Algorithms say more about what can be done with data than the other way around. They’re models fitted to data, but now the data and the world are being shaped to fit the model. The loop keeps tightening: systems learn from people, then reshape the environment so people behave in more predictable ways. There’s overfitting the model, and then there’s overfitting the world. A/B testing once helped fine-tune click rates or engagement; now it fine-tunes behavior. Feedback loops that were meant for optimization drift toward addiction like with most new generation of social media programs. Surge pricing, content feeds, affiliate rewards, and small discounts that keep you around a bit longer all steer behavior in ways that help the model learn faster and earn more for the companies behind it.
When multiple vendors follow the same algorithm, that’s cooperation without talking. Collusion without a meeting. Imagine everyone in a market following the same “independent pricing tool.” The coordination happens automatically. The algorithm becomes the shared rulebook, and competition turns into synchronized motion. You can already see this online. Prices pulse between a few tested levels. Large discounts have disappeared because the model decided they’re no longer needed. Retailers adjust prices by region, income level, or browsing history; I’ve seen differences between the Netherlands and Germany from the same seller. Price discrimination today isn’t about supply and demand anymore; it’s about prediction. The system learns how much you can tolerate, not what something costs. Online shopping grew because of convenience and trust, but now those same systems use their power to shape markets and customer behavior directly.
Even advertising feels like it’s running out of road. People have learned to tune it out, while bots keep clicking away, inflating metrics that no one believes. The testing continues, but the data has drifted too far from what’s real. Maybe this is why ads feel less like communication and more like a background process keeping the system alive, it once did with TV and radio networks and now with every streaming service out there including when we pay for it. Some already predict that the online personalized advertising does not really work if metrics are not gamed. Eventually it will collapse when the advertisers figure it out from the data and pull their ad budgets from the platforms.
If algorithms can’t find enough clean data, they start to make the world cleaner for themselves. They shape what people see, what they choose, and what they buy, just to make the data more consistent. That’s when optimization stops being analysis and becomes manipulation. When everyone follows the same models, the models stop being neutral and define the world as monocultural. They rewrite the conditions they measure. I am glad that the future LLMs have that monocultural data as their training data.