AI models today are built on a single repeating block—the Transformer—applied to text, images, and speech alike. But a new paper argues this “structural monoculture” is a mistake. The brain, by contrast, uses different structures for different tasks: dense layers for vision, thick layers for motion. The authors propose a “system of systems” where specialized modules communicate through standard interfaces.
The Research
Jaeho Seol, author of The Giant Hippocampus: From Structural Monoculture to a System of Systems (arXiv:2607.19973), reviews a century of cytoarchitecture—from Brodmann to single-cell Patch-seq—to show that cognitive functions rely on qualitatively different neural structures. He argues the Transformer is functionally analogous to the hippocampus (involved in memory), not to the cortex (involved in general processing). Applying it to tasks like auditory processing or working memory is like using a hippocampus for vision. The paper traces how the “Hardware Lottery” made the Transformer dominant despite this mismatch, and notes that even Mixture-of-Experts models fail to add diversity because they use identical expert modules. The evidence includes convolutional neural networks (CNNs) that achieved strong image recognition with far less data by encoding visual inductive biases—a lesson later discarded.
Why It Matters
This matters for your brain too: the research suggests that specialized, diverse neural structures are more efficient than one-size-fits-all approaches. In cognitive training, this implies that exercises targeting specific abilities (e.g., working memory vs. spatial reasoning) may be more effective than general brain games. Your brain is not a single giant hippocampus—it's a mosaic of specialized regions. Understanding this can help you choose training that matches the task, just as AI should match its architecture to its problem.
What You Can Do
Seek out cognitive challenges that target different skills: puzzles for spatial reasoning, memory games for recall, logic problems for executive function. Avoid “one-size-fits-all” brain training programs. The paper’s lesson is that diversity—in structure and in exercise—is key.
Source: arXiv q-bio.NC
Curious about your own brain? Take our free adaptive IQ test or try 306 brain training levels.
