Returning to experience
“I don’t think I’m any good at leading people.”
Example AI responseCould you tell me about a particular moment when you felt that way?
Tracing the event behind the words before treating them as a description of the person.
A dialogue app in development
Reconsidering how we see
ourselves and the world.
“This is who I am.” “That is how the world works.” I am developing an AI dialogue app that returns to the experiences behind these feelings and words, and explores how we might engage with the world from here.
We bring many experiences together into a way of seeing. It helps us navigate everyday life. Yet as those experiences settle into a few words, the feelings and details of our encounters with others can become harder to see.
The aim is to trace those experiences carefully and encounter new ones, reconsidering how we see ourselves and the world through dialogue.
01 / A CONVERSATIONStarting with a conversation
“I’m not cut out for this.”
What happens when we say that?
“I’m no good at leading people” may bring together events in a meeting, feelings during a silence, other people’s reactions, and expectations about what a leader should be. We explore what is held in those words by listening to the person’s experience.
This imagined conversation illustrates the design direction. It is not a record of actual use or a comparison with ChatGPT.
Attending to the details
“Everyone went quiet in a meeting, and I panicked. I thought I had to make a decision right away.”
Example AI responseDo you remember what that silence felt like?
Listening to what the person felt and how they understood the moment, without rushing to explain its cause.
Examining a way of seeing
“I felt they thought I was unreliable. But no one actually said anything. Maybe they were thinking, too.”
Example AI responseSo what the silence meant is still uncertain. If something similar happens again, how would you like to respond?
Neither explanation is taken as settled. The conversation explores what could be checked through experience.
Engaging in the world
“At the next meeting, I asked if people wanted time to think. After a pause, they offered some ideas. But I still felt uneasy while I waited.”
Example AI responseThe waiting still felt uneasy. Now that you’ve experienced people speaking up after a pause, how do you see the silence in the earlier meeting?
Returning to an earlier understanding through a new experience, attending both to what has shifted and to feelings that remain.
Following words back to experience.
Returning to those words through new experience.
02 / EXPERIENCEAcross conversations
Thinking and experience
can change each other.
In this dialogue, tracing the experiences and assumptions behind a judgment is part of analysis. New experiences can also unsettle what seemed clear in words. The conversation moves among these three activities as needed.
01 Unpacking a judgment
What has been gathered
into these words?
Events, feelings, others’ reactions, and standards taken for granted. Together, we trace what is held in a judgment such as “I can’t do this.”
Distinguishing what has been brought together may reveal uncertainties and other ways to understand the experience.
02 Revisiting an experience
What was happening
in that moment?
Alongside considering explanations, we return to feelings and interactions that may have been overlooked. We explore them in the person’s own words, rather than fitting them to the AI’s interpretation.
Retelling an event may reveal new meanings, as well as feelings that do not readily change.
03 Returning with new experience
Reconsidering a view
after trying something.
Asking someone, waiting a little, discussing roles. We explore ways of engaging that the person wants to try, then bring what actually happened into the next conversation.
Unexpected responses and changes in how things feel can become reasons to reconsider an earlier understanding.
Analysis opens up experience; experience changes analysis. This movement between the two is central to the work.
03 / POSSIBLE IMPACTThe change I hope to support
Seeing ourselves and the world differently.
Finding more ways to act and relate.
We may find another way to engage with something we once thought we could not do.
Someone who understood silence as evidence of their own inadequacy may become able to ask others or wait a little. That experience may change how they understand their role and other people’s responses. I am interested in how small shifts like these might extend into everyday choices and relationships.
Change will not necessarily look the same for everyone. Some people may begin to act; others may step back from expectations or ask for help. What changes become more likely through repeated conversations of a particular kind, for whom, and under what conditions? That is itself a question this work aims to explore.
Our options are not determined by perspective alone. Time, money, institutions and relationships shape what we can try. Dialogue may also help make those conditions visible and support people in finding the assistance and connections they need.
04 / DESIGNWhat guides this AI
Helping these conversations continue
when they are needed.
Conversations like these can also happen with ChatGPT. Offering advice, asking questions and reflecting on experience are all things both AIs can do.
I am working toward dialogue that can continue beyond a single exchange, taking earlier conversations and subsequent experiences into account. That means revising an understanding in response to what someone says now, rather than fixing a picture of them from past words. I am developing the memory and response mechanisms to support this.
Concrete help when someone needs to know what to do. Time to explore together when their thoughts are still taking shape. What response is needed depends on the person and the situation.
A closer look: response tendencies and distributions
The difference concerns which responses are more likely in a given situation, rather than whether a response is possible at all. Across many conversations, we can think of this pattern of responses as a distribution.
Two AIs may give similar replies. Over time, differences may also emerge in what they tend to attend to and how conversations develop. Both the overlap and the differences in tendency inform the design.
Thinking of response patterns as overlapping distributions
The aim is not to steer every conversation in the same direction. It is to make room for reconsidering one’s understanding and connecting with experience when those opportunities are needed.
05 / OPEN QUESTIONSWhat remains to be explored
Examining the distance between
intended change and what happens.
This page describes design intentions and hypotheses about possible change. Whether response patterns shift as intended, and whether those shifts help people, are separate questions to investigate.
Are patterns of dialogue changing?
Compare multiple situations and the course of conversations, rather than judging by one striking reply. Examine how opportunities to revisit self-understanding and reflect on experience arise.
What happens in everyday life?
Look beyond agreement during a conversation to what people subsequently try and how their choices and relationships change, drawing on their own words and concrete experiences.
For whom, and when, is it helpful?
Sometimes people want to reflect; sometimes they need immediate, practical help. Include situations where questions become a burden and cases where no change occurs.
Can the AI’s understanding be corrected?
Can the person challenge the AI’s view? Can it revise its understanding from what is said now, without clinging to the past? Examine whether the conversation steers people toward the AI’s interpretation.
A question connecting research and practice
What kinds of dialogue
open up what possibilities?
From the AI’s responses to repeated experiences, and from there to everyday actions and relationships. I am exploring these connections by building, using and reflecting on dialogue.