General
The FMP builds on the design research from M2.1, shifting the focus from early personalization exploration to AI memory control in self-disclosure contexts. Inspired by the Ebbinghaus forgetting curve [1] and its application in AI memory research (MemoryBank) [2], the project designed and implemented a five-level memory-strength mechanism that allowed users to regulate how strongly Hola recalled prior conversations without directly managing stored content (T&R, MD&C). A seven-day within-subjects field study compared fixed and adjustable memory-strength conditions through questionnaires and semi-structured interviews, based on 21 valid participant pairs (U&S, MD&C). The study found that memory-strength control did not directly lead to higher anticipated communication quality (ACQ) or willingness to disclose, but revealed that users' preferences reflected how they positioned their relationship with Hola, and that recall manner and memory boundaries shaped whether the experience felt supportive or unsettling.
Contribution to PI&V
This project reinforced that design should lead product development early, framing uncertain questions into researchable ones and feeding grounded insights back. In Hola, this meant shaping AI personalization from a design research perspective rather than treating it as a feature to copy. It also clarified that user-centered design in emotionally sensitive contexts requires careful interpretation of needs rather than open-ended user control over AI memory. This revealed a growth area for me in quantitative analysis, which I want to strengthen as I continue in AI product design.
Involved Expertise Areas
- U&S: within-group field study, ACQ/self-disclosure scales, thematic analysis, semi-structured interviews
- T&R: five-level memory-strength mechanism design and specification, memory scheduling logic design (iOS prototype and pipeline implemented in collaboration with Nexo Intelligent Technology Co., Ltd.)
- C&A: product interface and interaction design, memory-strength panel design
- MD&C: Ebbinghaus-inspired computational model, paired t-test, Wilcoxon signed-rank test
- B&E: commercial alignment of the research
Links
More project details see Dribble
Details related to product feature design are not available for this project due to a Non-Disclosure Agreement (NDA).
References
[1]: Ebbinghaus (1885), H. 2013. Memory: A Contribution to Experimental Psychology. Annals of Neurosciences. 20, 4 (Oct. 2013), 155-156. https://doi.org/10.5214/ans.0972.7531.200408
[2]: Zhong, W., Guo, L., Gao, Q., Ye, H. and Wang, Y. 2023. MemoryBank: Enhancing Large Language Models with Long-Term Memory. arXiv.