Researchers David Gifford and Zheng Dai, who earned a PhD in 2024, have found that artificial intelligence can mimic an artist’s style even when that artist’s work isn’t included in the AI’s training data. This phenomenon, as described by the researchers, suggests a weakening connection between the source data and the resulting AI-generated images.
Growing Datasets Dissolve Links Between AI Art and Artists
Jesus Diaz of Fast Company explains that the study, conducted at MIT’s Schwarzman College of Computing and CSAIL, demonstrates that the connection between training data and generated images weakens as datasets expand. Surgically removing examples from a model revealed this dissolving link, meaning tracing the origin of an AI-generated style becomes increasingly difficult and raises questions about authorship and intellectual property.
Researchers found that larger datasets do not necessarily strengthen the connection between source material and output, but rather obscure it, complicating efforts to identify the influences behind an AI’s aesthetic choices. The work was completed within the Electrical Engineering and Computer Science department and has implications for technology and policy discussions surrounding artificial intelligence and its impact on creative fields.
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.
Jesus Diaz, Fast Company writer
See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
