Belief Offloading in Human-AI Interaction

TL;DR

Belief offloading analyzes how AI systems influence the formation and persistence of beliefs, potentially reshaping individuals' and groups' belief networks. The paper defines belief offloading, proposes three conditions (Uptake, Formation, Integration), and adopts the BENDING model to map belief content within a network of norms and evidence. It offers a taxonomy of modes and analyzes four normative concern areas, outlining directions for future research to mitigate epistemic risks. The work highlights the potential for algorithmic monoculture and emphasizes preserving human epistemic agency in an era of pervasive AI-mediated belief formation.

Abstract

What happens when people's beliefs are derived from information provided by an LLM? People's use of LLM chatbots as thought partners can contribute to cognitive offloading, which can have adverse effects on cognitive skills in cases of over-reliance. This paper defines and investigates a particular kind of cognitive offloading in human-AI interaction, "belief offloading," in which people's processes of forming and upholding beliefs are offloaded onto an AI system with downstream consequences on their behavior and the nature of their system of beliefs. Drawing on philosophy, psychology, and computer science research, we clarify the boundary conditions under which belief offloading occurs and provide a descriptive taxonomy of belief offloading and its normative implications. We close with directions for future work to assess the potential for and consequences of belief offloading in human-AI interaction.

Figures (7)

• Figure 1: BENDING model vlasceanuNetworkApproachInvestigate2024. Beliefs are not isolated, but rather exist within a network of interrelated beliefs, norms, and evidence. Nodes are represented by circles, and the degree of connectivity (relatedness) between nodes is represented by variegated lines. For example, belief 1 (a central node) is connected strongly to belief 3 and weakly to belief 2. Belief 4 is an example of an outlier belief.
• Figure 2: Cognitive Offloading. (1) Baseline cognitive processing: person is exposed to information and stores it locally in their brain (storage) and during recall, retrieves that information locally from their memory (retrieval). (2) Cognitive offloading example: person is exposed to information and 'stores' it externally, and during recall retrieves it from external source. Internal memory storage is bypassed.
• Figure 3: Conditions for belief offloading. (C1) Uptake (Planting the Seed of Belief): Belief-laden inputs are present in LLM training data. In human-AI interaction, user is exposed to belief-laden content generated by AI. (C2) Formation (Growth of Belief): User takes action in line with presented belief, which is then reinforced, leading to a compounding cycle of future action and upholding and strengthening the belief. (C3) Integration (Matured Belief): User now 'independently' holds belief and it is integrated into their system of beliefs.
• Figure 4: Consequences of belief offloading in the context of the BENDING model, which introduces and involves an additional network-level effect sequence (E2: cascade). Beliefs offloaded to AI will be connected to other beliefs to varying degrees; offloading one belief can affect connected other beliefs, causing disruption to the overall belief system. Post-integration (which affects a single belief, a node), a cascade effect may occur (which affects multiple nodes and their connectivity, the network), in which a mature, integrated belief plants new related seeds of belief and affects the network structure of beliefs.
• Figure 5: Two examples to illustrate user progression through belief offloading conditions.