organise ideas.
It helps me write.
It helps me challenge my own thinking.
But I also discard an astonishing amount of what it produces.
Not because I dislike the answers.
Not because the machine is stupid.
But because I know perfectly well that I asked the wrong question.
I wrote an ambiguous prompt.
I failed to define the boundaries.
I unintentionally pointed the model in the wrong direction.
The mistake was mine.
That is simply part of learning to use the tool properly.
The problem is that many people find this remarkably difficult to admit.
They assume that because a model produced an answer, the fault must lie with reality rather than with the assumptions that shaped the question.
That has always been the danger of models.
Long before AI.
Long before machine learning.
Long before neural networks.
The more sophisticated the model becomes, the greater the temptation to believe that its output is reality itself rather than an interpretation of reality.
The map quietly replaces the territory.
The spreadsheet replaces the factory.
The simulation replaces the experiment.
The forecast replaces observation.
And eventually, if we are not careful, the model stops describing the world and starts describing only itself.
Artificial Intelligence did not create that problem.
It simply inherited one that has been growing ever since we first discovered that reality could be squeezed into rows, columns and formulas.
The technology has changed beyond recognition.
Human nature has not.