Reflection Chatbots

How can('t) chatbots support family members and caregivers on reflecting on moments of reality disjunction with people with dementia?

2024-2026
Part of PhD programme at Eindhoven University of Technology
Activities: chatbot development, focus groups, thematic analysis.
Published at DIS 2026:
Ruitenburg, Y., Lee, M., Markopoulos, P., IJsselsteijn, W. (2026, June). My Reality or Yours? The Role of Chatbots in Reflecting on Reality Disjunction in Dementia Care. In Proceedings of the 2026 Designing Interactive Systems Conference (pp. 2725-2744). https://doi.org/10.1145/3800645.3812931

What is reality disjunction?

Imagine the following situation. You are talking to your father and suddenly he asks: "Hey, when is Mom coming home?" You are taken aback, because your mother passed away over a decade ago. How would you respond?

The situation I just described is called reality disjunction. It occurs when two people present contradicting worldviews as true. Your father might live in a reality in which his wife is still alive, while you live in one in which your mother has already passed away.

These situations can occur between any two people. For example, my boyfriend and I could think we had arranged to meet at 5:00, while he thought we would meet at 6:00. However, when someone has dementia, reality disjunction can occur more frequently and more severely. Dementia can cause memory loss, hallucinations, delusions and confabulation, so the reality of the person with dementia slowly diverges from how other people perceive the world. Dementia can also cause communication difficulties, which further hinder people from discussing their realities with each other.

As a result, people can start to doubt their own and each other's experiences. They can feel lonely as their realities slowly diverge, and frustrated as they continuously try to realign them.

What we already knew

In previous research, we looked at how people respond to these situations, specifically couples with dementia. When they valued truth, they contradicted the other person. When they valued comfort within the interaction, they tried to let go of the disjunction, either by ignoring what was said, distracting from it, or going along with the other person's reality. And when they valued connection with the other person, they tried to understand the other person's reality by asking questions.

However, the main thing we found was that there is no single right response to reality disjunction. It is highly contextual. It depends on the people involved, the topic of the interaction, and the interactional goals at hand. There is no golden rule to follow in these situations.

That is why we turned to reflection, to help people learn from their own situated and contextualised experiences of reality disjunction. We also added chatbots, since prior Human-Computer Interaction research has shown that chatbots can be beneficial for reflective conversations. So our research question became: how can chatbots support reflection on reality disjunction in dementia care?

What we did

To investigate this, I held six focus groups with 14 family members and 10 formal caregivers of people with dementia in the Netherlands. During these sessions, we explored three different chatbots together, each supporting reflection in a different way. We used scenario probes, which were illustrations of the use cases of the chatbots, and technology probes, which participants could access on their phone to interact with the chatbot. These included a fixed introduction followed by generative responses.

The three chatbots were based on prior Human-Computer Interaction literature on reflection and on common activities in dementia communication training:

The role-play chatbot offered a role-play scenario of reality disjunction so that participants could practise, and then gave step-by-step advice on how to improve next time.

The dilemma chatbot presented a dilemma of reality disjunction, together with diverse opinions on whether a particular approach was the right one.

The analysis chatbot asked detailed questions about an experience of reality disjunction and then provided a compact summary of that experience.

What we found

I analysed the data through reflexive thematic analysis. I found that participants continually compared the chatbots with the human support they already received: from colleagues, from relatives, and from peers in online support groups. They compared them on four dimensions, all related to how involved the chatbot or the human was in their reflection. Participants said that humans were highly involved contextually, emotionally, experientially and relationally, while the chatbots were not. At first I thought this was a limitation of the chatbots, but in some cases this lack of involvement turned out to be a benefit.

Contextual involvement. Participants liked that humans were contextually involved. They knew the person with dementia and the topics being discussed, so they could support reflection within the lived context. The chatbot sometimes asked irrelevant questions or gave unsuitable advice.

On the other hand, humans were sometimes limited by all of this contextual knowledge, and could not think of new ways to approach the situation anymore. They could only think in limitations, not in possibilities. The chatbot could offer reflection that was not constrained by these limitations, and suggest endless new possibilities.

Emotional involvement. Participants liked that humans could offer reflection with emotional attunement. Because humans have their own emotions, participants felt that they could understand and support them with their emotional experiences. When talking to the chatbot, participants felt creeped out when it said it understood them, and did not even want to discuss their emotions with it.

However, there was a positive side to this: because the chatbot has no emotions of its own, it did not add an extra emotional burden to the conversation. Family members and relatives sometimes added their own sadness, frustration or anger to these conversations, and participants did not want to deal with those extra emotional layers.

Experiential involvement. Participants liked that humans could reason from their own lived experience with dementia. They understood the emotional and ethical complexity of these situations and gave guidance and asked questions that suited that understanding. Participants often asked: why would I listen to this chatbot? It has no experiences of its own, so why would I care?

But the chatbot could reason from the internet, from theories and from other people's experiences found online, so it could draw on a wider range of perspectives. Humans, especially relatives, often lacked these theories and the larger understanding of the topic.

Relational involvement. Finally, participants liked that humans could offer reflection with relational responsibility. They had ongoing relationships with their relatives and colleagues, so they felt these people actually cared about them and would give advice and information they could trust. With the chatbot, they felt: it does not care whether I have a good relationship with my father or not, so why would I listen to it?

The benefit was that participants did not have to maintain a relationship with the chatbot. They did not have to worry about taking up its time or sharing controversial opinions, because they did not care what the chatbot would think or feel. With humans, they were often hesitant to reach out because they did not want to take up their time or burden them emotionally.

What this means

First, we show the benefits of reflecting on reality disjunction. During the sessions, participants described how these reflective activities helped them better understand the causes and consequences of reality disjunction, imagine other strategies and perspectives, and evaluate whether their response had been suitable for the situation. Moreover, doing this in a group, where people could share experiences, advice, dilemmas and perspectives, made people feel less isolated in their experiences. It also provided the emotional validation that participants were looking for.

As for how to design reflective experiences, we argue that humans should stay on as reflective partners. Participants were adamant that humans could never and should never be replaced by chatbots, because only humans fully offered contextual understanding, lived experience, emotional attunement, and relational responsibility. But chatbots can be used as reflective tools that add to these reflective conversations. They are great for asking reflective questions, generating summaries, practising role play and giving feedback based on theories. Participants also liked that chatbots could help expand their thinking by generating alternative strategies, alternative perspectives, and different explanations for why something occurs. But they wanted humans, meaning themselves, their family members and their relatives, to decide which option was contextually appropriate. They felt that the chatbot did not have the judgment to do that for them.

Lastly, we argue that we should move on from designing chatbots that aim to replicate or replace human support. Participants were adamant that they do not want their current human support to be replaced by chatbots. They want to stay connected to their loved ones and to the support system around them. Instead, I recommend first identifying how chatbot support differs from the support people can already receive from their human support system, and only then introducing a chatbot where its strengths can complement the gaps in the human support currently available.

Next project

Page updated

Google Sites

Report abuse