This paper highlights autonomous systems as the ultimate stage of AI development and argues that achieving this requires a hybrid of connectionist and symbolic AI. We advocate embedding AI within a systems‑engineering framework rather than treating it as a standalone component.
A generic agent architecture is introduced, where behavior is expressed as a composition of cognitive functions organized around a long‑term memory that stores the agent’s evolving knowledge. The perception module maps raw sensor inputs to structured memory entries, the decision module generates plans to achieve goals based on these entries, and coordination mechanisms enable the fusion of individual and collective intelligence in multi‑agent settings.
Implementation challenges focus on three fronts: (1) bridging perception and memory by converting high‑dimensional sensory data into searchable symbolic representations; (2) goal‑directed decision‑making and planning under uncertainty, balancing exploration and exploitation; (3) multi‑agent collaboration, which demands communication protocols and consistent knowledge‑update rules. We further note that an agent’s trustworthiness extends beyond behavioral safety to include cognitive validity—how appropriately the agent uses its knowledge during decision making.
For trustworthiness assessment, we propose several avenues: building verification frameworks based on knowledge consistency, introducing interpretability metrics, and designing long‑term monitoring to quantify cognitive reliability.
In conclusion, we compare the aspirational vision of autonomous multi‑agent systems with the current state of the art, identifying substantial gaps in perception‑memory mapping, cross‑domain knowledge sharing, and cognitive trust guarantees that require interdisciplinary advances.
Review