Making Decisions With No Data

Have you ever looked at the sky and thought how the heck did the Ancient Greeks see a monster out of a few, random bright dots?

Let’s strip away the mystique for a moment, and what we have is just people hallucinating patterns in randomness. Take scattered, unrelated points (stars light-years apart with no physical connection), ignore many things like three-dimensional space, perspective, and basic logic, oversimplify it into a 2D shape that aligns with your own system of beliefs, and declare it a divine truth. Now, just replace “stars” with “training data” and “mythology” with “neural network” or “data-driven model”, and you’ve got modern overfitting. Both processes involve stitching randomness into a narrative so convincing that you forget it’s fiction. The ancient folks didn’t have the internet, but they had something better: TikTok-free lives, time, and an unlimited imagination. These constellations became the backbone of “science” for centuries. Sailors navigated by them (All models are wrong but some are useful, I guess). Kings waged wars based on their alignment. All because nobody stopped to say, “Hey, maybe that ‘Great Lion’ is just six effing stars that’ll stop looking like a “lion” in 10,000 years when their positions drift?”.
If constellations are humanity’s first overfit model, astrology is the half-baked “insight” built from the wrong model. It’s the ancient equivalent of training a model on 12 data points (the zodiac signs) and claiming it predicts human behavior. Astrology’s premise is simple: the position of planets and stars at your birth determines your personality. This means, billions of humans, each with unique lives, are reduced to twelve zodiac signs based on which constellation the sun appeared to be in when they were born. The output? Sunday newspaper telling you that you are “adventurous but struggle with commitment”. Astrology mistakes coincidences for causal relationships. Jupiter was in retrograde when you failed your driving test? Sure, blame the gas giant instead of your parallel parking skills. Horoscopes, by being so vague that they could apply to anyone, are a classic sign of a model that’s all variance and no substance. Ever met a “typical Scorpio” who defies every Scorpio stereotype? Of course you have. I am a Scorpio by the way.
Astrology’s ability to survive to these days is a constant reminder of humanity’s addiction to overfitting. We’d rather believe the universe is a bespoke storyteller than admit that sometimes, things just happen randomly.
Nowadays, we’ve traded star charts for GPUs¹ and market tickers, but our habit of seeing patterns in chaos remains. Machine learning engineers now fight the same battle ancient shepherds did: restraining the human urge to turn noise into a narrative. The tools have changed, but the stupidity hasn’t. The same psychological flaws that made Babylonians see omens in comets make data scientists see “insights” in manipulated correlations. We’re hardwired to prefer a wrong answer over no answer; a survival mechanism that backfires in the age of big data.
The solution to overfitting is simple: humility. Acknowledge that not every blip in the data is meaningful. That we need to collect a convincing amount of data to make informed decisions. But humility is boring. Why admit ignorance when you could instead deploy your overfit model and blame “edge cases” when it fails? Astrology, for all its absurdity, at least owns its delusions. Modern tech culture markets overfitting as “actionable data”. Ancient overfitters at least had an excuse: they didn’t know about space, gravity, or the scientific method. But today, with all our knowledge, we still cling to the same fallacies. Overfit models ruin credit scores and lead companies to chase the wrong products because of flawed market forecasts².
Somewhere, a future civilization will dig up our TensorFlow code and laugh, just as we laugh at the idea of Scorpios being “passionate”. The lesson? Whether we’re mapping stars or training GANs, let’s remember: the universe is under no obligation to make sense to us. Sometimes a cluster of stars is just a cluster of stars. Sometimes a data spike is just a glitch. And sometimes, my personality isn’t “Gemini energy”—it’s just me being an asshole. So next time we’re tempted to see meaning in the noise, let’s step away from the telescope and touch some grass. The cosmos will carry on.

1.

2.