From a Neural-Symbolic Means to the Inductive-Instructional End

Loizos Michael

Keywords: neural-symbolic, induction, instruction, coachability, educability

TL;DR: The integration that ultimately drives open-ended AI is epistemic, not architectural: inductive-instructional rather than neural-symbolic.

Abstract: Debates around neural-symbolic AI have productively explored how different representational substrates may be integrated within a single system. This paper argues that such debates risk mistaking a family of architectures for the underlying objective they serve. We propose a conceptual reframing: the central requirement for open-ended intelligence is not the integration of neural and symbolic representations per se, but the integration of inductive learning from experience with instructional learning from communicated information. Neural-symbolic approaches are best understood as a promising class of means toward this deeper epistemic end, which we characterize as inductive-instructional learnability. Clarifying this distinction helps unify disparate research threads and sharpens the criteria by which future AI systems should be evaluated.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 5