There is something slightly strange about the way Hollywood makes films today. We have reached a point where studios can analyse audience preferences, track streaming habits and use predictive algorithms to estimate which stories are most likely to succeed, sometimes before a camera has even been switched on. On paper, this sounds like progress. Why spend millions on a film nobody wants to watch when data can help identify what audiences enjoy? Yet I wonder whether, in our eagerness to understand what people want, we have become a little too comfortable giving them what they already know.

The endless sequels, familiar franchises and recycled storylines are not exactly new, but the technology behind filmmaking has made predicting audience behaviour increasingly sophisticated. Machine-learning models identify patterns in historical data, helping studios assess which genres, actors and narrative structures might generate returns. The problem is that recognising a successful pattern is not the same as understanding why it succeeded. When commercial decisions become too dependent on historical performance, experimentation begins to look like an unnecessary risk. Hollywood starts behaving less like a creative industry and more like an algorithm optimising for predictable outcomes.

Perhaps the most interesting way to understand this is through the concept of a feedback loop. Recommendation systems use our viewing history to suggest films we might enjoy, and our subsequent choices provide more data for future recommendations. This is not inherently harmful, but it can reinforce existing preferences at the expense of discovery. The Truman Show offers an interesting parallel: Truman’s carefully constructed world is designed around predictable behaviour, leaving little room for experiences beyond its boundaries. Real algorithms are hardly Truman’s producers, but the comparison raises an important question about how much our environments influence what we come to expect.

Yet some of cinema’s most memorable experiences emerge precisely from uncertainty. Before watching Inception, we might never have imagined that a complex narrative about dreams could become such an engaging cinematic experience. Its layered storytelling challenges audiences rather than simply reproducing familiar formulas. This illustrates a fundamental distinction in computer science: exploitation involves using what is already known to work, while exploration involves testing possibilities whose value remains uncertain. An industry that prioritises the former indefinitely risks overlooking the latter.

There is also a deeper problem with measuring success. Opening-weekend revenue, streaming completion rates and audience ratings are useful metrics, but they remain proxies for something far more complicated. Interstellar, for instance, offers spectacular science fiction while exploring time, grief and human connection. A model can measure engagement with a film, but those measurements cannot fully explain why its ideas might remain with someone for years. The issue is not that human responses cannot be studied; it is that no single metric can adequately represent artistic value.

I understand the logic. Studios have employees, investors and financial risks to consider, and data can help unconventional films find their audiences. But there is a difference between using technology to inform creative decisions and allowing measurable outcomes to determine the boundaries of imagination. Perhaps cinema’s greatest strength lies in its ability to change our preferences rather than merely satisfy them. The real danger is not that computers will eventually become capable of predicting what we want to watch, but that we might become so dependent on their predictions that we stop discovering what we never knew we wanted.