Introspective AI

A multi-year research project that explores opportunities and tensions for AI-mediated introspective practise.

My Roles
Product Designe
UX Design & Research
Study Facilitation
Video Animation & Edit
Analysis & Writing
Outcome
Keywords
Explainable AI
Interactive ML
Teachable ML
AI as a co-performer
Uncertainty & Unpredictability

Promise and Peril of AI-mediated Introspection

Introspection is the practice of looking inward and examining our thoughts, values, and emotions [2]. It requires us to reflect on our past experiences and contemplate questions about the future [13]. Our digital footprints can serve as reference points to our past realities, enabling us to actively contemplate them. Personal data plays an exciting role in this process as it captures our life experiences on an unprecedented scale. However, navigating through this vast amount of information can be challenging. Artificial intelligence (AI) presents promising possibilities to surface, explore, and identify patterns in our data records. However, in a space as personal and sensitive as self-development, it also raises several concerns.

◉ Research Question 1
In what ways can personal data support the practice of introspection?
◉ Research Question 2
What roles might AI play in surfacing life experiences and behavioral patterns bound up in people’s data?
◉ Research Question 3
What potential benefits and frictions exist in this emerging design space?

Experiments & Ideation

In the first phase of this research project, I conducted several activities simultaneously to familiarize myself with introspective practice, machine learning, and personal data. Further, I engaged in early design concept generation, manifesting emergent insights, ideas, and thoughts across this experimental phase.

  • I adopted a designer-researcher approach that originates with and concerns first-person-oriented, design-led research in HCI (e.g., [6,7,9–11,15]).
  • I leveraged methods of design ideation and speculative design.
  • I arrived at 7 fleshed out design proposals.

Introspective Explorations

I have tried and tested various apps, methods, and services to understand better the full extent of technological intervention in this space. I have noted their unique features, assessed their impact on my practice, and used them to inspire future stages.

  • I tried various journaling and dream journaling apps and meditation apps.
  • I spent several months using the ReplikaAI chatbot.
  • I began creating timelines of my life corresponding to different data forms.

AI & Machine Learning Explorations

This is a short glimpse into some of the self-experiments I conducted where I integrated my data with machine learning tools to reflect on these experiences and consider how they might shape my practice of and orientation toward introspection.

  • I tested the IBM Watson API using their personality insights service to process personal digitalized journal entries.
  • I created an interactive canvas that plots speech-mined data using a t-SNE algorithm.
  • I created personal styleGAN2 models synthesizing personal aesthetics over time.

Design Concept Ideation

As more and more insights from first-hand experiences emerged across the experimental phase, I documented them by sketching and annotating various design concepts.

  • This design ideation focused on creating services that use personal data in alternative and explorative ways.

Framing

We framed Introspective AI as a context-aware agent mediating interactions between users and their personal data. Our approach was to move away from the idea of AI as a single, all-knowing agent that is infallible and takes on a human-like form [1,14].
We also speculated on future forms of personal data mining where deeper layers of data extraction exist and reach into humans "affective, cognitive and physical worlds" [4].

7 Design Proposals

After several rounds of developing, refining, and reflecting on Introspective AI design concepts, we arrived at seven distinct design proposals for Introspective AI products.

1 – Everyday Personality

Everyday Personality presents a chatbot interface that delivers short and contextualized introspective prompts. This service uses its deep understanding of your behavioral data to intervene in everyday life with tailored prompts delivered at “opportune” moments.

2 – Music Reflection

Music Reflection is a Spotify integration that generates introspective prompts based on personal data correlated with the music one is listening to.

3 – Mind Probes

Mind Probes is a smartphone app that uses external sensors for sound, color, smell, haptic, and vision to prompt users to collect sensory stimuli reflecting their social and emotional associations.

4 – Vision Shrine

Vision Shrine is a device that visually tracks and updates a user's goals, dreams, and desires in real time. The device creates a data collage that represents the user's "ideal self" and can be interacted with through answering questions and resizing content.

5 – Hello, Cyberself

Hello, Cyberself offers a conversational window into the assumptions (and biases) that a personal Introspective AI has developed over time. It leverages real-time voice cloning technology [3,8] to speak to you in your own voice. It expresses introspective prompts to you as you— embodying your personality traits and beliefs and then reveals the data ‘under the hood’ that generated these inferences.

6 – Dream Streams

Dreams offer a way for your subconscious mind to communicate with your conscious self. Dreaming offers an important window into phenomena that shape our innermost desires, fears, and goals [5,12]. Dream Streams combines a dreamcatcher-like device paired with mobile applications to offer windows into one’s subconscious and open new pathways to self-awareness.

7 – Deeptalk Report

Deep Talk Report is an application that audits verbal and written conversations to find and classify deep exchanges. These analyzed accounts are curated guided introspective sessions and are also woven together to generate broader thematic reports as well as customised introspective activities.

Elicitation Study

We adopted a design fiction approach for this study and created a fictional company called Meta.Aware, to contextualize the various Introspective AI concepts in video sketches. We used this platform to conduct interviews with 17 participants, using the videos as prompts for discussion. Our Participants had a range of reactions related to perceived benefits and tensions in this emerging design space, and we report on the results as a guide for designing – or not designing – future AI tools within this space.

The Participants

We recruited a diverse sample of people to elicit a wide range of discussions about and beyond our design proposals. This sample is not meant to be statistically representative but instead generative.

  • Occupations
    Filmmaker, Artist, Big Data Analyst, Spiritual Counselor, Environmental Education, Professor of Human Geography, UX designer, and AI & UX Researcher.
  • Experience with Introspection
    11 were frequent practitioners, 4 moderate, and 2 communicated interest but did not practice often.

Context for an Alternative Present

To help participants suspend disbelief, we designed the fictional company Meta.Aware and created an Introduction Video where the "CEO" explains the company's purpose. Participants could view this video on the Meta.Aware YouTube channel, along with each of the video proposals.

Study Procedure

We interviewed all 17 Participants via Zoom sessions that lasted 2 to 3 hours.

  • Introductory Interview
    We asked participants to reflect on their use of apps in this space and what “AI” means to them. Then, we showed the Meta.Aware introductory video.
  • Design Proposal Reactions
    After presenting each video scenario, we asked the participants for their initial and unfiltered impressions.
  • Final Interview
    Participants evaluated all design proposals in relation, selected their most and least favorite services, and explained how these services could support or not support their needs and values.

Findings & Implications

Participants had varying attitudes toward the potential role of AI services in their introspective practice. Our study showed that devising concepts with controversial design traits can improve understanding of the design space, boundaries, and social acceptabilities. Next, we summarize key findings and suggest opportunities for future HCI research and design.

Theme 1
Co-Creation with new introspective resources
  • Contributing data and anticipating results
    Participants valued having more agency over what and how data is collected and that these acts were often perceived as opportunities to prompt introspection. Participants appreciated the ongoing dialogue about their data and valued the qualities of IAI services that invited them to be creative with those accounts.
  • Balancing curational control
    They also appreciated the IAI services that made their data malleable and tangible through both direct and implicit forms of interaction.
  • Manipulation and revision as co-creation
    Participants valued tweaking, exaggerating, and re-contextualizing their digital profiles with the help of AI tools, which could lead to co-creative cycles of manifesting, discovering, and contemplating new aspects of their identity.
Theme 2
New perspectives and confrontations of self
  • Distrusting analytic perspectives
    Theme 2 located a nuanced tension concerning how an IAI presents information to the user. For participants, it was highly desired for an IAI to reveal patterns and introduce new perspectives on one’s life; however, it was perceived as unacceptable to be told what these patterns might definitively mean. Participants wanted to have the opportunity to be more involved in an IAI’s decision-making process and wanted options for ‘opting in’ and ‘tuning’ the degree to which an IAI service leveraged machine analysis for introspection.
  • Confrontational Perspectives
    Surprisingly, participants also largely appreciated when the IAI exhibited bold and cheeky attitudes. These qualities were perceived as productive in pushing them out of their comfort zones, diverting from the status quo affirmational aims of most commercial applications.
  • Taking perspectives in—or not
    Yet, the more seasoned introspective practitioners expressed concern for less experienced users who could fall prey to technologically determined “self-fulfilling prophecies,” marking areas that must be handled carefully.
  • Abstract & Interpretable Perspectives
    Participants voiced their desire to move beyond ways of rational self-analysis as they seemed to be unsatisfied with their current reflective practices with personal data (e.g., via journaling and self-tracking apps).
Theme 3
Desire for Presence, Scalability, and new Rituals over time
  • Supporting Presence and Habits
    Many participants gravitated towards IAI services that were embodied in physical devices for their intentional, solitary focus on specific methods. Here, persistent, self-contained, safeguarded experiences were envisioned to nurture new and novel introspective rituals—they were often described as desired alternatives to smartphone apps that dominate the current design space.
  • Supporting Temporal Connections and Trajectories
    Some participants envisioned longer-term human-technology relations with the IAI services where the contemplative rewards could deepen over time. They expressed enthusiasm for tools that could help them to explore temporal interrelations among their past-, present- and future selves.
  • More than just a user
    Lastly, participants expressed desires to socially share resources generated by IAI services as well as participate more directly in the development of IAI models that extend to broader communities of practice – both of which highlight key needs not well supported by currently available AI applications and services.
☆ Implication 1
Balancing curational control and autonomy through Ludic interaction design.
☆ Implication 2
Design for the desire to tweak and exaggerate parameters to help users better understand who they are in relation to the more magnified views that the IAI model may show.
☆ Implication 3
Embrace abstract, mystical, and irrational representations of data. It doesn’t need to make sense for it to be of value for introspection.
☆ Implication 4
Preserving and exploring temporal connections among a past, present, and future self. If users gradually enrich their archives with temporally relevant data. How would those different stages of self be brought into context?
☆ ...
More in the Paper ↓

CHI'23 Paper

Envisioning and Understanding Orientations to Introspective AI: Exploring a Design Space with Meta.Aware

Nico Brand, William Odom, and Samuel Barnett. 2023. Proceedings of SIGCHI Conference on Human Factors in Computing Systems. Hamburg, Germany, CHI’23. ACM Press.

Master's Thesis

A Design Inquiry into Introspective AI: Surfacing Opportunities, Issues, and Paradoxes

Nico Brand Spring 2022.

Design Pictorial

A Design Inquiry into Introspective AI: Surfacing Opportunities, Issues, and Paradoxes

Nico Brand, William Odom, and Samuel Barnett. 2021. In Proceedings of Designing Interactive Systems. Virtual Event. DIS ’21. ACM Press. 12 pages.

Acknowledgements

This research took place in the Greater Vancouver area in Canada on the unceded traditional territories of the Coast Salish peoples of the Katzie, Kwantlen, Kwikwetlem (kwikwəƛ̓əm), Qayqayt, Musqueam (xwməθkwəyəm), and numerous Stó:lō Nations.

This research is supported by the Natural Sciences and Engineering Research Council of Canada (NSERC), the Social Sciences and Humanities Research Council of Canada (SSHRC), and the Canada Foundation for Inno- vation (CFI).

We thank our participants for generously sharing their experiences with us and Aamir Ali, Chiara Schmitt, Chiara Ferrari, Julian Goto, and Vanessa Montoya for their assistance on this project. We also thank the anonymous reviewers for their highly constructive feedback on both publications.

Credits

Team: Leonard Weigand, Vincent Fischer, Nico Brand
Consulting: Benedikt Groß, Michael Schuster
University: Simon Fraser University , Fall 2019 → Fall 2022

back on top next project previous project

Sources

  1. Jesse Josua Benjamin, Arne Berger, Nick Merrill, and James Pierce. 2021. Machine Learning Uncertainty as a Design Material: A Post-Phenomenological Inquiry. arXiv preprint arXiv:2101.04035 (2021).
  2. Alex Byrne. 2005. Introspection. Philosophical Topics 33, 1 (2005), 79–104.
  3. Jemine Corentin. 2019. Real-Time-Voice-Cloning. Retrieved from https://github.com/CorentinJ/Real-Time-Voice-Cloning
  4. Kate Crawford and Vladan Joler. 2019. Anatomy of an AI System. Virtual Creativity 9, 1 (December 2019), 117–120. DOI:https://doi.org/10.1386/vcr_00008_7
  5. Daniel C. Dennett. 1976. Are dreams experiences? The Philosophical Review 85, 2 (1976), 151–171.
  6. Audrey Desjardins, Jeremy E. Viny, Cayla Key, and Nouela Johnston. 2019. Alternative Avenues for IoT: Designing with Non-Stereotypical Homes. In Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems, ACM, Glasgow Scotland Uk, 1–13. DOI:https://doi.org/10.1145/3290605.3300581
  7. William W. Gaver. 2006. The video window: my life with a ludic system. Pers Ubiquit Comput 10, 2–3 (April 2006), 60–65. DOI:https://doi.org/10.1007/s00779-005-0002-2
  8. Ye Jia, Yu Zhang, Ron J. Weiss, Quan Wang, Jonathan Shen, Fei Ren, Zhifeng Chen, Patrick Nguyen, Ruoming Pang, Ignacio Lopez Moreno, and Yonghui Wu. 2018. Transfer learning from speaker verification to multispeaker text-to-speech synthesis. In Proceedings of the 32nd International Conference on Neural Information Processing Systems (NIPS’18), Curran Associates Inc., Red Hook, NY, USA, 4485–4495.
  9. Carman Neustaedter and Phoebe Sengers. 2012. Autobiographical design in HCI research: designing and learning through use-it-yourself. (2012), 10.
  10. William Odom and Tijs Duel. 2018. On the Design of OLO Radio: Investigating Metadata as a Design Material. In Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems (CHI ’18), Association for Computing Machinery, Montreal QC, Canada, 1–9. DOI:https://doi.org/10.1145/3173574.3173678
  11. James Pierce and Eric Paulos. 2015. Making Multiple Uses of the Obscura 1C Digital Camera: Reflecting on the Design, Production, Packaging and Distribution of a Counterfunctional Device. In Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems, ACM, Seoul Republic of Korea, 2103–2112. DOI:https://doi.org/10.1145/2702123.2702405
  12. S. R. F. Price. 1986. The Future of Dreams: From Freud to Artemidorus. Past & Present 113 (1986), 3–37.
  13. Robert Van Gulick. 2000. Inward and upward: reflection, introspection, and self-awareness. Philosophical Topics 28, 2 (2000), 275–305.
  14. Qian Yang. 2018. Machine Learning as a UX Design Material: How Can We Imagine Beyond Automation, Recommenders, and Reminders?
  15. Revealing Tensions in Autobiographical Design in HCI | Proceedings of the 2018 Designing Interactive Systems Conference. Retrieved October 13, 2021 from https://dl-acm-org.proxy.lib.sfu.ca/doi/10.1145/3196709.3196781