The Embodied Intelligent Elephant in the Room

The central point made in this paper is this: human-level grounded meaning in an agent can only result from directly experiencing the world, which in turn can only be possible via embodiment (coupled with ‘embrainment’ - a suitable brain architecture). Via embodiment, we humans are able to internalize our direct interactions the world, in addition to being able to associate symbols with them—this allows us to communicate via symbols, thereby externalizing our individual representations for mutual, collective benefit. Lacking embodiment, in contrast, today's AI agents are able to only operate at a derivative, symbolic level, without being able to experientially understand, or relate to, the meaning behind the symbols they process. The only way to enable artificial agents to acquire ‘first-hand’ meaning for symbols, would be to provide suitable embodiment for them. Also, a case is made for machines capable of exhibiting non-symbolic intelligence via analog embodiment, to complement symbol-originated intelligence.