Tentacular Artificial Intelligence, and the Architecture Thereof, Introduced

Selmer Bringsjord, Naveen Sundar Govindarajulu, Atriya Sen, Matthew Peveler, Biplav Srivastava, Kartik TalamadupulaarXiv 2018 · OpenReview

Abstract

We briefly introduce herein a new form of distributed, multi-agent artificial intelligence, which we refer to as "tentacular." Tentacular AI is distinguished by six attributes, which among other things entail a capacity for reasoning and planning based in highly expressive calculi (logics), and which enlists subsidiary agents across distances circumscribed only by the reach of one or more given networks.

Introduction

Standard AI agents are typically designed for narrow, fixed environments with predetermined goals and actions. A robotic arm programmed to move a cup to a saucer operates within a strictly bounded "blocks-world." However, real-world environments are messy, dynamic, and populated by multiple actors—humans, pets, and other machines—each with their own beliefs and constraints. When a traditional agent's capabilities are exceeded or its environment changes unexpectedly, it becomes hamstrung.

To achieve true artificial general intelligence, we require agents that can break out of these narrow limitations. Tentacular Artificial Intelligence (TAI) is proposed as a solution: a distributed, multi-agent architecture where agents use highly expressive formal logic to reason, plan, and delegate tasks across the Internet of Things (IoT).

The TAI Architecture and Logic

The core of the TAI approach is a logicist framework powered by the Deontic Cognitive Event Calculus ($\mathcal{DCEC}$). Unlike standard extensional planning languages (like PDDL), $\mathcal{DCEC}$ is a sorted first-order modal logic. It integrates the "event calculus" for reasoning about time and physical states with modal operators for intensional states such as Knowledge ($K$), Belief ($B$), Intention ($I$), and Obligation ($O$).

The TAI agent decision-making flow.

As shown in the architecture above, a TAI agent acts as a central coordinator. When assigned a goal, it first attempts to mathematically prove whether a single-agent plan exists within its own constraints. If it proves that no such plan is possible, it formulates a multi-agent plan, recruiting other "tentacles"—subsidiary agents or devices across the network—to execute specific actions.

Planning, Delegation, and Theory of Mind

A prerequisite for effective delegation is a "Theory of Mind." To coordinate with others, an AI must be able to model the cognitive states of external entities. It needs to know what another agent knows, or believe that a human is unaware of a specific danger.

Temporal progression of TAI reasoning from goal identification to multi-agent execution.

The TAI protocol ensures that both goals and plans are represented in the same logical language. This allows for seamless communication and formal verification. The agent doesn't just output a black-box decision; it provides a formal justification and certification that its solution is both effective and adheres to ethical or legal constraints.

Ethical Reasoning in High-Stakes Scenarios

Because $\mathcal{DCEC}$ includes deontic operators (rules about obligation and permission), TAI agents can weigh conflicting rules. In a scenario involving a carbon monoxide leak in a smart home where physical alarms are broken, a TAI agent can formally prove that it cannot solve the problem alone. It can then formulate a plan to override a smartphone's "Do Not Disturb" mode or blast a bedroom TV to wake the residents. In this case, the agent reasons that the moral obligation to save a life overrides the legal or contractual prohibition against bypassing device privacy settings.

Conclusion and Future Directions

While the TAI research program is currently in an early state, it offers a robust framework for transparent and ethically-aware distributed AI. Future work will address the computational complexity of real-time automated theorem proving in expressive logics and explore privacy-preserving communication methods like zero-knowledge proofs to mitigate the concerns raised by the pervasive reach of "tentacular" agents.

Summary updated

Abstract

We briefly introduce herein a new form of distributed, multi-agent artificial intelligence, which we refer to as "tentacular." Tentacular AI is distinguished by six attributes, which among other things entail a capacity for reasoning and planning based in highly expressive calculi (logics), and which enlists subsidiary agents across distances circumscribed only by the reach of one or more given networks.
BibTeX
@misc{bringsjord2018tentacular,
  author = {Selmer Bringsjord and Naveen Sundar Govindarajulu and Atriya Sen and Matthew Peveler and Biplav Srivastava and Kartik Talamadupula},
  title = {{Tentacular Artificial Intelligence, and the Architecture Thereof, Introduced}},
  year = {2018},
  eprint = {1810.07007},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1810.07007},
}
Figure