Dynamic Agent Affiliation: Who Should the AI Agent Work for in the Older Adult's Care Network?

Link to paper

Conference: ACM Designing Interactive Systems Conference (DIS), 2024

Authors: Mai Lee Chang, Alicia (Hyun Jin) Lee, Nara Han, Anna Huang, Hugo Simão, Samantha Reig, Abdullah Ubed Mohammad Ali, Rebekah Martinez, Neeta M Khanuja, John Zimmerman, Jodi Forlizzi, Aaron Steinfeld

My Role: Lead Researcher

Research Overview

This research investigates a critical question in AI agent design for elderly care: who should an AI agent work for when supporting older adults experiencing cognitive decline? As populations age and the availability of care workers decreases, AI agents could help bridge the gap in elder care. However, their adoption depends heavily on trust, which becomes complicated when an agent might need to report information about an older adult that they don't want shared. Through a speed dating study with storyboards, we explored how agent affiliation (who the agent works for) affects acceptance among both healthy older adults and those experiencing cognitive decline, along with their caregivers.

Innovation

This research introduces the novel concept of "dynamic agent affiliation" - the idea that an AI agent's loyalty and decision-making priorities should shift over time as an older adult's cognitive health changes. Unlike previous work that treated agent affiliation as static, our research revealed that both older adults and caregivers expect and prefer the agent's affiliation to transition gradually from the older adult to their caregivers as cognitive decline progresses. This represents a significant innovation in how we conceptualize human-agent relationships in care settings, moving beyond simple one-to-one affiliations to more nuanced, context-aware relationships that evolve with changing circumstances.

Method

We used speed dating with storyboards - a scenario-based design methodology that allows exploration of potential futures without being constrained by current technological limitations. We developed 25 storyboards (16 primary, 9 secondary) illustrating various scenarios where an agent might face conflicts of interest between an older adult and their caregivers. These storyboards covered six key instrumental activities of daily living (IADLs): finances, transportation, socializing, physical safety, medication and appointments, and health. Each storyboard depicted familiar situations with provocative technology interventions, allowing participants to imagine and reflect on how they would want an agent to behave in these complex social situations.

Participants

We conducted semi-structured interviews with 38 participants across two groups:

• 18 healthy, independent older adults (ages 62-90)

• 10 older adults experiencing early stages of cognitive decline (age 60-78), along with their primary caregivers (age 20-70)

Participants came from diverse backgrounds and low socioeconomic status areas. 

Procedure

We conducted sessions primarily in private rooms at senior centers, with some in participants' homes or via video chat. Each session lasted about an hour and covered 2-9 storyboards, with participants responding to leading questions and semi-structured follow-up inquiries. Audio recordings were transcribed, resulting in approximately 38 hours of interview data that was analyzed using affinity diagramming to identify emerging themes.

Key Findings

Our research revealed four primary factors influencing how people think about agent affiliation:

1. Adoption Motivation: Both healthy and declining older adults showed positive reactions to adopting agent support. Healthy older adults viewed agents as enabling independence and providing security, while declining older adults were willing to trade privacy for safety and reducing caregiver burden.

2. Health Status Impact: Participants expected agent affiliation to shift dynamically as cognitive health declined. Healthy older adults envisioned the agent initially working for them and being altruistic, while both groups anticipated a gradual transition to working more for caregivers as decline progressed.

3. Support System Influence: Older adults with children expected agent affiliation to shift to their children during decline, while those without children envisioned a more intimate relationship with the agent, with greater willingness to cede control directly to the agent.

4. Perception of Agent Autonomy: Healthy older adults wanted the agent to mirror their decision-making processes, while declining older adults preferred the agent to mirror their caregivers' processes, demonstrating increasing concern about the burden they might place on caregivers.

Research Impact

This research makes two significant contributions to the field:

• It provides new knowledge about factors affecting adoption and use of cognitive support agents for aging in place, challenging previous assumptions that older adults resist new technology. Our findings show that when technology is presented as extending independence and reducing caregiver burden, both older adults and caregivers are receptive to adoption.

• It presents opportunities and considerations for designing agent affiliation to increase adoption likelihood. By understanding how affiliation expectations change with health status, support systems, and perceptions of agent autonomy, developers can create agents that better align with users' evolving needs and relationships.

The work has important implications for addressing the growing gap between the population of older adults experiencing cognitive decline and the availability of care workers. By designing agents that can navigate shifting affiliations appropriately, we may better support aging in place while maintaining trust among all stakeholders.

Future Directions

This research opens several promising avenues for future investigation:

• Quantifying Affiliation Changes: Developing metrics to detect when and how agent affiliation should shift, and identifying appropriate triggers for these transitions

• Communication of Affiliation: Creating interaction designs that effectively signal to all stakeholders when and how agent affiliation is changing

• Ethical and Legal Implications: Addressing questions around agent decision-making during affiliation transitions, including potential conflicts between safety, autonomy, and privacy

• Mediator Role Development: Exploring how agents can act as objective mediators in disagreements between older adults and caregivers, especially around sensitive topics like driving safety or living arrangements

Skills Demonstrated

Technical Skills

• Qualitative research design and implementation

• User experience research methods (speed dating with storyboards)

• Affinity diagramming for thematic analysis

• Audio transcription and qualitative data analysis

Research Skills

• Literature review and synthesis across multidisciplinary fields

• Human subjects research with vulnerable populations

• Semi-structured interview techniques

• Collaborative research in multidisciplinary teams including partnerships with local organizations

• Scientific Communication: Presented complex technical concepts clearly in academic writing and conference presentation

Domain Knowledge

• Aging in place technologies and adoption patterns

• Cognitive decline progression and impacts

• Care coordination and caregiver burden dynamics

• Human-agent interaction design principles

• Ethics in technology for vulnerable populations

• Trust and affiliation frameworks in human-agent relationships