As we should expect, fMRI shows many local regions of the brain involved in specific cognitive activities, such as recognizing a face or understanding language. Less expected is the complementary discovery of a constant pattern of brain activity associated with many different kinds of cognitive demand. This same “multiple-demand” or MD pattern is seen in tasks tapping stimulus discrimination, working and episodic memory, response inhibition, language, problem-solving, arithmetic and much more (see here). The MD network, we propose, constructs attentional control structures for complex behaviour of all kinds.

With fMRI, the MD network can be precisely defined in highly specific regions of frontal, parietal and occipito-temporal cortex. Our most recent definition uses the high-resolution cortical parcellation of the Human Connectome Project, with a widely-distributed core of 10 MD regions surrounded by a more weakly activated and connected penumbra.  Often, MD regions are immediately adjacent to regions with quite different functional properties, e.g. specific involvement in language. With identification of specific MD regions, we aim for detailed investigation of their functional properties and contributions to attentional control.

References:
Duncan, 2013
Fedorenko et al., 2013
Assem et al., 2020
Assem et al.,  2022
Chou et al., 2023

Once MD regions are defined, their physiological properties can be examined using a range of complementary methods. With fMRI, for example, detailed representations of task events can be studied using multivoxel pattern analysis or MVPA. Our results show broad coding of many different kinds of task information, including relevant stimuli, responses and rules. In our group at the University of Oxford, neurophysiological recordings in behaving monkeys have allowed MD activity to be examined at the level of activity in single cells and cell populations. In lateral frontal cortex, for example, neurons show high speed adaptation to code the specific information and events of current behaviour; explaining why, in fMRI, activity in the same regions is seen for many different kinds of task. This work shows how distinct regions of frontal cortex collaborate to control successive task steps, and how information shifts between attentional foreground and background.  Collaborations with clinical groups in several institutions worldwide are allowing us to conduct similar studies in human patients with implanted brain electrodes.

Outside cerebral cortex, we are also beginning to examine MD activity in specific regions of caudate, thalamus and cerebellum.  With strong connectivity of these cortical and subcortical regions, the MD system is well placed for cognitive integration.  We propose that, with “attention” to a cognitive operation, the MD system selects and assembles its component parts, assigning them to their required roles and relationships.

References:
Stokes et al., 2013
Woolgar et al., 2016
Assem et al., 2023
Kadohisa et al., 2023
Watanabe et al., 2023

Well-known tests of “fluid intelligence” – usually involving novel problem-solving – are of interest for their broad ability to predict success in many kinds of laboratory and real-world tasks. In fMRI, fluid intelligence tests show strong activity in the MD system, and in neurological cases, we find that impaired fluid intelligence is predicted by MD lesions. These tests, we propose, measure a fundamental aspect of MD activity, solving complex problems through attention to simpler, more tractable episodes.

To develop this hypothesis, we have used a bizarre form of error termed goal neglect. In goal neglect, a person knows and can describe task rules, yet these rules have no apparent influence on behaviour. Extreme cases are well known from the literature on frontal lobe damage, but when tasks are novel and complex, neglect is seen also in the normal population. Importantly, it is linked closely to fluid intelligence, and in behavioural experiments we have linked neglect to attentional control.

Fluid intelligence declines strongly in old age. Using MRI, we are asking how cognitive decline links to changed structure and function within and outside the MD system (camCAM, Henson).

References:
Duncan et al., 2008
Duncan et al., 2017 
Woolgar et al., 2018
Mitchell et al., 2022

Many clinical tests are used to measure executive control deficits in frontal lobe patients. Because fluid intelligence is correlated with all kinds of test, and because it is impaired after MD lesions, we are asking how far different clinical deficits are explained by fluid intelligence loss.

The results show an intriguing picture. For some common tests, such as Wisconsin Card Sorting and Trail Making, deficits in frontal lobe patients are entirely explained by fluid intelligence. Since much of the frontal lobe lies outside the MD system, however, we should not expect this for all tests. Confirming this prediction, some aspects of executive control (e.g. social cognition, complex multi-tasking) show deficits separate from fluid intelligence. In some cases, the evidence suggests link to damage outside the MD system.

In a wide range of neurological and psychiatric conditions, impairments in attentional control cripple return to everyday life. Their assessment, however, is notoriously uncertain. Working with Tom Manly, we consider the use of cognitive testing in patient assessment and management.

References:
Roca at el 2010

The brain’s “default mode network” or DMN is often considered complementary to the MD network, with reduced activity in many tasks compared to rest. A common interpretation is that DMN activity reflects self- or internally-directed cognition.   Our research, however, links DMN activity to  major changes in the focus of cognition, whether externally- or internally-directed.  We are investigating how MD and DMN systems combine in cognitive control.  Our core hypothesis is that the DMN establishes a broad cognitive ontext, including spatial, temporal and social aspects, with the MD system controlling individual cognitive operations within this broad context.

References:
Crittenden et al., 2015
Smith et al., 2018
Wen et al., 2020

Our ideas concerning attentional episodes are reminiscent of the concept of parental scaffolding in cognitive development.  In scaffolding, the parent shapes child cognition by focusing attention on useful parts within a complex activity.  With collaborators Lynne Murray and Peter Cooper from the University of Reading, along with Claire Hughes from the University of Cambridge, we are testing the prediction that parental skill in scaffolding predicts child fluid intelligence.

References
Duncan et al., in press

John Duncan: Programme leader
Danny Mitchell: Senior investigator scientist
Valentina Mione: Investigator scientist
Moataz Assem: Postdoctoral fellow
Jascha Achterberg: Research student
Ashley Zhou: Research student