When AI art has no author: Study finds generated images often can’t be traced to training data

MIT ^ | 08/18/2026 | Rachel Gordon

[BenLurkin]

MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) ... identified a phenomenon they call attribution decay, where the more data a generative model is trained on, the less any individual training example matters to any particular output. It feels counterintuitive, but at sufficiently large scales, they find, you can often remove any single image from the training data, or every image by a given artist, or every photograph of a given person, and the generated sample doesn't change.

And if removing something changes nothing, the researchers argue, it can't be said to be responsible for anything.

...

Their workaround is an architecture they built themselves, called a "diffusion ensemble." Instead of one monolithic model, it's made up of many smaller components, each trained on a different slice of the data. Want to know what the model would do without a particular image? Just switch off the parts that saw it. No retraining, no approximation. What's left is a true counterfactual model, not an estimate of one.

Of course, a clever architecture only matters if it still works as a generator. So the team put the ensembles head to head with 24 conventional diffusion models trained on the exact same data. The images came out looking about as good by standard measures.

One nice surprise in the numbers: The more training data, the better the ensembles held up against their single-model counterparts, a hint that they may actually be more data-efficient.

"When you have low amounts of data, they do very poorly," says Dai. "But if you have more data, it actually scales better compared to the vanilla diffusion model."

---

TOPICS: Computers/Internet
KEYWORDS: theinstitute

[BenLurkin]

Of course, a clever architecture only matters if it still works as a generator. So the team put the ensembles head to head with 24 conventional diffusion models trained on the exact same data.

[spintreebob]

Quality, accuracy of data is more important than Quantity.

AI does not seem to know the difference between fact and fiction. AI seems to regard Cabot Cove ME and Spart MS as equal in consideration to other cities in those states. But, of course, those cities are fictional.

AI places GA cities in the wrong county and the county on the wrong side of the state.

AI has many errors that are correctable. But the humans who define the algorithms for AI consumption of data seem to be much more interested in Quantity than Quality.

GiGO Garbage In Garbage Out.

[ClearCase_guy]

It would end up being a very long “book”, but I can imagine an AI repository that contains short, simple verified factual statements:

Boston is a city in Massachusetts.
Cabot Cove is a fictional city which is stated to be in Maine in the “Murder, She Wrote” series.
etc.

You could employ humans to verify a collection of these short statements and make sure that nothing goes into the repository unless it has been cleared by human.

Make that repository the reference of first resort for any AI tool and it would begin with a factual foundation (better than starting with reddit or wikipedia).

[higgmeister]

When AI art has no author: Study finds generated images often can't be traced to training data

Then Study is an idiot.

[spintreebob]

In traditional IT we have levels of objective/factual REFERENCE DATA. REFERENCE DATA is based on STANDARDS.

Example
Base 10 number system is a STANDARD. Base 16 number system is a STANDARD. Within a STANDARD 12+13 depends on the STANDARD.

Certain US and global GEOGRAPHIC facts are the standard.
Portand ME is not Portland OR. Thus a complex hierarchy exists.

Cabot Cove ME is not real but follows a Literary STANDARD which extends to Books, Movie, TV, Social Media, etc.

Things are TRUE within a CONTEXT. But AI often lacks the meaning of the context. I will specifically ask AI about Dalla GA, GA is clearly the limiting context. But AI will give me answers about Dallas TX outside the context.

Yes, I can badger AI into limiting its response to GA. But that requires me to have knowledge of TX and GA outside AI, which kills the usefulness of AI.

One problem is accepting STANDARDS. Humans have only 2 sex: Male and female; XX and XY chromosones. But try to get those who build AI to accept that.

We humans have many internal contradictions that AI can’t handle.

[minnesota_bound]

I had ChatGPT https://chatgpt.com create these two images. They are art...

It was 105 degrees today so: create hot smiley image
https://davidswebsite.com/freerepublic/hotsmiley.png

Wishing is was colder: create icy smiley image
https://davidswebsite.com/freerepublic/icysmiley.png