AI optimization leads to average, bland outputs, erasing uniqueness and delight.

Yet people instinctively prefer imperfect, human-made works over flawless AI outputs.

Preserve imperfection: create before using AI, and consciously keep quirks and flaws.

Slice open a supermarket tomato. Red to the edges, geometrically obedient, flavored like the faint rumor of tomato. Breeders nowadays “built” it for the truck, the shelf, the long wait under fluorescent light; they are optimized to offend nobody. Keep that tomato image at the back of your mind as you read on—the words, pictures, songs, and videos crossing our screens are ripening with the same logic.

The Likeliest Everything

Optimization has one favorite move: pick the likeliest next thing. The likeliest word, the likeliest pixel, the likeliest chord. Likeliest means most common, and most common means average. Every polished output drifts toward the same middle, the highest common denominator, with every corner sanded and every seam hidden. Each piece arrives wearing a fresh outfit over the same skeleton.

Researchers have begun to measure that polishing process. In a Science Advances study, writers were asked to compose short stories with and without generative AI. Writers with the most access to AI tools produced stories scoring higher for novelty and usefulness—their deliverables also resembled one another. While the individual stories improved on average, the shelf as a whole deteriorated as it grew monotonous. It’s a modern version of a classic social dilemma: each writer has good reason to use the tool and benefits in the short run; the sum of those........

© Psychology Today