In 2008, Wired editor Chris Anderson declared the scientific method obsolete. With petabytes of data and powerful algorithms, he argued, scientists no longer needed hypotheses or theories. Just feed the data into the machine and let the patterns emerge. Numbers would speak for themselves.

It was a provocative claim, and more than a decade later, it remains contested. Machine learning now drives discoveries in genomics, particle physics, and climate science. But has the logic of scientific inquiry actually changed? Or are massive datasets simply amplifying methods scientists have always used? The answer matters for how we understand what science is and what it can tell us about reality.

The promise of big data is seductive: with enough information, patterns reveal themselves. No need for clever hypotheses or elegant theories. Just collect, compute, and discover. This approach has produced genuine successes. AlphaFold predicted protein structures by learning from databases of known examples. Recommendation algorithms find connections humans would miss.

Yet a closer look reveals theory hiding in plain sight. Choosing which data to collect requires assumptions about what matters. Defining variables means carving the world at particular joints. Even the structure of a neural network reflects theoretical commitments about how learning should work. The data doesn't arrive pre-labeled by nature.

Karl Popper argued that observation is always theory-laden: we cannot see without a framework for seeing. Big data doesn't escape this condition; it multiplies it. Every dataset embeds prior choices about measurement, categorization, and relevance. The scientist who claims to follow the data is following theories they haven't examined.

Takeaway

Theory-free science is an illusion. The question isn't whether assumptions shape inquiry, but whether we make them explicit enough to criticize.

Big data excels at finding correlations. Given enough variables, patterns will appear—some meaningful, many spurious. The famous example: U.S. spending on science correlates almost perfectly with suicides by hanging. Mine enough data, and you find such pairings everywhere. They predict nothing useful because they explain nothing real.

Traditional science aims beyond prediction toward explanation. We want to know why phenomena occur, what mechanisms produce them, what would happen under counterfactual conditions. Knowing that smoking correlates with cancer was a beginning; understanding the cellular pathway through which carcinogens damage DNA is the achievement.

Machine learning models often deliver prediction without explanation. A neural network may forecast protein folding with stunning accuracy while remaining a black box. This is genuinely useful, but it's a different epistemic product than classical science offered. We trade understanding for capability—and sometimes that trade leaves us unable to know why our predictions work, or when they will fail.

Takeaway

Prediction and explanation are not the same achievement. A model that works without understanding is a tool, not yet a theory.

Algorithms feel objective. They don't get tired, don't have grudges, don't favor friends. Feed in the same inputs and you get the same outputs. This appearance of neutrality is part of why automated analysis carries scientific authority. The machine, we assume, just reports what's there.

But every algorithm encodes choices. Which features count as relevant? What counts as a successful prediction? How are errors weighted—is a false positive worse than a false negative? These questions have no value-free answers. A medical diagnostic system that minimizes overall error may systematically fail certain populations whose data was underrepresented in training.

Philosophers of science have long noted that scientific objectivity is achieved through social practices—peer review, replication, open criticism—rather than personal detachment. Algorithmic systems require similar scrutiny: examining training data, auditing outcomes, surfacing assumptions. Calling something data-driven doesn't exempt it from this work. It just hides where the values went.

Takeaway

Objectivity is not the absence of choices but the willingness to expose them to criticism. Code can conceal what conversation might reveal.

Big data hasn't replaced the scientific method—it has amplified its old questions in new forms. Theory still shapes observation. Correlation still falls short of explanation. Objectivity still depends on critical scrutiny rather than mechanical procedure.

What has changed is scale. With more data and more computation, we can do more—but we can also obscure more. The philosophical work of science remains: asking what our methods assume, what our results mean, and where reality might be pushing back against our models.