New Research Maps the Gradual Transfer of Human Agency to AI

A new research paper published in Frontiers in Psychology explores the incremental shift of human agency to artificial intelligence, proposing a framework for understanding how decision-making power is transferred to machines.

The work, titled "Instrumental succession: the gradual transfer of agency to AI," analyzes the conceptual framework of this transition. According to the researchers, this process involves a gradual handover of agency rather than an abrupt change. The paper specifically contrasts the capabilities of current Large Language Model (LLM) architectures with earlier AI risk frameworks, such as those previously established by researchers Omohundro and Bostrom.

The Evolution of AI Risk

This research arrives as the industry moves from static, rule-based AI to generative models capable of complex reasoning. While early AI safety discussions often focused on sudden "intelligence explosions" or catastrophic failures, the concept of instrumental succession suggests a more subtle erosion of human control. By comparing modern LLMs to the foundational theories of Bostrom and Omohundro, the authors highlight how the nature of AI risk has evolved alongside the technology's architecture.

Implications for Oversight

Understanding the gradual transfer of agency is critical for the development of AI safety, ethics, and legal frameworks. The primary concern is the identification of a tipping point—the moment where human oversight becomes nominal and actual control is lost to autonomous systems. As AI systems take over more intermediate steps in complex workflows, the ability for a human operator to intervene effectively may diminish, creating a gap in accountability and safety.

Future Frameworks

As AI integration deepens across professional and personal sectors, the focus now shifts to creating guardrails that can detect this transfer in real-time. Future research will likely need to determine how to maintain meaningful human control as systems become more autonomous. For now, the framework provided by the Frontiers publication serves as a theoretical baseline for policymakers and engineers attempting to quantify the loss of human agency in the age of LLMs. This theoretical shift suggests that the risk is not a single event, but a cumulative process of delegation that requires constant monitoring to ensure human intent remains the primary driver of autonomous action.