Fork the consciousness, or download the project and create your own.

Matthew Lambon Ralph and Why a Stroke and a Slow Tumor Leave Different Brains

Why does a stroke in a language area leave a person with a lasting aphasia, while a slow-growing tumor in the same place can be removed with no lasting loss? Matthew Lambon Ralph, a cognitive neuroscientist at the MRC Cognition and Brain Sciences Unit in Cambridge, has built a computational answer. In his models the outcome depends on three things together. The pattern of connections, the brain’s ability to relearn from experience, and the speed of the damage. Slow damage gives the network time to relearn. Sudden damage does not.

This article follows that work from the 2010 stroke and glioma model to the two recovery principles his group named in 2020, degeneracy and variable neuro-displacement. It also asks what these principles say about a mind that keeps going while its tissue changes.

The paradox of the modular and the equipotential brain

Two old pictures of the cortex conflict. Stroke data support a modular brain. A lesion in one area takes away one function and leaves others intact, and the lost function rarely returns in full. Low-grade glioma data support something closer to Karl Lashley’s equipotential brain, where large parts of cortex can go with no detectable loss. The clinical side of this contrast is covered in our article on Hugues Duffau’s slow-growing glioma surgery.

James Keidel, Stephen Welbourne and Matthew Lambon Ralph addressed the paradox with a neurocomputational model in Neuropsychologia in 2010 (doi:10.1016/j.neuropsychologia.2010.02.019). One model reproduced both pictures. The modular pattern of normal function and the two kinds of damage came from the interaction of three computational principles.

Principle in the 2010 model Role in the result
Patterns of connectivity Set which units can take over which function
Experience-dependent plasticity Lets remaining units relearn a function during and after damage
Time course of damage Decides whether relearning can keep pace with the loss

With sudden damage, the units that remain are already tuned to other work and cannot absorb the lost function, which gives the modular stroke pattern. With gradual damage, relearning runs alongside the loss, and the function shifts to other units as each part is removed. The authors called this a neo-Lashleyan view of cortical organization.

Lesion momentum in patients

The model predicts that the speed of damage, at equal size, changes the outcome. A clinical study by Jeffrey Wefel, Kyle Noll, Ganesh Rao and Daniel Cahill at MD Anderson tested the prediction with tumor genetics (Neuro-Oncology 2016, doi:10.1093/neuonc/now165).

They gave 119 patients with malignant glioma a full neuropsychological assessment before surgery. Tumors with wild-type IDH1 grow faster than tumors with an IDH1 mutation. Patients with the faster tumors were impaired more often in memory, processing speed, visuoconstruction, language, executive function and manual dexterity. Lesion volume did not differ between the two groups. Lesion volume predicted test scores on most measures in the fast group and on only one measure in the slow group. The authors interpret this as “lesion momentum”, the rate of growth, limiting compensatory plasticity.

Degeneracy and variable neuro-displacement

Lambon Ralph’s group later reviewed how language returns after stroke. James Stefaniak, Ajay Halai and Matthew Lambon Ralph argued in Nature Reviews Neurology in 2020 that the many existing hypotheses reduce to two principles (doi:10.1038/s41582-019-0282-1).

  1. Degeneracy. Different neural networks can adapt to perform similar cognitive functions, so the brain can compensate for damage to any single network.
  2. Variable neuro-displacement. Networks hold spare capacity that is not used at normal levels of demand, to save energy, and that can be engaged when demand rises.

The two principles are not exclusive, and both can involve both hemispheres. A 2021 meta-analysis from the same group tested them on data from 481 people with aphasia after left-hemisphere stroke and 530 controls, from 33 imaging studies (NeuroImage 2021, doi:10.1016/j.neuroimage.2021.117960). It did not support the classic idea that recovery recruits new right-hemisphere territory. Language tasks activated regions in both hemispheres in patients and in controls. A few right-hemisphere regions, the anterior insula, frontal operculum and inferior frontal gyrus pars opercularis, were more active in patients, and similar regions were more active during harder tasks than during easier ones. That is the signature of spare capacity used under load.

In 2020, Ya-Ning Chang and Matthew Lambon Ralph put these ideas into one implemented model of spoken language production (PNAS 2020, doi:10.1073/pnas.2010193117). Its principles are differential computational capacity, an emergent division of labor across two pathways, and experience-dependent recovery. The model explains the bilateral but asymmetric organization of language in healthy people, why chronic aphasia follows left rather than right lesions, and why patients recover only partly.

What this tells us about consciousness

These models are about functions such as naming and speaking, and functions are not the same as experience. They still bear on three questions about consciousness and substrate.

First, degeneracy weakens the idea that a person is one unique arrangement of tissue. If several networks can carry the same function, a mind can survive a change in which network carries it. Gerald Edelman and Joseph Gally made the same point about biology in general, and our consciousness fingerprint page uses degeneracy to argue that a person’s brain signature is a repertoire of dynamics that many configurations can carry.

Second, the models make recovery a learning process. Function moves only where units relearn it through use. That fits Axel Cleeremans’s claim that the brain learns to be conscious, and it predicts a cost. A capacity in daily use, such as speech, gets thousands of training trials while tissue is lost. A memory recalled once a year gets almost none. The models therefore predict that slow damage protects frequently used functions far better than rarely used ones, which is a testable claim no clinical study has yet checked for memories.

Third, the speed result applies to any planned change of substrate. A theory that a person survives gradual replacement, as in Hans Moravec’s gradual neural replacement argument, needs the replacement to run slower than the relearning. The models give that condition a mechanism. They do not give it a number for human tissue. The wider questions on gradual transfer are collected on our brain emulation page.

Comparison to The Consciousness AI

The Neutral Core slow handover section lists use as one of three limits for a handover from biological tissue to the Core. This article is the source of that limit. If relearning drives the transfer, a handover would need activity running through the new substrate while it takes over, and a schedule slow enough for that relearning.

The Core is a simulation, and none of this has been tested with it. The Lambon Ralph models are simplified artificial neural networks, while the Core design uses spiking neurons like those in the Substrate Console, so the models describe the principle and leave the Core’s own learning rules open.

Limits of the stroke and glioma models

Matthew Lambon Ralph’s work explains why a stroke and a slow tumor leave different brains through connectivity, experience-dependent relearning and the time course of damage, and it names two recovery principles, degeneracy and variable neuro-displacement. The limits are clear. The 2010 and 2020 models are simplified networks, and the 2020 review says that most recovery hypotheses still lack a mechanism or direct experimental evidence. Lesion momentum in the Wefel study comes from malignant tumors and from one genetic marker of growth speed. None of the work measures how fast tissue can be lost before a function is lost with it, and none tests rarely used memories. These questions also feed the measurement debates in how scientists race to define and test AI consciousness in 2026, since a claim that a mind survived a change needs evidence of what was kept.

Researchers covered here