An algorithm is a set of rules that takes information, weighs it, assigns values, and makes a call. Sometimes it’s simple. Sometimes it’s dressed up in machine learning and enough math to make investors excited.
They chew through historical data, recent results, trends, weighted variables, guestimates, and whatever else you feed them. Then they try to tell you what’s likely to happen next.
The challenge is that history doesn’t sit still and while it rhymes, history rarely repeats.
Recent events get inherited and amplified. A hot streak starts looking like destiny. A bad loss leaves fingerprints on matches that haven’t happened yet. The model doesn’t know fear, pride, ego, or questionable decisions.
It just knows the numbers.
And because AI, algorithms, and predictive models are the shiny things in the news right now, I built one for the World Cup.
And by “one,” I mean an algorithmic predictive model. Not AI. I’ve got neither the patience nor the hardware to build that kind of beast.
Side note: Yes, it can be argued that AI is just an algorithmic predictive model written large, dressed in more data, more layers, more computing power, and better marketing. That’s a fair argument. I hear you. But when most people hear AI, they're thinking JARVIS. Not a math geek and history nerd sharing a dorm room.
It’s been fascinating to watch it develop.
That’s the thing about algorithms. They’re not perfect. At best they’re the embodiment of that old line from John Wayne in Trouble Along the Way: “The race isn’t always to the swift nor the battle to the strong...but if you are a betting man, that’s the way to bet.”
That’s what the model does. It weighs recent history against past performance, looks at scoring, quality of group, form, and matchup strength, then makes a prediction with its python powered heart.
By the time group play ended, the algorithm had it dialed in.
It missed Germany (55.6%) and Paraguay, though it had the match closer than I would have (70/30). It missed Australia (52.5%) and Egypt. Fair enough.
But in the Round of 16, so far, it’s been clean.
France over Paraguay, 89.2%. Morocco over Canada, 85.4%. Spain over Portugal, 61.9%. Belgium over the United States, 64.7%. Norway over Brazil, 63.4%, which I thought was too optimistic until Norway went out there and made the algorithm look like a genius. England over Mexico, 60.7%.
At the time of this writing, two matches are still waiting: Argentina (85%) over Egypt. Colombia (50.5%) over Switzerland.
The second one is barely a prediction at all. More like a coin spin with the algorithm calling it tails because the head side of the coin weighs more.
For the curious, the model has France (65.5%) winning the whole thing over Argentina.
What fascinates me, and annoys me in equal measure, is that while the algorithm has gotten pretty good at picking winners and very good at predicting the scoring differential.
But it still can’t predict the actual score.
It picked Norway over Brazil by one goal. Got that right.
But it said the most likely score was 3-2, at 7.5%.
The actual score, 2-1, came in at 0.9%.
For some additional context, the model thought Brazil beating Norway 5-3 (5.4%) was more likely than Norway winning 2-1 (0.9%).
Not a little more likely. A lot more likely.
So the algorithm saw Norway. It saw the win. It saw the one-goal margin. It traced the outline of the thing almost perfectly. Then, when asked to give the score, it coughed up smoke and called it a day.
That’s probably the lesson.
Algorithms can read the tracks. They can smell the rain coming. They can get the shape of it, but the specifics, contrary to what Hollywood and venture capitalists would have us believe, remain out of reach.
Disclaimer: Someone with deeper statistical knowledge, more experience building predictive models, better data, and more computing power could almost certainly tune this better than I did. They could probably account for volatility, referee weirdness, matchup quirks, and some of the misses my model made along the way.
But I still stand by the larger conclusion.
Algorithms can help you see the shape of what’s coming. They just can’t always give you the specifics.