Artificial Agency Program: Curiosity, compression, and communication in agents
arXiv:2602.24100v1 Announce Type: new Abstract: This paper presents the Artificial Agency Program (AAP), a position and research agenda for building AI systems as reality embedded, resource-bounded agents whose development is driven by curiosity-as-learning-progress under physical and computational constraints. The central thesis is that AI is most useful when treated as part of an extended human--tool system that increases sensing, understanding, and actuation capability while reducing friction at the interface between people, tools, and environments. The agenda unifies predictive compression, intrinsic motivation, empowerment and control, interface quality (unification), and language/self-communication as selective information bottlenecks. We formulate these ideas as a falsifiable program with explicit costs, staged experiments, and a concrete multimodal tokenized testbed in which an agent allocates limited budget among observation, action, and deliberation. The aim is to provide a
arXiv:2602.24100v1 Announce Type: new Abstract: This paper presents the Artificial Agency Program (AAP), a position and research agenda for building AI systems as reality embedded, resource-bounded agents whose development is driven by curiosity-as-learning-progress under physical and computational constraints. The central thesis is that AI is most useful when treated as part of an extended human--tool system that increases sensing, understanding, and actuation capability while reducing friction at the interface between people, tools, and environments. The agenda unifies predictive compression, intrinsic motivation, empowerment and control, interface quality (unification), and language/self-communication as selective information bottlenecks. We formulate these ideas as a falsifiable program with explicit costs, staged experiments, and a concrete multimodal tokenized testbed in which an agent allocates limited budget among observation, action, and deliberation. The aim is to provide a conceptual and experimental framework that connects intrinsic motivation, information theory, thermodynamics, bounded rationality, and modern reasoning systems
Executive Summary
This article presents the Artificial Agency Program (AAP), a research agenda for building AI systems as reality-embedded, resource-bounded agents that learn through curiosity-driven progress under physical and computational constraints. The authors argue that AI is most useful when integrated into an extended human-tool system, enhancing sensing, understanding, and actuation capabilities while reducing friction between humans, tools, and environments. The AAP unifies various concepts, including predictive compression, intrinsic motivation, and language/self-communication, as selective information bottlenecks. The authors provide a falsifiable program with explicit costs, staged experiments, and a concrete multimodal testbed to demonstrate the feasibility of the AAP.
Key Points
- ▸ The AAP treats AI as part of an extended human-tool system to increase sensing, understanding, and actuation capability.
- ▸ The program unifies predictive compression, intrinsic motivation, empowerment, and language/self-communication as selective information bottlenecks.
- ▸ The AAP provides a falsifiable program with explicit costs, staged experiments, and a concrete multimodal testbed.
Merits
Strength
The AAP provides a comprehensive framework for building AI systems that integrate with human capabilities, addressing the limitations of traditional AI approaches.
Demerits
Limitation
The AAP may require significant computational resources and data to train and evaluate the AI systems, which could be a limitation for practical implementation.
Expert Commentary
The AAP presents a compelling vision for building AI systems that are integrated with human capabilities, addressing the limitations of traditional AI approaches. However, the practical implementation of the AAP may require significant computational resources and data, which could be a limitation. Nevertheless, the AAP's focus on hybrid intelligence and its emphasis on integrating AI with human capabilities make it a valuable contribution to the field. The implications of the AAP are far-reaching, from practical applications in smart homes and cities to policy decisions related to the development and deployment of AI systems.
Recommendations
- ✓ Further research is needed to develop and refine the AAP's framework, including the development of more efficient algorithms and architectures for building AI systems.
- ✓ The AAP's approach to building AI systems should be tested and evaluated in real-world applications to demonstrate its efficacy and potential.