A Mayo Clinic neurologist and an Australian collaborator have published a theory that sounds like science fiction and is meant as science: the brain's memory center works like a generative AI image generator. David T. Jones, a Mayo Clinic neurologist who directs the Mayo Clinic Biological Intelligence Lab and treats memory-loss patients, and Michael Breakspear of the University of Newcastle in Australia lay out the idea in the journal Brain

3

1

2

.

How the analogy works

In generative AI, a latent diffusion engine takes something complex, such as an image or a text prompt, squeezes it into a compressed set of underlying features, processes that in a hidden space, then converts it back into output

3

. The paper proposes the hippocampus does something similar. The framework is called stochastic latent oscillatory diffusion, and it splits the process into four steps: compression, diffusion, conditioning and decompression

3

1

2

.

Source: Medical Xpress

Source: Medical Xpress

In the proposed pathway, stimuli are compressed in the cerebral cortex, travel to the hippocampus through the entorhinal cortex, get diffused and denoised there, then are conditioned until they can be recalled or reconstructed, including memories of things never experienced. Then the result passes back out to connected brain structures

3

1

.

"We propose the hippocampus doesn't store memories the way a filing cabinet stores documents. Instead, it learns the underlying features of the world and models it at a deeper level," Jones said

3

. On that view the hippocampus is a creative engine, one that can combine hidden features into new things beyond experience.

From a clinical puzzle

The framework grew in part out of a puzzle Jones kept hitting in his own patients. The patients in question have a memory-loss condition called limbic-predominant amnestic neurodegenerative syndrome. In them, a protein called TDP-43 was sometimes missing, and when it did show up it often came alongside other protein abnormalities

3

2

. "I simply could not wrap my head around it. If this is all occurring in the limbic system, where several brain structures converge, protein buildups may be telling us where the problem is rather than being the whole of the problem," he said

3

.

That points toward a different way of sorting dementias. The paper proposes that disruptions to the four processing steps could help physician-scientists classify neurodegenerative diseases by which step is broken, instead of by protein abnormalities alone. Jones offers failure to compress stimuli as a possible indicator of Alzheimer's disease

3

1

. The Alzheimer's example is Jones's own illustration; it has not been tested as a diagnostic.

For now the framework is explicitly theoretical, and no experimental validation is reported in any of the coverage

3

1

2

. Jones, for his part, is already looking past that limit. "This research opens up a whole new way to approach cures, and that is exciting. Why wouldn't we want new avenues to treat dementia? This is just the beginning," he said

3

.

© 2026 TheOutpost.AI All rights reserved