The binding problem goes away not because we solved it, but because we never needed it to begin with.
By James Vornov, MD PhD
Neurologist, drug developer and philosopher exploring the neuroscience of decision-making and personal identity.
A little change of pace this week. My views on brain maps were changed recently by an important new reframing of the binding problem in a review by H. Steven Scholte and Edward H.F. de Haan in Trends in Cognitive Sciences (2025). Their paper Beyond binding: from modular to natural vision has helped me understand how it’s possible that the many maps we find across the cerebral cortex could ever provide a unified model of the world without them ever coming together in a theater of the mind.
What is the binding problem?
When we look at a scene and see a red car and a blue bicycle, how does the brain associate the right color with the right object? I was taught that the visual system is a pipeline that extracts features. During my career, the process has been mapped in great detail using recordings from awake behaving animals and using non-invasive measures in people like fMRI. We have a very good idea of how we detect color and identify objects in the cerebral cortex.
Scholte and de Haan talk exclusively about the visual system, so let’s stick to that, realizing that this binding issue applies more broadly when we consider the coordination of both neighboring and distant cortical areas in presenting the world in awareness. Now we know that after preprocessing of contrast and edge in the retina and thalamus, the primary visual cortex is essential for detecting edges and separating the binocular depth information. From there, visual information is further processed by nearby areas, each with its own mapping of the visual field and its unique response pattern. V2, V3, V4, MT.
How are features bound together in perception?
And so we see the problem. If form is in V3 and color in V4 and the motion of the bicycle relative to the car is in MT, how do you associate the extracted features together as a unified perception? Red car and blue bicycle even though red and blue are extracted by one module and car and bicycle by another? This is what has been called the binding problem.
This has intrigued me for many years. I’m not so bothered by emergent qualities like free will and subjective experience. I feel comfortable exploring the underlying mechanism that supports these emergent experiences. I don’t think one can easily explain how the neural activity gives rise to the emergent phenomenon. As Weinberg said, “The arrows of explanation point downward”.
But the binding problem is one of neural activity. How can a modular system give rise to unitary experience? Gamma synchrony was a popular explanation. The idea is that neurons representing features bound to the the same object would fire in sync at gamma frequencies (30-70 Hz) while staying desynchronized from neurons representing other objects. Wolf Singer and others pushed this hard in the 90s. But it turned out gamma synchrony is too weak to bind neurons across different brain regions, or even between neurons more than a few millimeters apart. Actually, neuronal activity can be perceived as a single event in awareness whether signals arrive synchronously or spread across 100 ms. Synchronization is not the answer.
There were other proposed mechanisms like enhanced firing rates for bound activities or convergence zones where features meet. But they all shared the same assumption: features are processed separately and need to be glued back together somehow.????????????????
But what if we were mistaken from the start? What if the brain isn’t really modular in nature? Scholte and de Haan argue that this was a misguided framework. It seems to have arisen mostly arose from how we looked for cortical specialization in our experiments. Going back to Hubel and Wiedel, we went on a hunt for specialization by presenting simple stimuli and finding the area that reacts most consistently, ignoring all the weaker responses seen more broadly.
Mistaken from the start? An artifact of the search?
In fact, as Scholte and de Haan review the evidence, it becomes quite clear that recent experimental and modeling data present a strong case that the brain is not strictly modular at all. In each case they cite, the impression that there is modular feature extraction arose from how our experimental paradigm reinforced the conceptual framework of modularity and feature extraction.
- Modern electrophysiology – Large-scale recordings show that neurons aren’t finely tuned to particular features. In reality, when you do unbiased sampling across broader regions with complex stimuli, it’s clear that neurons respond to multiple features simultaneously. The idea that they were tuned to isolated properties came from presenting stimuli with isolated properties, finding the strongest responding area, and declaring that area the feature extractor. All of the more broad, weaker responses were literally ignored because they didn’t fit with the conceptual framework.
- Advanced neuroimaging – Similarly, unbiased analysis of activity from fMRI shows that areas of the brain engage in processing multiple features at once. Again, we force the data into our frameworks by doing subtraction to highlight the most unique responses to assign function to brain areas. Clearly, this is important to understand how information is processed, but it doesn’t prove strict specialization.
- Reexamined neuropsychological cases – As a neurologist with an interest in stroke, the reevaluation of classic cases of “pure” deficit patients was fascinating. In the cases discussed, the patients actually had co-occurring impairments that were overlooked. I was always surprised at how rare these kinds of deficits were in the patients I saw in the hospital. They were memorable, but I always wondered why other patients with the same distribution of injury did not show the “classic” deficit, which was probably not really so classic. But again, these cases were presented for didactic discussion of modular brain function. We taught the cortex was a mosaic of functions, but we never were really able to explain why some regions are “eloquent”, meaning that when injured, they always result in deficit. Like V1 in the visual system, the primary motor cortex.
- Large-scale lesion study – The motivation for this review seems to come from their study of over 300 patients trying to supposedly specialize brain regions to specific deficits. Just as my clinical experience showed, some areas are eloquent and show up reliably, but for the most part, using validated tasks of the kind we use for advanced imaging in humans or electrophysiology in animals, you can’t detect a deficit when the supposedly specialized area is injured.
- Deep neural networks – By way of explanation, the review presents the evidence that Deep Neural Networks that are specifically designed to carry out the tasks of the nervous system can achieve excellent functionality like robust vision without feature segregation or binding mechanisms??????????????. The features are obviously represented in the model, but not segregated in any way you can see examining or perturbing the model. Actually, this criticism cuts much deeper than that. If you train Deep Neural Networks (DNNs) to match actual firing patterns in the brain areas, you get the same broad responsivity and you don’t need binding mechanisms. So if you constrain the network to do what neurons actually do, you see that there’s not any need for binding.
So the idea of binding of features not only is contradicted by the evidence, but there’s no need for it to begin with. I’d say that’s just about the perfect definition of a useless framework.
So why specialize at all?
Yet, even without binding, we’re left with the question of why specialize at all? If lesions of specialized areas like MT don’t destroy perception of coherent motion, why is it there at all? What’s it doing? Or broad areas of parietal cortex specialized for cross-modal integration of, say, vision and joint position sense, what are they doing if lesions have no effect on what we can measure by sensitive testing?
As a clinician, let me guess at a somewhat metaphorical answer. I think it’s like a research group where each member has some special expertise that they bring to the joint activity to make discoveries and publish papers. Let’s say the molecular biologist gets a faculty offer elsewhere and leaves. What does the group do? Do they stop their research and publication activities? No, they use what knowledge they have of molecular biology and carry on. Maybe their productivity goes down for a while as they study up on techniques and one or more of them gets better at molecular biology. Not their original training and passion, but there’s enough capability there that maybe at the publication level you barely notice a decrease in output.
Would we say that the molecular biologist wasn’t necessary? Wasn’t a brilliant specialist and a major contributor to the group? Of course not. Now if the lab head who wrote all the grants and brought in the money got sick, then productivity might take a hit for a while until the group got its act together. If it could and didn’t just shut down. The lab head is an eloquent part of the group. The molecular biologist is not.
And so this is how I conceptualize the loss of a specialized area due to a stroke. There’s redundancy and the network continues to function, perhaps a bit degraded. The specialization was providing a boost to function, but the function was never fully dependent on it no matter how brilliantly it worked. In fact, that very brilliance diffuses out, providing resilience. And so we see disruption and recovery after a stroke when these “silent” areas are damaged. The system loses its specialized boost from the area, but it can function. We see this kind of specialization and resilience in all kinds of ecologies, and I think of these complex brain meshes as ecologies, with each area contributing to a whole, being a part of the broader representation of the self in the environment. Can’t function alone, but for the most part, the rest can function when lost.
So bye bye binding. The brain was never taking apart and reassembling the world. It was in a set of parcellated, redundant maps, with specialized areas acting as boosters rather than keepers of secrets. The binding problem dissolves not because we solved it, but because we never needed to begin with.
Prefer to follow via Substack? You can read this and future posts (and leave comments) at On Deciding… Better on Substack: Brain, Self, and Mind
© 2025 James Vornov MD, PhD. This content is freely shareable with attribution. Please link to this page if quoting.
