Most people look for lost keys under the streetlight—not because that's where they dropped them, but because that's where the light is. Science has the same problem, sometimes called the 'streetlight effect': researchers gravitate toward already well-studied areas where equipment, references, and funding are readily available. The brightest minds crowd under a few lamps, while the dark fields where truth actually hides go unexplored.
But a May 2026 paper in *Nature*—《Artificial intelligence redirects collective attention toward novel scientific research》 (Artificial intelligence redirects collective attention toward novel scientific research) by Sun et al.—uses hundreds of thousands of data points to reach a counterintuitive conclusion: AI is pushing human scientists out of their comfort zones and into the unknown.
Winter under the Streetlight
Before AI intervened, structural biology was suffering a kind of aesthetic fatigue. Researchers tracking 245,396 experimental structures in the Protein Data Bank (PDB) found a discouraging trend: over time, scientists' interest in studying new proteins steadily declined. More and more, people preferred to circle within known protein families, doing incremental patchwork.
The root cause is that exploration costs are too high. Studying a completely unfamiliar protein could consume a PhD student's five years and still yield nothing.
AI fires a flare
Then came AlphaFold2 and its successors.
The authors found that AI does not merely parrot existing data, as many feared. Instead, it acts as a confidence booster.
When AI produces a high-confidence structure prediction for a human gene that has never been studied, everything changes:
1. Risk disappears: The dark territory is illuminated. Studying an 'unpopular' protein is no longer career suicide, because AI has already provided the blueprint. 2. Attention reallocation: After AlphaFold's release, the share of research on novel proteins rebounded significantly. The shift was most pronounced in papers that cited AI predictions.
In short, AI used its computational power to fill in the pits that humans once had to risk their careers exploring.
The chain reaction of knowledge
The most moving part of the paper is its demonstration that the joy of discovery is contagious. Analyzing nearly 250,000 downstream papers, the researchers found that this AI-guided 'attention shift' spread like dominoes: when structural biologists began studying obscure proteins, downstream drug development and pathology research followed.
This strongly rebuts the famous worry that 'AI will make science more conservative.' The evidence shows that AI is not reinforcing the old classics—it is aggressively expanding the frontier.
Why this paper matters
It reveals a new law of intelligence: true wisdom lowers the barrier to exploring truth.
In the past, humans chose mediocrity out of fear of failure; now, AI provides 'confident foresight' that gives people the courage to take risks.
Summary: Tools don't just extend our hands and feet—they reshape our ambitions. AI is driving us away from those dying streetlights and toward the once-daunting, now gold-strewn dark forest, where countless AI-lit coordinates await the first line of truth written by human hands.
Don't worry about being replaced by AI. Worry instead about whether, when the whole forest is lit up, you still hold on to that most primal curiosity—the instinct to go looking for your keys.