Ontology of Cognition

In this episode, we delved into the question of cognitive ontology. While some cognitive functions such as sensory processing or memory are intuitively well-defined, others such as empathy or hierarchical processing remain more cryptic. With the emergence of modern neuroscience, the story only gets more complicated, as brain activity of different areas does not neatly fit with our set of cognitive functions- while some areas are specialised for functions, like the amygdala being involved in emotional processing, most areas are actually multifunctional. How can we integrate our neuroscientific knowledge of how the brain functions into our psychological understanding of our mind?

What do we mean by cognitive ontology?

Cognitive ontology refers to the contents of our minds, the components of psychological theories and vocabulary of functional organisation in the brain. We can also frame our cognitive ontology based on things that can go wrong in mental health disorders. Some examples include working memory, attention, perception, executive function, reward prediction, and cognitive control.

Defining a global shared cognitive ontology may be somewhat inspired by gene ontology, which categorises genes based on their function and the biological pathways they are involved in.

History of cognitive ontology

Plato's theory of the soul posited a tripartite organisation, where the soul is composed of reason (logos), spirit (thumos) and appetitive (epithumia) parts.

Thomas Reid (1710-1796) proposed a “faculty psychology” where the mind is divided into distinct innate mental powers each responsible for certain cognitive functions. These mental faculties included memory, abstraction, judgment, perception, imagination, reasoning, and others.

Franz Gall (1758-1828), a phrenologist, advocated for 27 mental faculties, including memory for words, memory of people, sense of place, sense of colour, love of offspring, affection/friendship, and some stranger ones like talent for poetry, ability to mimic, talent for architecture, sense of connectedness between numbers, sense of sound

Why do we need an ontology of cognition?

A clear shared ontology is necessary to integrate research findings from different studies.

Currently, psychologists frequently use different terms to refer to the same cognitive process. For example, the concept of mindreading is also commonly called mentalising, perspective taking, cognitive empathy, or theory of mind.

Good psychological concepts are supposed to explain current behaviour, predict future behaviour or behaviour in different contexts, and be distinct from one another (Feest, 2020). With the field of neuroscience continuously expanding, and the increasing integration of neuroscientific findings into the field of psychology, there may be a need to revise our cognitive ontology if we hope to predict brain activation patterns on the basis of cognitive function. Psychological findings also have the power to transform neuroscience research. For example, we design tasks based on which cognitive functions we want to better understand and isolate.

Where is there controversy?

Some argue that a large-scale ontological revision is necessary because psychological categories are folksy, intuitive, and unscientific, and the vagueness of the ontology may be holding cognitive neuroscience back (see Poldrack, 2010)

The proposal that scientific progress will require replacing our current set of mental concepts with neuroscientific concepts is not entirely novel- Patricia and Paul Churchland famously argued for eliminative materialism in the 1980s, which holds this view and argues that common-sense concepts of mental states do not correspond to anything real.

Thus, many argue that we should strive to align our cognitive ontology with functional neuroimaging data that tells us which brain areas are active during specific tasks. If we are able to decode the task being done purely based on brain activity, then this is a cognitive ontology we can be certain about. Would you agree?

A key problem here is that functional neuroimaging (e.g. fMRI) has revealed that many processes specified by cognitive ontology do not map onto structural or functional features of the brain.

For example, one study (Lenartowicz et al., 2010)  carried out a meta-analysis of hundreds of fMRI studies of cognitive control constructs (working memory, response selection and inhibition, task switching). They then trained a machine learning classifier to see if it could reliably discriminate patterns of brain activation associated with each construct. Promisingly, the classifier could readily distinguish several cognitive control concepts, but not all of them. It was found that task switching could not be readinly distinguished. Whether this is because of limitations in the study designs (how tasks are designed to isolate the function of task switching), or because task switching should not be considered a cognitive function remains up to debate.

Ultimately, it remains unclear whether we have failed to find one-to-one mappings between brain regions and cognitive functions because of the limitations of the methods we have to read out brain activity, because we have the wrong ontology, or because such mappings do not exist (ontology is represented in some other way).

There's a good review article about this: "Cognitive ontology and the search for neural mechanisms: three foundational problems"

https://www.humanbrainproject.eu/en/follow-hbp/news/individual-brain-charting-a-high-resolution-brain-map-of-cognitive-functions/

Alternative representations of cognitive functions

https://iep.utm.edu/phen-psy/

Perhaps rather than attributing specific cognitive functions to regions, for example the amygdala as a fear centre or cerebellum as motor control, we should define regions based on their mechanism or computational ability. That way many different regions can each have their own contribution to a predefined cognitive function.

For example, Broca’s area is a region known to be specialised in speech production, but is also implicated in other things like tool use, comprehending arm gestures, syntax comprehension, and even musical syntax. Tettamanti & Weniger (2006) claim that Broca’s involvement in these seemingly different processes can be explained by the fact that all of them involve representing hierarchical structures (e.g. syntax, grammatical rules).

Following this framework, it is becoming increasingly common to use data-driven approaches to try to decode functional outcomes based on complex brain activation patterns. For example, Yao et al. (2015) analysed ~10,000 fMRI experiments from the BrainMap database and built a mathematical model (hierarchical Bayesian) linking fMRI activation patterns to 83 task categories via abstract cognitive components. The model estimated the probability that a task would recruit a cognitive component, and the probability that a cognitive component would activate a particular region, which uncovered 12 hypothetical cognitive components that predicted their data set accurately. They varied the number of components from 2 to 20- when there were more components, additional components emerged as subdivisions of lower-order components, corresponding to a nested ontology. How we can define these categories in a way that is easy to understand, rather than abstract concepts, may be a separate question.

Taxonomic pluralism: another option?

An assumption of “taxonomic monism”, that there is one true taxonomy of human cognition, pervades the literature. It still remains in question whether the “right” taxonomy describes the structure of the mind or encapsulates how cognitive elements map onto brain structure. Taxonomic pluralism holds that we will need multiple cognitive ontologies to capture the diverse aims of researchers.

https://easy-peasy.ai/ai-image-generator/images/illustrated-brain-map-cognitive-function-exploration

Check out this chapter for a comprehensive review on this topic: "Neuroscience and cognitive ontology: a case for pluralism"

Finally, one promising approach going forward is to have shared platforms or atlases of our cognitive ontology that researchers can refer to when studying cognitive functions. One such platform currently does exist. It's called the cognitive atlas (cognitiveatlas.org) and is like a Wikipedia for cognitive science.

We hope you enjoyed the episode on this topic. If you haven't yet listened to it, find the episode here: https://spotifycreators-web.app.link/e/QyRKWMD7YSb