Papex

Open problem · Neuroscience · Open

Binding problem

How does the brain combine separate features into a single experience?

Summary

How objects, background and abstract or emotional features are combined into one unified experience is not explained; no complete model exists.

Why it is open

The features of a scene are represented in the visual system in a highly distributed way, so something must establish which features belong to the same object [12]. Errors show that this is not trivial: when attention is diverted, people report "illusory conjunctions", for example seeing a shape in the colour of a neighbouring object [2, 15]. Several mechanisms have been proposed, and researchers still disagree about which one the brain uses, and whether a special mechanism is needed at all [12, 13, 15].

Part of the difficulty is that "the binding problem" names several problems. Jerome Feldman distinguishes at least four: the general coordination of activity across brain areas, visual feature binding, variable binding (attaching roles to fillers, as when "Mary" is the one who gives and "John" the one who receives), and the subjective unity of perception [11]. In his review, plausible neural networks for coordination and for visual feature binding are no longer a mystery; the neural realisation of variable binding is "completely unsolved, but is not unsolvable"; and the subjective unity of perception "remains mysterious" [11].

What would count as a resolution differs by version. For visual feature binding, Michael Shadlen and Anthony Movshon set a bar: a theory must say how binding is computed, not only how its result is signalled [7]. A convincing account should also predict when binding fails, and show that disrupting the mechanism produces binding errors while leaving single features intact. Some researchers would count a different outcome as a resolution: showing that no separate binding step is needed at all [15]. For the subjective unity of experience there is no agreed test; that version overlaps with the neural basis of consciousness.

What is known

Attention and binding errors (1980–1995). Anne Treisman and Garry Gelade proposed in 1980 that simple features are registered early, automatically and in parallel, while attention must be directed serially to each item whenever a conjunction of features is needed to identify it [1]. In 1982 Treisman and Hilary Schmidt showed that illusory conjunctions are frequent among unattended items varying in colour and shape, and also occur with size and solidity [2]. In 1995 Stacia Friedman-Hill, Lynn Robertson and Treisman described patient R.M., who had bilateral parietal–occipital lesions: he miscombined colours and shapes even when viewing freely and could not judge where things were; the authors proposed that a damaged spatial representation explains both deficits [5].

Synchrony (1981–2001). In a 1981 report, von der Malsburg proposed that neurons representing features of the same object are grouped by firing in synchrony, within a few milliseconds. The same neurons could then join different groups at different moments, avoiding the combinatorial explosion of needing a dedicated cell for every combination of features [3, 7, 12]. In 1989 Charles Gray, Peter König, Andreas Engel and Wolf Singer reported that neurons in separate columns of cat visual cortex synchronise oscillatory responses in the 40–60 Hz range, and that this depends on the cells' separation, their preferred orientation and global properties of the stimulus [4]. In 2001 Engel and Singer argued that millisecond-precise synchrony, with gamma-range oscillations, may be crucial for conscious processing [8].

Tests and critique (1999–2005). Shadlen and Movshon concluded in 1999 that the synchrony theory describes the signature of binding without saying how binding is computed, that the cortex lacks mechanisms to decode synchronous spikes as a special code, and that little evidence links correlated activity directly to binding; they judged the theory "as proposed" untenable [7]. Direct tests in the motion area MT of monkeys were negative. Alexander Thiele and Gene Stoner found in 2003 that plaid patterns seen as one coherent surface produced significantly less synchrony than patterns seen as two surfaces, the opposite of the prediction [9]. Ben Palanca and Gregory DeAngelis found in 2005 that spike synchrony barely depended on whether features were grouped [10].

Firing-rate labels (1998–2023). In 1998 Pieter Roelfsema, Victor Lamme and Henk Spekreijse trained monkeys to trace one of two curves. Neurons in primary visual cortex (V1) responded more strongly to segments of the traced curve than to the other curve, even where the two crossed [6]. In a 2023 review, Roelfsema separates two routes [12]. "Base groupings" are combinations hard-wired in neurons tuned to feature conjunctions; because there are more possible conjunctions than neurons, a second, flexible route of "incremental grouping" labels the features of one object with enhanced firing and is thought to rely on feedback and horizontal connections. The review concludes that enhanced firing binds features and that oscillations and synchrony are unrelated to binding [12]. It also reviews optogenetic experiments in mice: silencing V1 just after its initial feedforward response left grating detection intact but prevented the mice from separating figures from their background [12].

The question is still contested (2024–2025). With intracranial recordings in people, Jacob Garrett and colleagues reported in 2024 that brief oscillations near 90 Hz ("co-ripples") occur together across language areas during reading and semantic decisions, phase-locked at zero lag over more than 12 cm; they call the question of whether such oscillations help integrate information controversial [13]. In a 2025 opinion article, Steven Scholte and Edward de Haan argue that neurons and brain regions respond to many features at once, that lesion studies have not found purely selective deficits, and that the binding problem may be an artefact of a modular view of vision [15]. In artificial intelligence, Declan Campbell and colleagues reported in 2024 that vision–language models fail at simple multi-object tasks such as counting and localisation, and argued that many of these failures can be explained by the binding problem and resemble the limits of rapid, feedforward processing in humans [14].

Main approaches

Binding by attention (feature integration). Features are registered in parallel; focused attention to a location combines them, one object at a time [1, 12]. Strongest evidence: slow search for conjunctions, illusory conjunctions without attention, and binding failures after parietal damage [1, 2, 5]. Main difficulty: it is a cognitive account; it says when binding needs attention, not which neural signal marks the result. Much of its evidence comes from brief displays of arbitrary feature combinations; Scholte and de Haan argue that such displays create errors that natural vision rarely meets [15].

Binding by synchrony (temporal correlation). Neurons coding features of one object fire in synchrony, separately from those coding other objects [3]. Strongest evidence: stimulus-dependent synchrony across cortical columns [4] and its proposed link to awareness [8]; human co-ripples keep the broader question open [13]. Main difficulty: it explains how binding might be signalled, not how it is computed [7], and targeted tests in MT did not find the predicted synchrony [9, 10].

Binding by enhanced firing (incremental grouping). Neurons coding the attended object's features fire more, and the enhancement spreads to elements that group with them [6, 12]. Strongest evidence: object-based response enhancement in V1 [6] and causal silencing results for figure–ground perception [12]. Main difficulty: by its own account, incremental grouping takes time and handles one object at a time [12], so it must explain how a scene with several objects is experienced as unified. Its evidence is from vision; variable binding in language remains, in Feldman's words, "completely unsolved" [11].

Conjunction coding, or no special problem. Neurons tuned to feature combinations ("base groupings") already code some groupings [12]. Scholte and de Haan go further: because features co-occur in predictable ways in natural scenes, the visual system represents those co-occurring patterns directly and needs no separate binding step [15]. Strongest evidence: multi-feature responses and the lack of purely selective deficits after lesions [15]. Main difficulty: the number of possible combinations exceeds the number of neurons [7, 12], so novel combinations still need a flexible mechanism; and current vision–language models still fail on simple multi-object tasks [14].

Open sub-questions

  1. How many groups at once? Synchrony allows several objects to be labelled at once; the enhanced-firing account predicts one incremental group at a time [12]. Can recordings during a two-object task show both objects labelled at the same moment in the same area?
  2. Causal test for feature binding. Does disrupting the late, feedback-dependent response of visual cortex (optogenetically in animals, or with brain stimulation in people) increase illusory conjunctions, not only figure–ground errors, while sparing discrimination of single features [2, 12]?
  3. Co-ripples. Are zero-lag co-ripples between distant human cortical areas [13] needed to integrate word form and meaning, or a by-product of co-activation? What result would tell these apart?
  4. Natural statistics. Do illusory conjunctions become rarer when feature combinations follow natural co-occurrence, as Scholte and de Haan's account implies [15], and do they return for novel combinations?
  5. Machine binding errors. Do vision–language models make human-like illusory conjunctions under load, and do architectures with explicit object slots or serial attention remove those errors [14]?
  6. Variable binding. What neural activity distinguishes "the dog chased the cat" from "the cat chased the dog", where the words are the same and only the roles differ [11]?

Work on this in Papex

If you have a testable answer to one of the sub-questions above, you can draft it as a private hypothesis in Papex, or prepare the draft from Claude or ChatGPT with the Papex connector. You review the exact version, sign it and decide whether to publish it; a public record that names this problem is listed on this page. A Papex record shows who signed which version and when. It is not peer review and does not mean the hypothesis is correct.

Key references

  1. Treisman, A. M. & Gelade, G. (1980). A feature-integration theory of attention. Cognitive Psychology 12(1), 97–136. 10.1016/0010-0285(80)90005-5 · Checked 06 Oct 2026
  2. Treisman, A. & Schmidt, H. (1982). Illusory conjunctions in the perception of objects. Cognitive Psychology 14(1), 107–141. 10.1016/0010-0285(82)90006-8 · Checked 06 Oct 2026
  3. von der Malsburg, C. (1994). The correlation theory of brain function. In Models of Neural Networks (Physics of Neural Networks), Springer, 95–119. First circulated as Internal Report 81-2, Max-Planck-Institut für Biophysikalische Chemie, Göttingen, 1981. 10.1007/978-1-4612-4320-5_2 · Checked 06 Oct 2026
  4. Gray, C. M., König, P., Engel, A. K. & Singer, W. (1989). Oscillatory responses in cat visual cortex exhibit inter-columnar synchronization which reflects global stimulus properties. Nature 338(6213), 334–337. 10.1038/338334a0 · Checked 06 Oct 2026
  5. Friedman-Hill, S. R., Robertson, L. C. & Treisman, A. (1995). Parietal contributions to visual feature binding: evidence from a patient with bilateral lesions. Science 269(5225), 853–855. 10.1126/science.7638604 · Checked 06 Oct 2026
  6. Roelfsema, P. R., Lamme, V. A. F. & Spekreijse, H. (1998). Object-based attention in the primary visual cortex of the macaque monkey. Nature 395(6700), 376–381. 10.1038/26475 · Checked 06 Oct 2026
  7. Shadlen, M. N. & Movshon, J. A. (1999). Synchrony unbound: a critical evaluation of the temporal binding hypothesis. Neuron 24(1), 67–77. 10.1016/S0896-6273(00)80822-3 · Checked 06 Oct 2026
  8. Engel, A. K. & Singer, W. (2001). Temporal binding and the neural correlates of sensory awareness. Trends in Cognitive Sciences 5(1), 16–25. 10.1016/S1364-6613(00)01568-0 · Checked 06 Oct 2026
  9. Thiele, A. & Stoner, G. (2003). Neuronal synchrony does not correlate with motion coherence in cortical area MT. Nature 421(6921), 366–370. 10.1038/nature01285 · Checked 06 Oct 2026
  10. Palanca, B. J. A. & DeAngelis, G. C. (2005). Does neuronal synchrony underlie visual feature grouping? Neuron 46(2), 333–346. 10.1016/j.neuron.2005.03.002 · Checked 06 Oct 2026
  11. Feldman, J. (2013). The neural binding problem(s). Cognitive Neurodynamics 7(1), 1–11 (published online 2012). 10.1007/s11571-012-9219-8 · Checked 06 Oct 2026
  12. Roelfsema, P. R. (2023). Solving the binding problem: assemblies form when neurons enhance their firing rate—they don't need to oscillate or synchronize. Neuron 111(7), 1003–1019. 10.1016/j.neuron.2023.03.016 · Checked 06 Oct 2026
  13. Garrett, J. C., Verzhbinsky, I. A., Kaestner, E., Carlson, C., Doyle, W. K., Devinsky, O. et al. (2024). Binding of cortical functional modules by synchronous high-frequency oscillations. Nature Human Behaviour 8(10), 1988–2002. 10.1038/s41562-024-01952-2 · Checked 06 Oct 2026
  14. Campbell, D., Rane, S., Giallanza, T., De Sabbata, N., Ghods, K., Joshi, A., Ku, A., Frankland, S. M., Griffiths, T. L., Cohen, J. D. & Webb, T. (2024). Understanding the limits of vision language models through the lens of the binding problem. Advances in Neural Information Processing Systems 37, 113436–113460. 10.52202/079017-3604 · Checked 06 Oct 2026
  15. Scholte, H. S. & de Haan, E. H. F. (2025). Beyond binding: from modular to natural vision. Trends in Cognitive Sciences 29(6), 505–515 (Opinion). 10.1016/j.tics.2025.03.002 · Checked 06 Oct 2026

Page reviewed 06 Oct 2026

Source

Open, as listed by Wikipedia · Unsolved problems in neuroscience. Checked 29 Sep 2026.

The status is the source’s. Papex does not decide whether a problem is solved.

Records on this problem

No public record addresses this problem yet.

More in Life sciences