Rethinking Grounding

The grounding problem is, generally speaking, theproblem of how to embed an artificial agent into itsenvironment such that its behaviour, as well as themechanisms, representations, etc. underlying it, canbe intrinsic and meaningful to the agent itself, ratherthan dependent on an external designer or observer.This paper briefly reviews Searle’s and Harnad’sanalyses of the grounding problem, and then evalu-ates cognitivist and enactive approaches to over-coming it. It is argued that, although these twocategories of approaches differ in their nature andthe problems they have to face, both, so far, fallshort of solving the grounding problem for similarreasons. Further it is concluded that the reason theproblem is still somewhat underestimated lies in thefact that modern situated and embodied AI, despiteits emphasis of agent-environment interaction, stillfails to fully acknowledge the historically rooted in-tegrated nature of living organisms and their envi-ronmental embedding.