Synthetic emotions and consciousness: exploring architectural boundaries
AI & SOCIETY May 1, 2026 Hermann Borotschnig
Artificial agents increasingly display emotion-like behaviors, raising the question of whether such systems risk instantiating consciousness. This paper proposes a hierarchical, dual-source control architecture for synthetic emotion that deliberately excludes architectural features associated with access-like consciousness, as defined by major theories. The architecture combines motivational signals from immediate needs with affective guidance from episodic memory to modulate action selection. The authors specify four engineering risk-reduction constraints: no global broadcast, no metarepresentation, no autobiographical consolidation, and bounded learning. They demonstrate that an emotion-like controller can satisfy these constraints, identify safe extensions, and map gradual transitions that increase access risk. The work provides a methodological template for converting consciousness-related questions into auditable architectural tests and preliminary audit indicators for governance frameworks.