Neural Network Independently Discovers the 'Memory Palace' Method: Nature Machine Intelligence Study

On July 20, 2026, a team led by Li (Li et al.) published a paper in Nature Machine Intelligence showing that recurrent neural networks trained on a free recall task independently discover several human-like memory strategies. The most accurate of them use an index-based mechanism that resembles the ancient 'memory palace' mnemonic technique.

What the study showed

Recurrent neural networks (RNNs) optimized for free recall found not one but several distinct memorization strategies. According to Nature Machine Intelligence, these strategies go beyond classical temporal context models, which for decades were considered the main explanation for how people recall lists of words in arbitrary order.

Free recall is a classic memory test: a person is shown a list and then asked to reproduce it in any order. For a long time, human behavior in this test was described by temporal context models, in which the probability of recalling a word depends on how close in time it was presented to its neighbors. The new work shows that the task has more than one optimal solution.

• Publication — July 20, 2026, Nature Machine Intelligence, DOI 10.1038/s42256-026-01274-0
• Authors — Li et al.
• Method — recurrent neural networks (RNNs) optimized for free recall
• Key finding — the top models use an index-based mechanism, an analog of the 'memory palace'
• What they outperformed — classical temporal context models

What does this have to do with the 'memory palace'?

The best-performing models relied on an index-based mechanism. They assigned a kind of address to each item to be remembered and looked items up by those labels, instead of scanning through the entire sequence. The approach resembles the 'memory palace' technique (the method of loci), which ancient orators already used: facts are mentally 'placed' in the rooms of an imaginary building, then retrieved by walking a familiar route.

Importantly, the network was not given this strategy ready-made. It arrived at the strategy on its own, simply by optimizing for recall accuracy. In other words, the index scheme turned out to be a good solution to the task itself, not a trick built in beforehand.

Why it matters

The work brings artificial and biological memory mechanisms closer together. If a neural network arrives, unprompted, at a strategy similar to a human mnemonic, that supports the argument that such strategies are an optimal solution to the recall problem, not merely a cultural invention. For neuroscience, it offers a testable hypothesis about how the brain organizes free recall.

For AI, long-term memory remains an open problem: language models hold context in a limited window and organize older memories poorly. A strategy that addresses items by index instead of re-reading them in sequence echoes how modern systems bolt on external databases and vector search. The Li et al. study gives this approach a biologically plausible rationale.

The top models use an index-based mechanism resembling the 'memory palace' technique. — from

Li et al., Nature Machine Intelligence (DOI 10.1038/s42256-026-01274-0)

What this means

Neural networks increasingly serve not only as practical tools but also as models of the brain. By optimizing a network for a specific memory task, researchers get testable hypotheses about how human recall works, along with hints for future AI systems with more structured long-term memory.