An Ecosystem for Scalable Symbolic Modeling in Neurosymbolic AI; or Shapes of Cognition

Marjorie McShane, Sergei Nirenburg, Sanjay oruganti, Jesse English

Keywords: theories of cognition, computational cognitive modeling, constructions, reasoning by analogy

Abstract: Symbolic AI has a bad reputation. When used alone, it is associated with small, brittle systems in narrow domains; and when incorporated into neurosymbolic architectures, it tends to serve as a minor flourish to fundamentally empirical systems. But it does not need to be this way. Here we present an ecosystem for developing transparent, scalable, neurosymbolic AI that serves agents, developers, system users, and outside stakeholders alike, while staying true to the scientific grounding of cognitive modeling. This ecosystem underlies Language-Endowed Intelligent Agents (LEIAs) developed within the OntoAgent cognitive architecture and the HARMONIC cognitive-robotic one. This paper has different objectives for different audiences. To readers outside of the symbolic modeling community, it explains why symbolic modeling is useful, feasible, and scalable. To readers within the symbolic modeling community, it proposes specific development methodologies that can help us to collectively make our case to a wide variety of stakeholders, with the goal of expanding the footprint of symbolic modeling in neurosymbolic systems to make the latter transparent, explainable and, ultimately, trustworthy.

Paper Track: Commentary

Submission Number: 18