Abstract

Compassionate Information Systems (CIS) are designed to notice suffering, feel empathic concern, and facilitate actions that alleviate suffering (Raman & McClelland, 2019; Raman et al., 2021). However, a critical challenge confronts CIS designers: compassionate actions mediated through IS do not always land as intended. We term this phenomenon compassion discord – a systematic misalignment between the compassion giver’s intended actions and the compassion receiver’s interpretation of those intentions. For example, a system broadcasting a colleague’s distress may be experienced as exposure rather than care; an AI offering informational resources may miss a user who needs silent acknowledgment. Compassion discord is a fundamental design problem: if the system cannot account for how individuals prefer to give and receive compassion, even well-designed features risk being experienced as intrusive, dismissive, or harmful. To address compassion discord, we introduce the concept of Compassion Languages – systematic patterns in how individuals prefer to express and receive compassionate communication. Compassion Languages are theorized to operate through three mechanisms: expression patterns (preferred ways of conveying compassionate intent), reception patterns (preferred ways of receiving compassion), and alignment processes (the degree of match between the two). Drawing on communication accommodation theory (Giles, 2016), we propose that Compassion Languages vary along four dimensions – directness, temporality, resource type, and relational orientation – yielding six types of Compassion Languages: Direct Support, Presence, Resource Provision, Information Sharing, Recognition, and Protective Action. The taxonomy has direct implications for CIS design. First, it enables personalized compassion matching, pairing support seekers with support providers whose expression patterns align with seekers’ reception preferences. Second, it enables adaptive system responses where AI-driven features adjust their compassionate engagement mode (e.g., silent presence vs. informational resources vs. direct intervention) based on the user’s profile. Third, it introduces discord detection as a design requirement – systems should monitor for signals that compassionate actions are being misinterpreted and adjust accordingly. Empirically, we propose a two-phase approach: computational analysis of large-scale online discourse (e.g., Reddit communities describing compassion conflicts) to identify naturally occurring patterns, followed by interviews to validate the taxonomy and develop a Compassion Language assessment instrument. This research contributes to IS by: (1) identifying compassion discord as a previously untheorized mechanism explaining why CIS features sometimes fail; (2) providing a taxonomy that designers can use to build personalization into compassionate system features; and (3) establishing new design requirements – compassion profiling, adaptive response, and discord detection – that extend Compassionate IS research. At AMCIS, we seek feedback on the taxonomy, empirical design, and operationalization in existing platforms such as peer support and healthcare communication tools.

Download

Share

COinS