Neuroscientists love the word "representation." They use it the way your uncle uses "literally" - constantly, confidently, and in ways that would make a linguist weep. A neuron "represents" a face. A brain region "represents" fear. A population of cells "represents" the number four. But here's the awkward part: when two neuroscientists say a neuron "represents" something, they might mean completely different things. It's like two mechanics arguing about a car's "performance" when one means horsepower and the other means cup holder accessibility.

A team of researchers led by Stephan Pohl and Wei Ji Ma at NYU, alongside philosopher Ned Block and collaborators across institutions, just published a framework in Nature Reviews Neuroscience that essentially says: "Hey everyone, can we please agree on what we're talking about?" (Pohl et al., 2026).

The "We Need to Talk" Moment

The concept of neural representation sits at the absolute core of neuroscience. Every time a researcher claims that neurons in your visual cortex "encode" the orientation of a line, or that cells in your parietal cortex "represent" how many dots you're looking at, they're making a representation claim. The problem? The field has been operating without a shared playbook for what that actually means or how to test it.

As Konrad Kording put it in earlier work on this exact headache: the term "representation" is probably one of the most common words in all of neuroscience, and "it might mean something very different from one professor to another" (Baker et al., 2022). That's not a minor quibble. It's like building a skyscraper where the architects can't agree on what "load-bearing" means.

Four Dimensions to Rule Them All

Pohl and colleagues propose four key dimensions that pin down what we mean when we say neurons represent something:

Sensitivity - Does the neural response change when the feature changes? If neurons fire differently when you see vertical versus horizontal lines, they're sensitive to orientation. This is the bare minimum - your neurons noticed something happened. Think of it as "did the smoke detector go off?"

Specificity - Does the response change only for that feature, or does it also react to a dozen other things? A neuron that fires for faces but also for cars, shoes, and particularly aggressive squirrels isn't very specific. This is the difference between a smoke detector and one that also screams at toast.

Invariance - Does the representation hold up when irrelevant stuff changes? If your face-detecting neuron still fires for a face whether it's big, small, upside down, or wearing a ridiculous hat, it's invariant to size, orientation, and fashion choices. This is how your brain recognizes your friend whether they're across a room or two inches from your face.

Functionality - Here's the big one. Is the representation actually used by the rest of the brain? A neuron might beautifully track some feature, but if nothing downstream ever reads that signal, it's like writing the perfect email and never hitting send. The brain, it turns out, is full of signals that look meaningful but might just be neural gossip that nobody acts on.

Why Your Analysis Method Matters More Than You Think

One of the paper's sharpest contributions is showing how different data analysis methods map onto these dimensions. Decoding analysis (training a classifier to read brain activity) tests sensitivity - can we tell what the brain saw? Encoding models (predicting brain activity from stimuli) test something subtler. Representational similarity analysis, or RSA, the method Nikolaus Kriegeskorte pioneered back in 2008 to compare how brains and models organize information, captures geometric structure but doesn't directly test functionality (Kriegeskorte et al., 2008).

The team walks through classic neuroscience examples - orientation tuning in visual cortex, numerosity cells, spatial location coding - and shows how different researchers have been talking past each other by emphasizing different dimensions without making that explicit. One group says "we decoded orientation, so it's represented!" Another says "but nothing downstream uses that signal, so is it really?" Both are technically right. They're just answering different questions.

So What Changes Now?

This isn't just philosophy cosplaying as neuroscience. The framework has teeth. It gives researchers a concrete checklist: before claiming something is "represented" in the brain, specify which dimension you're testing. Are you showing sensitivity? Specificity? Invariance? Functionality? All four? Because claiming "the brain represents X" while only testing sensitivity is like saying you've found the murder weapon because you found a knife in the kitchen.

The paper also highlights a gap that should make anyone pause: functionality - the evidence that a representation is actually used downstream - is the hardest to test and the least often demonstrated. We've gotten really good at showing that brain activity correlates with stuff in the world. We're less good at proving the brain actually does anything with that correlation (Favela & Machery, 2025).

For a field that runs on the idea that brains build internal models of the world, getting precise about what "building a model" means isn't just nice housekeeping. It's the difference between actually understanding how the brain works and just describing what it does while sounding very confident about it.

References

1. Pohl, S., Walker, E.Y., Barack, D.L., Lee, J., Denison, R.N., Block, N., Meyniel, F., & Ma, W.J. (2026). Clarifying the conceptual dimensions of representation in neuroscience. Nature Reviews Neuroscience. DOI: 10.1038/s41583-026-01030-8

2. Baker, B., Lansdell, B., & Kording, K.P. (2022). Three aspects of representation in neuroscience. Trends in Cognitive Sciences, 26(11), 942-958. DOI: 10.1016/j.tics.2022.09.003. PMID: 36175303

3. Kriegeskorte, N., Mur, M., & Bandettini, P. (2008). Representational similarity analysis - connecting the branches of systems neuroscience. Frontiers in Systems Neuroscience, 2, 4. DOI: 10.3389/neuro.06.004.2008. PMID: 19104670

4. Favela, L.H. & Machery, E. (2025). The concept of representation in the brain sciences: The current status and ways forward. Mind & Language. DOI: 10.1111/mila.12531

5. Dujmovic, M., Bowers, J.S., Adolfi, F., & Malhotra, G. (2022). The pitfalls of measuring representational similarity using representational similarity analysis. bioRxiv. DOI: 10.1101/2022.04.05.487135

Disclaimer: The image accompanying this article is for illustrative purposes only and does not depict actual experimental results, data, or biological mechanisms.