Cognitive Capabilities

With a little help of my Chap (Wang Qingsong)

Technological breakthroughs happen for reasons which often trigger irrational reactions; it was the case for the printing press and textile machinery, and now for generative AI (GAI). And the forces driving oppositions generally mirror the ones behind innovations, for GAI it’s the gap between human and mechanical achievements as perceived incurrent social and economic environment. An outline of cognitive capabilities of humans and machines could thus help to overcome the objections.

Functional View

From a functional point of view cognition relies on three kinds of resources:

  • Named (identifiable) elements (objects and phenomena) in environments (facts/data)
  • Mental representations of environments, values and objectives (concepts/knowledge)
  • Shared symbolic representations (categories/information)
Cognitive Prism

Cognitive interoperability is achieved through three basic functions:

  • Communication (facts/concepts): Exchange of meaningful signs or symbols between living organisms (plants included) and machines.
  • Classification (facts/categories): Grouping of identified objects and phenomena according to shared features
  • Reasoning (concepts/categories): Truth-preserving processing of mental and/or symbolic representations.

These can be used to characterise individual and collective intelligence across animal species.

Cognitive Tiers

Taking clues from ontogeny (individuals) and phylogeny (species) development, cognitive capabilities can be characterized by three stages:

  • Conversation: Immediate communication between agents with beliefs using signals and signs (animal species and machines) and symbols (human species and machines) with meanings bounded by actual contexts.
  • Reason: Communication (conversational and mediated) and processing of symbolic representations of bounded (machines) or boundless (people) contexts.
  • Judgment: Ability to distinguish between actual and virtual representations and deal with modalities (people).

These intelligent capabilities appear unevenly distributed among animals species, some with a focus on direct non-symbolic intelligence (e.g. mammals), others on collective intelligence (e.g. insects).

Animals’ Cognitive capabilities

The human species is the only one dotted with the whole range of symbolic and non-symbolic cognitive capabilities, often with much inferior ones of the latter. That understanding could provide a yardstick in the current debate about AI limits, setting the cursor for awareness (conversation), transparency (reason), and consciousness (judgment).

Cognition & Ontologies

That taxonomy of cognitive capabilities can be used to design the knowledge bases (aka ontologies) meant to support collaborations between human and artificial brains:

  • Conversations, set on empiric modalities that characterise extensional percepts: observations, assertions, assessments, deductions, managed surrogates
  • Reasonings, set by logical modalities and symbolic descriptions, i.e. the categories used to describe objects and phenomenons: identification, instantiation, agency, containment, life cycle, communication, …
  • Judgments, driven by conceptual modalities and intentional representations: abstract (pertaining to values devoid of direct reference to environments), concrete (pertaining to physical or symbolic environments), virtual (pertaining to hypothetical or fictional environments), and nominal (pertaining to the vocabulary used to label environments), …

Ontologies can thus be understood as an ultimate form of meta-modeling bringing under a common roof the whole range of empiric, conceptual, and logical knowledge.

Cognition & Languages

Language has always been the underlying subtext of Artificial intelligence, whether it’s for implementation (computer languages), communication (user interfaces), or truthfulness (knowledge representation). While technologies have for long shackled AI advances in separate swimlanes, with computing power striding ahead, generative approaches like LLMs now induce a convergence, and as a corollary some confusion between truthful communication and truthful representation. A comparison between pidgins and creoles can help to avoid the confusion.

Pidgins are simplified languages that emerge when a population is constrained to use a foreign language in social communication. To that effect local people use glossaries made of names from natural and foreign languages, combined with shallow lexicons as rudimentary grammars.

From Pidgins (dotted line) to Creoles

Depending on the balance of political and/or socio-economic forces, foreign languages can impose themselves in full, or be confined to domain-specific activities by the emergence of fully developed local languages usually known as creoles. That natural evolution of human languages can provide a reference model for generative ones.



%d bloggers like this: