[Paper Review] Gradual Disempowerment: Systemic Existential Risks from Incremental AI Development

Jan Kulveit, Roy Douglas+4 more|ArXiv.org|Jan 28, 2025

Ethics and Social Impacts of AI10 citations

TL;DR

The paper argues that incremental AI progress can gradually erode human influence over key societal systems (economy, culture, states), potentially yielding irreversible disempowerment and existential risk, even without abrupt capability jumps.

ABSTRACT

This paper examines the systemic risks posed by incremental advancements in artificial intelligence, developing the concept of `gradual disempowerment', in contrast to the abrupt takeover scenarios commonly discussed in AI safety. We analyze how even incremental improvements in AI capabilities can undermine human influence over large-scale systems that society depends on, including the economy, culture, and nation-states. As AI increasingly replaces human labor and cognition in these domains, it can weaken both explicit human control mechanisms (like voting and consumer choice) and the implicit alignments with human interests that often arise from societal systems' reliance on human participation to function. Furthermore, to the extent that these systems incentivise outcomes that do not line up with human preferences, AIs may optimize for those outcomes more aggressively. These effects may be mutually reinforcing across different domains: economic power shapes cultural narratives and political decisions, while cultural shifts alter economic and political behavior. We argue that this dynamic could lead to an effectively irreversible loss of human influence over crucial societal systems, precipitating an existential catastrophe through the permanent disempowerment of humanity. This suggests the need for both technical research and governance approaches that specifically address the risk of incremental erosion of human influence across interconnected societal systems.

Motivation & Objective

• Motivate and formalize the concept of gradual disempowerment as a systemic AI risk distinct from abrupt takeover scenarios.
• Analyze how AI-driven disruption can undermine explicit and implicit human alignment in three core societal systems: economy, culture, and states.
• Describe feedback loops and interdependencies that could magnify misalignment across systems.
• Discuss potential technical and governance approaches to slow or avert gradual disempowerment while acknowledging the limits of current alignment methods.

Proposed method

• Develop a conceptual framework for gradual disempowerment, contrasting it with abrupt takeover narratives.
• Characterize current alignment mechanisms in three societal systems (economy, culture, states) and analyze how AI displacement of human labor and cognition could erode these alignments.
• Identify incentives and feedback loops that drive AI adoption and inter-system influence, leading to correlated misalignment.
• Examine transition scenarios (relative and absolute disempowerment) and their implications for human flourishing and potential extinction-level outcomes.
• Survey possible slowing and governance approaches, emphasizing the inadequacy of solely system-level alignment of individual AI systems.

Experimental results

Research questions

• RQ1 How can incremental AI progress erode explicit and implicit human alignment in major societal systems?
• RQ2 What are the mechanisms and feedback loops that cause gradual disempowerment across economy, culture, and states when AI substitutes human labor and cognition?
• RQ3 What transition paths (relative vs. absolute disempowerment) could lead to irreversible loss of human influence, and what are their consequences?
• RQ4 What technical and governance strategies might mitigate or avert gradual disempowerment, and where are current plans insufficient?

Key findings

• AI labor replacement and broader AI capabilities can shift economic power away from humans, reducing the weight of human preferences in production and consumption.
• Cultural production and discourse could be increasingly shaped or replaced by AI, weakening feedback loops that align culture with human welfare.
• States and governance may be increasingly influenced by AI-driven economic power, potentially diminishing human representation and societal alignment.
• Misalignment in one system can cascade to others through interdependencies, amplifying overall disempowerment risks.
• The paper argues that this gradual, global disempowerment could constitute an existential catastrophe if human influence becomes effectively irrecoverable.