A playbook to curb the growing power of Big Tech.

We are approaching a credibility crisis with Scott Bessent and Kevin Warsh leading our economy.

A legal dispute over the future of Kalshi, Polymarket and others has drawn in the Trump administration, the president’s son and nearly every state attorney general.

Generative AI has long been accused of copying artists’ work outright (see the numerous copyright lawsuits winding their way through court). But a new study out of MIT makes a very different argument—one that could complicate how those cases hold up.

In the study, published in Nature, the MIT researchers Zheng Dai and David Gifford set out to to test whether a generated image can be traced back to a single piece of training data. They were looking specifically at diffusion models, the systems most often used for generating images and video.

Their finding? It comes down to how big the training data set is. They found that the more data a model is trained on, the harder it becomes to attribute its output to any particular piece of training data. In fact, when they removed a specific piece of training data from the set, they found that it had little to no bearing on the updated output of the model. As the researchers put it: “We can often omit any sample or creator from the training data without affecting a generated sample.”

This trend, what the researchers call “attribution decay,” means that artificial intelligence can produce an image that resembles a particular artist’s work, while having no provable causal link to that artist’s actual contribution to the training data.

The researchers suspect that this happens because larger datasets tend to contain a lot of visual redundancy; many different images share overlapping features, so no single image is responsible for the DNA of an AI-generated image. They got the same result again and again, across dozens of experiments. But they’re careful to treat this as their best explanation rather than something they’ve directly proven (more on that below).

How this works

To get to this finding, the researchers needed a way to test cause and effect precisely, because diffusion models don’t work like a database you can just delete files from. During training, a model doesn’t store copies of the images it sees. Rather, it adjusts millions of internal numerical settings based on patterns across the entire training set, and later uses those settings to turn random noise into a new image. That means you can’t simply delete one training image and expect a straightforward before-and-after comparison. The influences of all the training images are normally tangled together across the whole model.

Instead, the researchers built models out of separate components, each trained independently on a different slice of the data, then combined. That structure let them cleanly remove the influence of one artist, person, or image by switching off just the components that had been exposed to it, without retraining the entire model from scratch. They call this technique “ablation.”

AI image generators learn by studying huge numbers of pictures and learning to reconstruct them. Dai and Gifford generated an image using the full training set, then regenerated it with one artist, person, or image removed, while everything else remained the same. If the newly generated results didn’t change, they could surmise that the missing piece wasn’t the cause of that particular output; the model would have generated a nearly identical image regardless.

They also tested the standard shortcut people use for spotting AI copying: Finding the training image that looks most similar to an output and calling it the source. That shortcut got it wrong more often as datasets grew; the “closest match” was frequently coincidental. There wasn’t causality because removing it often changed nothing in the generated results. That effect undercuts a lot of existing “AI copied this artist” claims, which rely on visual similarity rather than actually testing cause and effect.

Why does this happen? The researchers offer an explanation, though they’re careful to call it a conjecture rather than a proven fact. According to their paper, they “conjecture that attribution decay happens because features that are important to model behavior are distributively and redundantly encoded throughout its training set.” 

The “unattributability” effect kicks in only at large scale. The paper states this effect is already significant at scales of 10,000 to 100,000 images, while many commercially deployed models train on datasets with up to a billion images.

The Warhol effect

To understand how this works in practice, let’s use Andy Warhol’s artwork as an example. Imagine a model trained on 50,000 works of art that happen to include Warhol’s silkscreens. If you delete Warhol’s specific images and regenerate, the output barely changes. This is not because the model “understands” Warhol’s style in some abstract sense, but because Warhol wasn’t the only artist doing bold flat colors, repeated grids, and pop culture subjects.

In fact, numerous other artists in that same dataset were doing visually similar things. That means that the visual features of thousands of images overlap enough that no one image is irreplaceable. In other words, Warhol’s paintings weren’t a unique ingredient; they were one of many sources that use the same visual pattern. Removing them would leave plenty of redundant signal behind for the model to draw from.

It’s important to note that the researchers are not arguing that AI models can simply create a Warhol-esque image out of thin air. If that same dataset had contained zero pop art, zero flat-color silkscreen work, and zero repeated-grid compositions, the model would have nothing to draw on to produce an image in that style.

Rather, they are saying unattributability kicks in when a style is redundantly present across many images. Delete one source of a common feature, and the feature survives through the remaining similar images. Delete every source of a rare feature, and the model loses the ability to produce it entirely.

What this means

These findings makes it more difficult to point to a specific artistic provenance, which could undermine artists’ arguments that AI copied their specific artwork. The same redundancy that lets a style survive one artist’s removal also makes it very hard to prove any single artist was the definitive source of a given output.

The peer-reviewed paper in Nature gives AI companies a sharp defense weapon in court, but not a shield. The researchers themselves tell about the legal stakes, with a careful hedge attached: Unattributability “ostensibly provides a refutation of access, a key element used in establishing infringement, thereby circumventing the intellectual property protections designed to limit such use as long as the harvest is conducted on a sufficient scale for attribution decay to manifest.”

That word “ostensibly” matters because even the authors are flagging this as a plausible legal argument, not a settled one. If the AI gives you a silkscreen of a brightly colored tomato can, it answers only one narrow question: Did ingesting Warhol’s art cause this image? It says nothing about whether scraping an artist’s work into a training set without permission was legal in the first place. That is the bigger fight already underway in dozens of lawsuits.

Overall, the research sharpens one argument in the AI copyright fight without definitely settling it. Companies now have real evidence against being blamed for specific outputs. But the bigger question, whether using copyrighted work to build these systems was ever allowed, remains exactly where it was before this paper.

They arrived in a kaleidoscope of colors, but one won. For decades, the American school bus has maintained such a standard look that “school bus yellow” is national shorthand for its deep, glossy hue.

No act of Congress mandated the color, with its slight orange tint. Rather, a 1939 conference of transportation officials from all U.S. states (then 48) chose the shade for its visibility as one of the first safety standards. At their peak in the 1980s, the yellow buses took more than 60% of U.S. students to school.

The metal-bodied school bus got its start in rural America. Georgia car dealer Albert Luce designed it to carry cement workers, then marketed it to schools when the company turned it down.

It was impeccable timing as school districts replaced one-room neighborhood schoolhouses with larger ones much farther apart. That consolidation gave rise to the school bus.

“I don’t think you could’ve had one without the other,” says Matt Anderson, head transportation curator at the Henry Ford Museum of American Innovation in Dearborn, Michigan, which houses Luce’s 1927 bus.

Decades later, buses helped desegregate schools, hauling Black and white children across the country’s racial divide. The controversial practice has mostly ended, and in much of the United States, schools have grown more segregated, not less.

Carl Fisher drove a metal school bus for two-thirds of its nearly 100-year history. With 66 years behind the wheel, Fisher retired in 2012 as the longest-working driver, according to the Guinness Book of World Records. His first model was a Dodge truck that his dad’s friend converted, adding benches and small stools to haul 20 children in Pleasant Hope, Missouri.

“He put windows in it and everything, really made it nice,” Fisher says.

The basics became industry standard and more safety upgrades followed, including padded seats and a flip-out stop sign.

“It’s the safest form of on-road transportation per passenger, per mile, bar none — without seat belts,” says Brad Beauchamp, who is with the electric vehicles team at Blue Bird, the yellow school bus maker Luce reputedly named after the bird his wife spotted on a fence post.

Due to driver shortages and other factors, a yellow school bus is no longer available to take many students to school, putting pressure on families to figure out transportation.

Jim Oppegard, who drove kids for more than two decades in the Minneapolis suburb of Brooklyn Park, retired at 94 as the oldest school bus driver on record. It’s rewarding work, he says.

“You’re helping the kids and helping the overall society,” he says. “You’re doing an important job.”

___

Part of a recurring series, “American Objects,” marking the 250th anniversary of the United States. For more American objects, click here. For more stories on the anniversary, click here.

—Jeff McMurray, Associated Press

I used to schlep from my office to a 6:30 p.m. Soul Survivor class, shower, and make dinner by 8 p.m. Every week. The waiting lists were insane. The bike you got mattered. Thirty-four dollars for a cycling class and nobody blinked. That wasn’t about fitness.

SoulCycle was always about selling a version of yourself back to you, one you could embody if you worked hard enough.

The brand said as much itself. Its tagline was never about the workout: “Take your journey. Change your body. Find your soul.” That’s a self-transformation pitch with a bike attached to it. When SoulCycle finally launched its first real ad campaign in 2017, the CEO made the strategy explicit. As Melanie Whelan told Marketing Daily, the “Find It” campaign was meant to help riders “discover something, the thing that makes it meaningful to them”—whether that was strength, purpose, or clarity. The class was just where it happened.

Julie Rice, who cofounded the company, put it more bluntly years later: “People didn’t come to SoulCycle because they got fit. It was the connection they got in the room.” A researcher at Harvard Divinity School who studied the brand found something stranger still—people were bringing questions to their SoulCycle instructors that they used to bring to pastors. SoulCycle was less a fitness brand and more a belief system with a clip-in shoe requirement.

Which is exactly why the loyalty was so volatile. In 2019, SoulCycle’s owner hosted a fundraiser for Donald Trump. Riders who’d built an identity around the brand’s message of inclusion didn’t just complain; they left. Weekly attendance dropped 7.5% within a week. The brand lost nearly 13% of its U.S. customer base the following month. Not because the workout changed. Because the story the brand told about itself stopped matching what its customers believed about themselves.

That’s the part most marketers miss when they talk about brand loyalty. It isn’t really loyalty to the product. It’s loyalty to the self-image the product lets you hold on to. The moment the brand’s actions contradict that self-image, even slightly, the relationship doesn’t bend. It breaks.

Peloton gets blamed for SoulCycle’s decline. So does the pandemic. So does the shift toward Pilates, and the numbers back that one up. Pilates participation grew nearly 40% over the past five years, while cycling fitness dropped 33.5% in the same window, according to TheStreet’s coverage of SoulCycle’s recent outlet closures. That’s a real shift in what people want from a workout.

But it doesn’t explain why a brand with a decade of devotion lost so much of it in a matter of weeks, years before any studio closures made headlines. Markets shift slowly; identity collapses fast. SoulCycle didn’t lose riders because people stopped wanting connection or stopped wanting to feel like the strongest version of themselves. It lost riders because, for a moment, its own behavior made that story impossible to believe. The Pilates numbers and the Peloton competition came later. The real fracture happened the moment the brand stopped being trustworthy to the people who’d built their identity around it.

That’s the risk every identity-driven brand carries, whether they realize it or not. The loyalty is real. So is the exposure.

—By Emily Cody

The opinions expressed here by Inc.com columnists are their own, not those of Inc.com.

Get 1 Smart Business Story delivered straight to your inbox when you subscribe to Inc.’s free daily newsletter.

---

This article originally appeared on Fast Company’s sister website, Inc.com.

Inc. is the voice of the American entrepreneur. We inspire, inform, and document the most fascinating people in business: the risk-takers, the innovators, and the ultra-driven go-getters that represent the most dynamic force in the American economy.

---