‘I already forgot half of it’ – Interviewing people with dementia for co-designing an intelligent system

ABSTRACT During the pre-design phase of supportive intelligent systems, interviews with future users, such as people with dementia, aim to explore the problems for which technology may present a solution. However, interviewing someone with dementia in this unstructured design phase is reported to be challenging resulting in systems that are not truly addressing the needs of the people with dementia. Existing guidelines on creating engaging interviews are often based on post-hoc reflections of researchers and participants. In this study we performed line-to-line sequential in-depth analyses on transcripts of 10 interviews conducted to co-design an interactive intelligent gaming system. Both interviewers and interviewees were subsequently asked to reflect on their experiences. Results show that information exchange with people with dementia can be supported by carefully constructed questions, well-designed probes, and an appropriate length of the interview. To create a pleasant atmosphere during the interviews, researchers should design the interview as a fulfilling experience, be prepared to respond respectfully to negative responses and be able to adapt to different signs of consent. Improving the engagement with people with dementia during interviews can enhance the design and development of intelligent systems that cater to their unique challenges and requirements.


Introduction
The number of people with dementia in western societies has been increasing over the past decades and in the absence of an effective cure will likely continue to do so (World Health Organization, 2017).Dementia describes a group of cognitive symptoms related to pathological changes in the structure and function of the brain, such as Alzheimer's disease.People with dementia may experience changes in memory, attention, reasoning, and communication abilities that eventually require an increasing need for care by family members and care professionals.The number of people who are employed in dementia care, however, is not increasing, leading to a shortage of formal carers in the near future.This calls for new ways of meaningful support for independent living and overall well-being of people with dementia and their family members.The use of technology is frequently proposed as a solution for providing necessary support.Artificial intelligence (AI) is increasingly becoming a key element of these technologies in care for older adults (Ho, 2020;Lukkien et al., 2023;Rubeis, 2020) resulting in more intelligent assistive technologies (IATs).Such intelligent systems can enhance human abilities and activities while responding to the user's needs and adapting to changing environments (Ienca et al., 2017).They utilize various algorithms such as activity recognition, context modeling, anomaly detection, location and identity differentiation, and automatic planning and scheduling (Rashidi & Mihailidis, 2013).For these systems to be intuitively usable and truly suited, it is essential to iteratively develop them together with the intended endusers: both the person with dementia and their caregivers.
Earlier research highlights the importance of actively involving people living with dementia in all phases of development of supportive technologies (Gibson et al., 2015;Holbø et al., 2013;Suijkerbuijk et al., 2015;Wang et al., 2019).During the earlier phases, participants can influence the conceptual idea, rationale, and functionalities of the system.Involving people with dementia in the later phases provides insights into interface design or implementation issues.Interviewing is a commonly utilized method for actively engaging people with dementia in these different phases of technology development (Span et al., 2013;Suijkerbuijk et al., 2019).However, attention is needed in creating a safe context for people with dementia to exchange information during interviews (Pratt, 2002).In the current study, we carefully analyzed transcripts of interviews with people with dementia and their (in)formal carers in the pre-design phase of the development of an intelligent gaming system.Moreover, we will provide a reflection on what we can learn from these analyses for using interviews in the process of co-designing more meaningful intelligent systems with people with dementia.

Intelligent systems for people living with dementia
A substantial body of work demonstrates the added value of intelligent systems to support people with dementia and their close relatives in their social contacts, daily activities, and feelings of safety (Bastoni et al., 2021;Evans et al., 2015;Ienca et al., 2017;Tsertsidis et al., 2021).Such intelligent supportive technologies utilize computing capabilities, robotics, and machine intelligence.They can also provide adaptive responses that maximize benefits for users, such as people living with dementia.Touch-screen multimedia systems, for example, can assist family members and caregivers in engaging people with dementia to reminisce on significant life events (Crete-Nishihata et al., 2012;Samuelsson & Ekström, 2019).The research by Crete-Nishihata and colleagues, also concluded that more user-friendly interfaces are required to enable people with dementia to fully utilize such systems.The use of tactile objects that automatically play a soundscape to stimulate conversation can overcome the difficulties associated with touchscreen operation for this user group (Houben et al., 2020).Other examples of potentially more intuitive user interfaces pertain to intelligent systems such as virtual assistants, robotics, and voice user interfaces.König and colleagues interviewed multiple people with dementia on the use of a virtual assistant which was aimed to aid in daily activities (König et al., 2016).Most of the remarks made by the participants concerned the need for personalization in such a device, which is in line with general needs in technology for people with dementia (Hirt et al., 2019).The challenge with virtual assistant interfaces is that they can be perceived as condescending, often focussing on tasks that people with dementia cannot perform independently.Similar insights are found in the use of different forms of robotic interfaces (Begum et al., 2015), although there are also positive experiences reported such as entertainment, stimulation, and calming effects (Alonso et al., 2019), and possibly even human-robot friendships (Torta et al., 2014).These positive experiences may be particularly significant if social robots incorporate environmental information through smart sensors (Casaccia et al., 2019).In addition, voice user interfaces might be of interest for people with dementia as well, although it is suggested that further investigation is needed into the perception and adoption of these interfaces by older adults (Sin et al., 2022).Researchers have suggested that an inclusive design process fosters much needed contextsensitiveness when developing AI technologies, which seems especially relevant in the long-term care domain (Lukkien et al., 2023) to avoid AI ageism (Stypinska, 2022).Additionally, involving people with dementia in the design of intelligent systems might also improve the adoption rates of such systems (Astell et al., 2009).

Active involvement of people with dementia in design of intelligent systems
Technology development can be divided into four iterative design phases; pre-design, idea generation, evaluation and post-design (Sanders & Stappers, 2014).Insights from literature reviews indicate that people with dementia are still only marginally involved in most of these phases, their role being particularly small in the pre-design and post-design phase (Span et al., 2013;Suijkerbuijk et al., 2019).The pre-design phase, also known as the fuzzy-front end, is the first step in the development process.At this starting point, interviews with future users are often the core means to explore the fundamental problems and opportunities from an end-user's perspective.This will determine the context of innovation, and the true nature of the challenges for which intelligent systems may present a solution.However, interviewing, or otherwise involving someone with dementia in this open and quite unstructured phase is reported to be challenging, as they may find it difficult to answer explorative questions (Dethlefs et al., 2017) and to reflect on their own experiences over time (Mayer & Zach, 2013;Meiland et al., 2012).Early in the design process, these challenges can affect the direction of intelligent system design.As a result, designers may prioritize the needs and wishes that are more clearly expressed by formal and informal caregivers (Miranda-Castillo et al., 2013) resulting in systems that tend to focus on safety and monitoring instead of promoting social and leisure time activities that foster inclusivity (Lazar et al., 2017;Smith & Mountain, 2012).We are, therefore, in dire need of better understanding how to elicit the perspectives of people with dementia in interviews early in the design process of intelligent systems.

Interviewing people with dementia
General guidelines to make interviews work and enable people with dementia in giving their input have been formulated before (Astell et al., 2009;Brankaert et al., 2015;Carmody et al., 2015;Cridland et al., 2016;Hendriks, 2019;Mayer & Zach, 2013;Nygård, 2006;Wang et al., 2019;Williams et al., 2020).These scholars advise adapting the length, time, and place of each interview with a person with dementia to their capacity to concentrate and to make them feel at ease.Furthermore, they provided insights in how to frame questions to enable people with dementia in sharing their experiences, such as repetition and rephrasing of questions, developing single-faceted interview questions, and the use of visual prompts or stimulus material to support recall (Cridland et al., 2016;Nygård, 2006).In addition, creating a good atmosphere within the research activity is highlighted by McKillop, who himself lives with dementia and has many experiences with being interviewed (McKillop & Wilkinson, 2004).He argues that in addition to securing a comfortable setting, the interviewer should have a caring personality, without showing pity.The presence of family members or formal carers can also be supportive to some (Nygård, 2006;Pratt, 2002).Nevertheless, although these generic guidelines exist for interviewing people with dementia and their carers, it is still important to deepen our understanding of the flow and dynamics between interviewers and participants and how this can be further improved to steer the design of intelligent systems toward the actual needs and wishes of this user group.An improved understanding will help researchers not to marginalize or exclude older persons, and especially people with dementia, as end users of intelligent technology (Stypinska, 2022).

The current study
In this study we analyzed transcripts of 10 interviews conducted to co-design an interactive intelligent gaming system.The gaming system was intended to be adaptable to the cognitive abilities of the person with dementia and fit their personal interests.Based on the above-mentioned literature, two main themes related to active involvement in interviews were defined: Information exchange and atmosphere.Information exchange pertains to the quality and the amount of information that is exchanged between interviewers and interviewees (Patton, 2002).This is important in the pre-design phase of technology development, as confusion by complex information or questions may disempower a person with dementia in the interview, potentially resulting in an inaccurate understanding of issues relevant to the design of the intelligent system (Cridland et al., 2016).Atmosphere pertains to actions that support open and undistorted communication between the interviewer and respondents (Pratt, 2002).This helps interviewees to talk freely so that an interviewer can tap into what is really important to them (Gubrium et al., 2012;Nygård, 2006).When the interview is a pleasant experience in the first phase of the development of an intelligent system, people are potentially motivated to participate in subsequent phases, such as interacting with early prototypes.This involvement of people with dementia in multiple phases will contribute to the development of intelligent systems which are actually in line with their needs and wishes.
The following main research questions were considered: (1) What actions contribute to the information exchange between an interviewer and a person with dementia?(2) What actions contribute to the atmosphere during an interview with a person with dementia?

Materials and methods
For the purpose of this study, we analyzed transcripts of interviews from the collaborative European Active and Assisted Living (AAL) development project MagicTable (Projectnumber: AAL-2018-5-130-SCP).The aim of the project was to develop a personalized game experience for people with dementia (Anderiesen, 2017) living at home to enhance their quality of life.The intelligent gaming system called "MagicTable" would invite users toward a combination of gamified cognitive training and social interaction personalized to their needs and wishes by use of algorithms.The intended users of this platform were the people with dementia and their informal or formal carers.During the pre-design phase of this project, interviews were conducted to explore game experiences that would be suitable for individuals living with dementia at home.The interviews covered both contentrelated aspects and initial ideas on what would be technologically feasible.The interviews resemble those in similar supportive technology development projects (Span et al., 2013;Suijkerbuijk et al., 2019).

Participants
Data were collected from 10 interviews with people in the early to mid-stage of dementia and (one of) their caregivers.The stage of dementia was defined by the involved care organization.Of the ten interviews, six interviews were conducted with spouses, two interviews with parent and child and two interviews together with a formal caregiver (see Table 1 for an overview).In total, four interviewers were involved.The interviews were primarily conducted by two student interviewers (both female, 24 and 25 years of age) who had some experience with interviewing for design purposes, but not specifically with people with dementia.Two additional interviewers (including the first author of this paper) were involved, each with over five years of experience with interviewing persons with dementia in technology development projects.They helped in setting up the interview questions and procedure and participated in two of the ten interviews.The interviewers in this project followed the guidelines in the Declaration of Helsinki and ethical approval for the interviews was granted by an independent board including an ethicist of the involved care organization.

Materials & procedure
Interviews were held at the homes of participants living in The Netherlands between March 25, 2019, and April 30, 2019.The interviews were recorded by means of a voice recorder.Information was given about the interview procedure, and, after this, participants were asked to read and sign an informed consent form.The informed consent stated what was being studied, ensured pseudonymized analysis, announced that audio and image recordings were going to be made, and made clear to the participants that they could withdraw their participation at any point in time during or after the session.The interviews took, on average, 88.5 minutes (SD = 20.9) and included 35 pre-selected main questions (with sub questions) probing themes that were relevant to the project concept such as daily structure, pleasurable and enjoyable activities, social relations, mental health, and experiences with technology.A video of a related existing product was shown and discussed as well.
To strengthen our insights from the transcript analysis, we implemented two reflective activities around the interviews.First, the interviewees were asked directly before and after the interviews about their reasons for participating and their experiences of being interviewed.Second, the two main interviewers were each invited for a post-interview after the project ended to reflect on their experiences with doing the interviews.These interviews started with pre-selected open questions on their interviewing strategies, and then addressed the two themes (information exchange and atmosphere) specifically.These one-on-one interviews (POST_1 and POST_2) took 45 and 38 minutes respectively.Again, the post-interviews were recorded by means of a voice recorder, information was given about the interview procedure and the two interviewers gave verbal consent.All interviews were transcribed verbatim.

Analysis
To study actions and responses in the ten interviews and the post interviews, three types of analyses were performed: (1) An open analysis of actions of interviewers and interviewees and the interviewees' responses, leading to an overview of all interviews and a structured coding scheme (2) A closed coding analysis, using a structured coding scheme, including quantitative analysis of the frequency of codes and sequential analyses (3) A deductive analysis on experiences with the interviews, from the perspective of interviewees and interviewers.
In Figure 1 a summary of the three types of analyses is given.

Open coding
In the open analysis, the Framework Method (Gale et al., 2013) was adopted, centering on emergent patterns and placing these within a theoretically informed framework.This framework consisted of the two themes "information exchange" and "atmosphere".In the current study we defined Information Exchange by the question type that enabled people with dementia to answer themselves, the level of elaboration, and the relevance of answers to the questions.Atmosphere included actions that supported a good atmosphere such as making jokes and showing empathy with the stories that were shared (Hellstrom et al., 2007;Murphy et al., 2015;Nygård, 2006;Pratt, 2002).To develop the coding scheme, two researchers engaged in a process of immersion in and familiarization with the transcripts and audio recordings.For two of the ten interviews, both researchers independently identified key codes and themes for the interviewees' responses, and all participants' actions.To increase the reliability and complement the framework, all authors were involved throughout the process to discuss these codes.The final coding scheme (Appendix) was then used for a closed coding analysis.

Closed coding
To increase the reliability and validity of the coding process, two new coders were involved of whom one coded all ten interviews, and one coded five.Two meetings were facilitated to increase intercoder agreement.The final coding procedure started with the identification of the question type in the interviews (including length, answer categories, addressee, sequence, nature of questions and content).Then, the person who answered was coded, either the person with dementia or the caregiver after which additional relevant actions were categorized that contributed to or neglected the flow and atmosphere of the interview.After the categorization of the actions, all responses of the persons with dementia were identified as either positive, negative, or neutral.Signs of a positive atmosphere in the current study, for example, entailed positive laughter, cheerful words, and indications that the person with dementia guided the pace of the interview.The coding procedure ended with two codes that indicated a chunk of actions and responses.All coding was performed by means of the qualitative data analysis software, MaxQDA2020.
We conducted sequential analyses to identify temporal patterns in sequences of coded behaviors or stimulus events (Bakeman & Gottman, 1997).Sequences were identified by combining line numbers of coded segments and linking these with either a preceding or a subsequent segment.For example: to identify the questions with given answers, the line number of a question was linked with the line number of the subsequent segment.When this segment included an answer response, then this was linked with the question.When the subsequent segment was not coded with an answer response, the question was regarded as unanswered.Coded segments from MaxQDA2020 were exported to Microsoft Excel 365 to perform these sequential analyses.

Deductive analysis
In the third analysis, two coders deductively analyzed the interviewees' responses to the questions evaluating their experience of the interview.In addition, the two interviews with the interviewers were deductively analyzed as well.

Quantitative overview
The word count of the interviewers and respondents in each of the interviews served as a first quantitative indication of their contribution (Table 2) and with that their potential impact on the design direction for the intelligent gaming system.Based on the word counts, on average, caregivers contributed 41.8% of the words (SD = 20.7) to the entire interview.Persons with dementia (PwD) had an average of 27.2% of the total words (SD = 16.8).Interview 3 was as an outlier as the interview took only 42 minutes and the interviewers had the highest contribution, which contrasted with the other interviews (52,9% + 6,1% = 59% of the words).In general, spouses contributed more to interviews (e.g., interviews 4 and 8) than did the formal carers (e.g., interview 3 or 9).

Sequential analysis
Fifty-seven codes were used to identify the question type, the answer type, other actions by interviewer, carer or person with dementia, the type of responses of the person with dementia and chunks of actions and responses.The coders achieved an interrater agreement of k n = 0,66 (Brennan & Prediger, 1981).This is regarded as a substantial agreement (Landis & Koch, 1977).As the latter chunk-codes (Uncomfortable Interaction and Off-topic) had the highest disagreement, they were left out of further analysis.In this interview two persons with dementia were involved.

Information exchange
To understand how the level of information exchange was influenced by the question type, we analyzed which questions were often answered by the person with dementia and which questions were not.A total of 2506 questions were identified in the 10 interviews, ranging from 168 to 382 questions per interview (1,5 to 6,1 questions per minute per interview).Table 3 shows a quantitative overview of the different questions and the related responses in all interviews.Most questions were directed to the person with dementia (1291), were yes/no-questions (1388), follow-up questions (1056), recall questions (937), contained neutral wording (1261) and were asked by the interviewer (1623).Of all questions 51,5% were directly answered by the person with dementia (in the first segment after the question was raised).When leaving out the one-word questions and the questions specifically addressed to the caregiver or interviewer, this percentage increased to 63,0%.3.2.1.1.Question type.To gain a better understanding of the types of questions that enabled persons with dementia to respond, we identified seven categories of questions.In Table 3, (column "Answered by PwD (%)") we highlighted elements that often elicited an answer.These elements had a positive impact on answerability and included questions that had more than five words, were directed toward the person, were closed-ended with multiple options, included follow-up questions, were reflective in nature, had a positive tone, and were asked by caregivers.Similarly, we highlight questions that were more often not answered by the person with dementia in column "Not answered by PwD (%)" of Table 3.These include questions containing only one word, questions that were caregiver/interviewer directed, open questions, new questions, recall questions and neutral wording.Clearly, the caregiver/interviewer questions were hardly answered by the participants with dementia.Furthermore, we found that questions that introduce new topics tend to restrict the ability to provide answers (37,0% of these questions were not answered by the person with dementia).People with dementia frequently state that they require assistance in responding to these types of "new-topic" questions or that they are unable to provide an answer altogether.
Example question including the aspects: undirected (=no pronoun) and introducing a new topic (=no introduction or structure term on this topic of less joyful days): Interviewer: [Laughs] You can call it like that.Informal carer: [Laughs] Indeed.Interviewer: Yes.And what makes a day less joyful?Informal carer: Well, my husband also has a lung disease and eye problems.So, his physical conditions affect our daily lives.[. . .]To determine which question types led to longer and more informative responses, we analyzed the length and content of the answers provided during the interviews.Our findings are summarized in Figures 2, 3 and 4. Of the 1832 questions, 1154 questions received an answer from the person with dementia; 326 (17,8%) were one-word answers, 353 (19,3%) were short answers (2 to 5 words) and 475 (25,9%) were long answers (more than 5 words).Closed questions with multiple options received most often a long answer (38,0%) while yes/no-questions resulted most often in oneword answers (21,9%).Also rephrased questions and reflective questions seemed more successful in generating longer answers.The nature of the questions and answers were identified as recall, reflective, hypothetical, or rhetorical.Table 4 shows how the nature of the answers given by persons with dementia was connected to the nature of the questions asked by the interviewer.This gave an idea of whether the interviewer allowed the individual to provide answers that matched the nature of the questions.Recall and reflective questions were particularly effective in eliciting matching recall and reflective answers from the people with dementia.

Other related aspects.
Apart from questions and answers, other relevant actions were also analyzed to understand their impact on the flow and atmosphere of the interviews.These actions included the use of attributes, joking, laughter, and supportive comments.A table (Table 5) was created to present the findings.Out of the actions analyzed, four were found to be related to the information exchange during the interviews, as indicated with an asterisk (*).These actions were the use of structure terms or preludes, introducing attributes, staying silent, and the use of difficult wording.The other actions analyzed will be discussed later.
The use of structure terms or preludes, such as referring to an earlier discussed topic during the interview, was identified 92 times in the interviews, of which 39 were in a question.Of these questions, 23 were answered by the person with dementia and 4 were unanswered.Also, the use of attributes was identified in the interviews.As the interviews were conducted in the homes of the participants, the attributes consisted of elements in the room, such as an agenda or a game.In 12 We highlighted the cells that show potential linked nature of questions and answers.Eg 50,4% of the recall questions received a recall answer.To able to compare the interviews, it was decided to present percentages of the actions.E.g. of all the actions coded in interview 1 7,3% was a "Story interviewer"-code.Compared to the total average percentage of "Story interviewer"-code (5,2%), it can be assumed that there were somewhat more stories shared in interview 1 than on average.In this Table the actions are accompanied with * were expected to relate to information exchange.Other actions were expected to relate to atmosphere.
out of 58 instances, attributes were used in a question as a probe, of which seven of these questions were answered by the person with dementia.Out of 58 instances, there were only three where the person with dementia used an attribute in their answer, possibly to add further detail to their verbal accounts.Another action that was coded in the 10 interviews was the use of difficult wording, such as "ipad" or "gps-system."Difficult wording was identified in the interviews only 22 times.11 out of 22 were used in a question, of which 8 of these questions resulted in a non-response by the person with dementia.
Apart from the actions previously mentioned, some actions performed by caregivers also had the potential to contribute to the information exchange during the interviews.These actions included confirming the answer given by the person with dementia, extending the answer, or correcting the answer.Of the 1945 answers given by the person with dementia (including answers not directly given in the first line after a question), the caregiver confirmed an answer 239 times, extended an answer 255 times and corrected an answer 81 times.When the caregiver extended the answer given by the person with dementia, it opened up new avenues for the interviewer to ask follow-up questions or for the person with dementia to provide additional information.

Recovery strategy after breakdown of information exchange.
The analysis of the sequence of questions and actions leading to answers provided valuable insights into what facilitated the exchange of information with people with dementia.Additionally, it was important to examine cases where the exchange of information was disrupted and how the interviewer or caregiver managed to overcome these breakdowns of information exchange.In this analysis, such disruptions were identified when the person with dementia expressed a need for assistance in answering.The person with dementia asked for help in the interviews 77 times after receiving a question.In 30 instances of these 77 situations, a rephrasing question followed.When a rephrasing question was used, it involved using simpler language or providing examples in a closed multiple-choice question format.In 23 other instances, a new question (3 instances), a follow-up question (10 instances), or a confirming question (10 instances) was asked.In 11 instances, the caregiver provided an answer instead of the person with dementia (of which 7 pertained only to the person with dementia).
Example of the interviewer rephrasing a hypothetical question about the possibilities of technology, after a person with dementia mentioned difficulties in understanding the question: Interviewer: For instance, have you ever been bored and wished you had a system that could entertain you?Would you be interested in using such a system?Person with dementia: I have no idea what you are saying.Interviewer: The system could be installed in your home and offer suggestions or invitations for interactions to keep you from feeling bored in the afternoon.Would you like to try it?Person with dementia: Well, in the afternoon, the day is almost over.
3.2.2.Atmosphere.The second topic of interest in our analyses pertained to the atmosphere during the interviews and what type of actions influenced positive or negative responses of the person with dementia.Responses coded as positive included clear positive laughter, cheerful comments, enthusiastic replies, and explicit indications of receiving time to answer a question.Signs of negative responses included, for example, expressions of unease, discomfort or shame and the use of negative wording.A total of 5521 responses of the person with dementia were identified in the 10 interviews.Of these responses, 621 (11,3%) were positive, 528 (9,6%) were negative and the remaining were coded as neutral (79,2%) (Figure 5).In three interviews the percentage of negative responses was somewhat higher than that of positive responses (interviews 3, 4 and 8).
Actions and questions.Actions such as sharing a personal story, providing help, telling jokes, laughing, showing empathy, and understanding, making positive statements, and giving compliments, could impact the atmosphere positively or negatively.In the following paragraphs we elaborate on these actions and the questions that led to positive or negative responses.To study the relationship between these actions and the atmosphere, an analysis was conducted to identify which actions were followed by positive or negative responses.(Figure 6).
More than half of the positive (316 of 621) and of the negative (353 of 528) responses could not be linked with the action-or question coding from the current analysis.Either these did not directly follow one of the actions, as we only analyzed subsequent segments, or other unforeseen actions had an influence on these responses.
Example of action "making a joke"(=he forgets about videos, so he can watch them again) and positive response (=laughing): Informal carer:" [.]He enjoys watching airplane videos and often forgets that he's seen them, allowing him to watch them again after a week.[Laughs]" Of the 621 positive responses, 153 were preceded by action-codes.Actions that were most frequently identified preceding positive responses were: laughing (56 times), responding with positive statements (39 times), and stories shared by the interviewer (32 times).Especially the stories and the positive statements of the interviewer seemed important strategies as these were most often linked with a positive response.Laughing occurred also in many other situations with no clear positive response.In 24,2% of all shared stories by the interviewer, a positive response of the person with dementia followed.For the positive statements this was 9,6%.
Of the 528 negative responses, 66 instances directly followed an action from our analysis.For example, laughing (33 times) and empathizing statements (11 times) were often followed by a negative response.Of these actions, only the empathizing statements appeared to be related more with negative responses in general (7,5% of the total amount of 146 empathizing statements were linked with a negative responses) compared to the other actions (e.g.2,6% of a total of 1253 instances of laughing).
We also studied whether the framing of the raised questions (positive or negative) was related to the responses.In the current interviews, a relatively high percentage (17,2%) of the positive questions were followed by positive responses, while 13,0% of negative questions were followed by negative responses.3.2.1.4.Recovery strategies after negative atmosphere.Through sequential analysis we also examined what occurred after a person with dementia expressed a negative response.Of the 528 negative responses, 131 instances were followed by one of the actions from our analysis, and 81 were followed by a question.In 326 instances, we could not identify a specific action or question from the current analysis that followed a negative response.Interviewers or carer often briefly responded with "Yes" or "OK."The most frequent subsequent actions after a negative response included laughing (55 times), the caregiver confirming the answer of the person with dementia (19 times), the use of structure terms or preludes (21 times) and empathizing statements (13 times).Also asking a question is a relatively common strategy after a negative response; often even a negative question (11,5% of the negatively framed questions were asked after a negative response).
Example of recovery through an empathizing statement (='that is not a good feeling') after negative response (='I feel stupid'): Interviewer 1: [. ..]Do you actively participate in the quiz?
Person with dementia: Well, I usually feel stupid because I don't know any of the answers these days.
Interviewer 1: I am sorry to hear that.It's not a good feeling.
Interviewer 2: I don't know any of the answers on TV-quiz [name] either.
Person with dementia: "Yeah, the questions are rather difficult."

Participant reflections
The analyses above described in more detail what type of actions were related to the response of the person with dementia in either positive or negative ways.To further understand how these dynamics were experienced by participants and interviewers, both were asked to reflect on the interviews directly afterward.The results are discussed in this section with illustrative evidence from the transcripts.

Interview experiences by interviewees
Most people with dementia and their caregivers reported that they did not know what to expect of the interview beforehand.In one of the interviews, the informal carer explained that the person with dementia felt concerned before the interview, as she did not have prior experience with such an interview and was therefore not fully aware what the interview entailed.Interviewees mentioned that they participated because of the trust they felt toward the recruiter (someone from the care organization).
Example of person with dementia explaining the lack of information prior to the interview: Interviewer: What were your first thoughts when you were asked to participate in this interview?
Person with dementia: "I didn't know what to expect and I asked [name of recruiter], but she couldn't tell me much.So, we just waited." According to some of the interviewees, participation was a way to offer help to an acquaintance (one of the recruiters) or to contribute to research.The informal carer in the example below explained that he felt that he could learn something as well from participating.This was also mentioned by two other informal carers.
Example of informal carer explaining additional benefits to him as participating in research: Interviewer: This was the first time you participated in such a research project?
Informal Carer: This type of project, yes.We've received quite a few requests to participate in clinical research in the hospital, though.Our son obtained a PhD so we understand the need for participants in research studies.As former school teachers, we always welcomed interns as it was a great learning opportunity for everyone involved.Participating in research is some sort of crosspollination where I learn from the researchers as well.
When asked afterward how they experienced the interviews, all interviewees reported positively about the experience.One of the participants with dementia specifically mentioned that he would not be able to recall the interview (see example below).
Interviewer: Okay, so what did you think about the interview?
Informal Carer: It was a really nice chat.It was better than expected.
Interviewer: What about you, sir?
Person with dementia: "Yes, yes.I already forgot half of it, but that is a different story."

Interview experiences by interviewers
Overall, both interviewers experienced the interviews as positive, which was also their intention when conducting the interviews in the first place.Interviewers reported that they considered interviews as successful when the atmosphere was good and resulted in -according to the interviewerselaborative sharing of experiences by people with dementia and their caregivers contributing to a rich understanding of this information exchange.Both interviewers mentioned that interview 10 was an example of a less successful interview as the person with dementia was triggered by one of the questions about a negative experience and became emotional during the interview.The interviewers reported that additional distractions during that interview (e.g. a cat walking over the table) made it difficult for them to proceed with their interview set-up.One of the interviewers also mentioned that she experienced interview 3 as more difficult.The interviewer did not feel confident about the extent that the involved person with dementia was able to convey her full story.In the experience of the interviewer, the formal carer who was present was not able to fully support the conversation.Both interviewers reported they experienced difficulties talking about negative experiences or dealing with disagreement among participants.
Example of an interviewer reflecting on their experiences of a negative atmosphere (=frictions between person with dementia and informal carer) and recovery strategy (=asking a new question): Question: "[.] Were there moments where you felt that the interviews were more negative?
Answer: "Yes, there were some moments where the informal carer and the person with dementia had some friction, especially when they disagreed with each other.When this happened, I felt concerned because it could highlight the person with dementia's limitations, which can be uncomfortable for them.To defuse the situation, I would start a new topic or ask a new question to make the conversation more neutral again." One of the interviewers reported the challenges of valuing the different perspectives in the interviews, as not all answers were coming from the person with dementia directly.She explained her strategy of using person-centered questions to check the experience of the person with dementia, after the informal carer answered a question.The other interviewer also suggested that having separate interviews with the person with dementia and with the informal carer should have been considered, because this would make it easier to define which answers reflect whose perspective.
The interviewers reported that distractions or off-topic talking by the interviewees complicated keeping track of the questions of the interview.However, one of the interviewers additionally mentioned that a pause in the interview could also be experienced as improving the atmosphere by lightening up the conversation.Furthermore, this interviewer explained that it helped them to get back to the interview set-up (see example below).
Example of an interviewer reflecting on their experience with taking brakes during interviews (=creating air and getting back to the interview guide): Question: "How did the physical attributes and distractions impact the flow?Answer: "Some people struggled with reflecting on their situation during the interviews, so the distractions sometimes provided a helpful break [.]Although it could be disrupting as well, a physical attribute such as the agenda, was useful for discussing their daily activities [.] Additionally, taking coffee breaks during the interviews allowed us as interviewers to prepare for the next line of questioning." Later in this post-interview, the interviewer mentioned that in hindsight, she would have incorporated more pauses during the interviews.She mentioned that it was difficult to fully attune to the information exchange (getting her questions answered) and atmosphere simultaneously during an interview.

Discussion
In the current study, we analyzed how existing guidelines are put into practice while interviewing people with dementia during the pre-design phase of an interactive intelligent gaming system.The study focused on two main themes: information exchange and atmosphere.The goal was to gain insight into how people with dementia are enabled to exchange information with interviewers and caregivers by a supportive atmosphere, thus positively contributing to their active involvement in technology development research (Hellstrom et al., 2007;Murphy et al., 2015;Nygård, 2006;Pratt, 2002).The following main research questions were considered: 1) What actions contribute to the information exchange between an interviewer and a person with dementia?; 2) What actions contribute to the atmosphere during an interview with a person with dementia?Below we will first discuss the strength and limitations of this study, then we will elaborate on the different contributing actions for information exchange and atmosphere and how this affects the design of intelligent systems.Finally, we will discuss how successfully collected information from users with diverse cognitive abilities can be used in the development of intelligent systems and how our insights further support inclusive development of intelligent systems serving users with diverse cognitive abilities.
One of the strengths of this study is the line-to-line sequential in-depth analyses on transcripts of a representative set of 10 conducted interviews.Existing studies on how to construct a meaningful interview with people living with dementia were mainly based on post-hoc reflections of researchers and participants.Our unique analyses identified positive and negative dynamics in information exchange and atmosphere related to actions and recovery strategies of the interviewer and caregiver.Findings were enriched by additional qualitative data of experiences of participants (people with dementia and caregivers) and interviewers to understand their perspectives of these interview dynamics.This resulted in a broad and novel perspective on how these interviews were conducted and experienced.
No study comes without limitations.The data used in this study came from one design phase and from one type of intelligent system development project only.Nevertheless, the current detailed indepth analyses were based on a relatively large set of data (almost 15 interview hours).Future research could enrich our insights by including similar projects with other interviewers and participants.In addition, the participants in the interviews (both people with dementia as well as the caregivers) were obviously not simple information providers who idly comply to interact with the interviewers (Clark, 2009).The results should therefore be acknowledged as an attempt to further understand these interactions, which inevitably contain multiple factors beyond what is analyzed in the current study.In-depth analysis of video recordings (e.g.together with people living with dementia (Williams et al., 2020) or detailed observation methods during interaction with early prototypes of the intelligent system, could have enriched the current insights.However, the unique approach of the current study contributes to existing methodological guidelines with an actual indepth analysis of interview data compared to previous analyses based on retrospective experiences of researchers in the field.This brings additional evidence for some of these guidelines and new insights to or further specification of these guidelines.We discuss these results and implications below.

Carefully introducing new topics
The number of direct answers given to questions in our analyses show the willingness of the participants with dementia to share experiences which informed early design directions for the intelligent system, such as important themes for the people with dementia in their quality of life.Closed multiple-choice questions, reflective questions and questions containing positive wording seemed to encourage people with dementia to share.Earlier research highlighted the importance of single-faceted interview questions (Cridland et al., 2016).The higher success of reflective questions compared to other question types in our analyses supports the claim that people with dementia are often able to share their opinions when asked appropriately (Ablitt et al., 2009).Additionally, reflective questions in the interviews were more likely to be follow-up questions and did not introduce new topics per se.Our research suggests that when conducting interviews with people who have dementia, it is crucial to allocate sufficient time to introduce new topics.This is particularly important when introducing novel concepts, such as new forms of user interfaces, to avoid any confusion or misunderstandings resulting from the order of questions.It is essential to ensure that people with dementia can effectively communicate their thoughts and ideas to designers.In the example of new user interface, this can be critical in improving the accessibility of the system through interface redesigns or addressing underlying system issues.From the reflections of the interviewers in the current study we learned that not only for the person with dementia, but also for (novice) interviewers themselves, slowing down or using pauses can help to create a more fruitful interview for information exchange.The deliberate use of structure terms, or preludes (Cridland et al., 2016), can be a helpful strategy to guide people through the different interview topics as well.

Guiding caregivers toward a supporting role
The contribution of informal carers was higher than that of formal carers.As they have their own experiences with living with dementia, it is expected that the informal carers too were eager to share their views and opinions on how systems could be supportive and accessible.The perspectives of carers can contribute to understanding the lives of people living with dementia and they can enrich the information shared by them (Dai & Mofatt, 2021).The results in the current study indeed show that caregivers can support the interviewer, by for example rephrasing questions, which were then often answered by the person with dementia.Nevertheless, interviewers from our study also mentioned that the presence of caregivers could diffuse or suppress the answers given by the people with dementia, as carers sometimes answered instead of or corrected them.This was also reported in earlier work (Pesonen et al., 2011) and in similar research in technology development (Lindsay et al., 2012).This can result in intelligent systems that are more attuned to the needs of carers instead of the needs of people with dementia themselves, which can be substantially different.Consequently, this could affect the prioritization of functionalities in or personalization of intelligent systems.From our study, person-centered questions seemed effective in balancing the contribution to the interview when informal carers overshadowed participants' voices.Dai and Mofatt (2021) suggest assessing the power dynamics among participates and carers prior to the research activity and develop concrete recovery strategies.If possible, the type of carer might be something to reconsider in future research, as the current study illustrated that children or formal carers seemed to be less inclined to interfere in the information exchange between researcher and people with dementia.

Using probes to involve people in abstract thinking
Exchanging information around hypothetical situations, such as how personalization, flexibility or proactivity of the system would be of value for the person with dementia, presented challenges for both interviewers and persons with dementia.As these types of questions are common in pre-design phases of technology development, this calls for further exploration on whether people with dementia can engage in hypothetical questions and with what kind of tools and materials this can be supported.This exploration is especially relevant to involve people with diverse cognitive abilities in co-designing increasingly complex intelligent systems as older adults in general show low algorithmic awareness (Gran et al., 2021).In the current study, the application of physical attributes was still rather limited and not always deliberately used by the interviewers.The video, for example, that was shown to participants to envision how the interactive intelligent gaming system could be designed, did not evoke much recognition or motivation to reflect on possible solutions.This indicates that these prompting tools should be piloted first and designed in such a way that they evoke a certain level of interaction.The use of materials such as design artifacts might have allowed to understand participants in different ways (Branco et al., 2017;Morrissey et al., 2016;Wallace et al., 2013).Also, various forms of technological probes and prototypes (Houben et al., 2022;Thoolen et al., 2019), can lead to richer and contextualized insights in the needs and wishes of people with dementia on user experience, functionalities, performance and accessibility.This is in line with existing guidelines on interviewing with tangible prompts to enable people with dementia in elucidating their thoughts (Briggs et al., 2003;Nygård, 2006).In addition, design artifacts or probes could enhance a more collaborative atmosphere between people with dementia and their carers during a research activity (Suijkerbuijk et al., 2015).

Personalizing the length of the interview
The interviews analyzed in the current study took on average almost an hour and a half.Within this duration of the interviews, interviewers asked hundreds of questions per interview.Lengthier interviews may be regarded as a strategy to enhance the rigor of qualitative studies (Rubin, 2000).However, this seems to conflict with general recommendations to involve people with dementia in shorter interactions (Nygård, 2006) or equally value short-during interviews of only a few minutes (Cridland et al., 2016;Gebhard & Mir, 2021).It seems quite challenging to find the right balance in exchanging sufficient information with people with dementia for further design phases (e.g., requirement elicitation, refinement, validation), breaking interviews in multiple meetings and still not overburden these participants and their caregivers.Future endeavors in this field should highlight how flexibility could be adopted to find a suitable way for different participants.Hence, flexibility is not only a crucial characteristic of intelligent system design but also a key factor in the process of gaining insights for such designs.The word counts per participant in the current study provided an indication of the variety in contribution among the different persons with dementia.This calls for personalized participation (Branco et al., 2017) that respects different needs in how long and how often people with dementia can and wish to contribute to these types of interviews.

Designing an interview as a fulfilling experience
Interviewers reported that their intention was to focus on positive aspects in life and enable the person with dementia to talk about their strengths and wishes, as in line with recommendations of Vernooij-Dassen (Vernooij-Dassen et al., 2011).Some actions, such as sharing stories or using positive statements, were often linked with positive responses of the people with dementia.This suggests that these actions contribute to a positive atmosphere and support the researcherparticipant relationship (Hendriks et al., 2018) needed in order to keep people engaged for further phases in the development of intelligent systems.Earlier work reflects on the potential benefits for people with dementia to participate in interviews, such as the opportunity to talk about and make sense of experiences and give a sense of purpose and altruism (Pesonen et al., 2011).Next to these larger effects, the level of enjoyment as indicated by the number of jokes and frequency of laughter in the current analyses, show that interviews can also be pleasurable experience as they happen.A study on the appreciation of humor and laughing as a naturally occurring phenomenon in a design process with teenagers shows that it can encourage an open, relaxed atmosphere (Iivari et al., 2020).When interviewees feel comfortable and at ease during the interview process, they are more likely to speak openly and honestly about their needs and preferences.This can help designers gain a better understanding of user needs and design more effective and intuitive interactive systems.Future research could further define how humor and other ingredients for a relaxed atmosphere can be of specific value in research activities, such as interviews, with people living with dementia in the design process of intelligent systems.

Preparing for a respectful response to negative experiences
From the current study, it seems inevitable that interviewers unintentionally touch upon the lost abilities of the person with dementia and therefore sometimes can induce negative responses.Balancing positively and negatively framed questions is an important consideration to encourage open and honest discussion on difficult topics (Cridland et al., 2016).Interestingly, from the current analyses, we found that negative responses were related to empathizing statements.Possibly, the interviewers gave room to a person with dementia to ponder on negative experiences.Similarly, interviewers or carers asked a question with negative wording after a negative response of the person with dementia.This could be regarded as a positive action to respectfully give attention to a negative response.When participants in interviews feel understood and validated, this can possibly promote empathy on the part of the designers, helping them to better understand the needs and experiences of their users and create more meaningful and relevant interactive systems.It was natural to the interviewers in the current study to empathize with their participants, which is also claimed to be a necessary skill in doing qualitative research (Brinkmann, 2007).Rigid protocols on how to implement these kinds of actions do not seem appropriate (Hendriks et al., 2020) but the expertise and reflections of experienced researchers in the field could be helpful to understand how they adapt their strategies and how this affects the involvement of people with dementia.

Adapting to different ways of consent expressions of participants
One of the informal carers in the current study reported that the person with dementia experienced some level of anxiety before being interviewed.Another participant specifically mentioned that he did not recall much about the interview immediately, thereafter, indicating that it could be difficult to reflect on the experience.This highlights the importance of paying sufficient attention to written, verbal and graphical informed consent and the interview dynamics as they happen.Continuous consent has therefore been proposed as a necessary strategy (Dewing, 2002).Repeatedly checking with and ensuring that the person with dementia feels positive about their contribution should be practiced, for instance by deliberately spending time with the person after the interview (Hellstrom et al., 2007).Other researchers, however, reported that participants with dementia could also experience this continuously checking as patronizing (Span et al., 2017).This, again, calls for personalized approaches in the research practice with people with dementia (Suijkerbuijk et al., 2022).When specifically asked afterward about their experience in the interviews used in the current study, both interviewers and interviewees reflected positively about it.The current study shows that analyzing the responses of the interviewees can be an additional triangulating tool to strengthen the reliability and to better understand their experiences beyond, perhaps socially acceptable, answers when explicitly asking them.The insights mentioned above could be applicable to general ethical considerations surrounding the use of intelligent systems among individuals with dementia, including issues related to informed consent, autonomy, privacy, data security, and affordability (Wangmo et al., 2019;Zwijsen et al., 2011).

Future work
The qualitative outcomes of the interviews in our "MagicTable" project included nine directions for the development of the gaming platform: going out, being physically engaged, having time for myself, not having too many commitments on a day, having real contact, being among people, reliving the past, having equivalence in decision-making and being part of the outside world (Nap et al., 2019).Needs assessments by means of interviews with larger populations (Miranda-Castillo et al., 2013;Reilly et al., 2020) show similar outcomes.These directions spark inspiration for developing intelligent gaming platforms for people with dementia for training memory and promoting inter-generational bonding (Robert et al., 2014) or safe ways of exercising and gaming (Van Santen et al., 2018).However, if not carefully analyzed how these interviews were constructed, as shown in the current study, there is a risk for developers of intelligent systems to focus on directions that might not be the most important to the people with dementia themselves.Close collaboration with people with dementia might unravel other directions, such as the physical design of systems, having shared experiences, include additional sensory experiences, personalizing the old and the new and not excluding the participant with dementia by means of technology (Hodge et al., 2018).Moreover, to engage people effectively in subsequent development phases (such as the generative phase, evaluation phase, and post-design phase), it is crucial to ensure they feel completely heard and understood in the first interactions during the predesign phase.In future research, when utilizing interviews to gather information from individuals with diverse cognitive abilities, it is essential to establish an inclusive research framework by incorporating existing guidelines (Cridland et al., 2016;Hendriks et al., 2013) and integrating the additional insights obtained from this study.Additionally, researchers should thoroughly analyze the interview data to confirm the qualitative outcomes directly attributed to individuals with diverse cognitive abilities.To ensure that supportive intelligent systems meet the needs of people with dementia, it is imperative to iteratively involve them throughout design processes.Especially since current data-and understanding challenges for intelligent systems for people with diverse cognitive abilities (Lewis, 2020) still request some level of interaction of the person through the various voice-or touch interfaces of such systems.This can be beneficial as supervised learning loops, for example, can also foster users' understanding and transparency (Takeda et al., 2019).

Conclusions
To develop personalized and adaptive intelligent systems within this field, respecting actual user values, we need a thorough understanding of how we can create inclusive research to involve people with all sorts of cognitive abilities.Not only to create meaningful and responsible systems, but also to empower people living with dementia to voice their perspective (Williams et al., 2020), as our research shows that they can indeed verbalize their ideas early in the development process.In addition, co-designing intelligent systems with people with dementia contributes to the muchneeded change in the societal views on aging and cognitive diversity (Lazar et al., 2017).
Interviews seem to be a useful method to invite people with dementia to share their experiences.Being involved in research, however, has an impact on participants as they actively negotiate, manage and experience their involvement (Clark, 2009).Hence, attention is needed from researchers to create the best possible experience, both in enabling them in sharing information and in supporting a positive experience of participating in research.Information exchange in interviews with people with dementia can be supported by carefully constructed questions; intentionally guiding carers to an amplifying role; the use of appropriate probes; and adapting the length of the interview to the preference of the participants.To create a pleasant atmosphere during the interviews, researchers should design the interview as a fulfilling experience, be prepared to respond respectfully to negative responses and be able to adapt to different ways of consent expressions.We think that our analyses of both successful and less successful interview strategies contribute to the actual understanding of what happens when interviewing people with dementia in the pre-design phase of technology development.This might not only be relevant for the research method of involving people with diverse cognitive abilities in co-design, but also spark some inspiration for algorithms supporting personalized interaction with intelligent systems and their different users.We invite other researchers to integrate the insights from our work in their interview guides, practice flexibility to adhere to personal differences and share their experiences in successfully involving people with dementia in the design process of supportive technologies.
Mirella Minkman is a distinguished Professor of 'Innovation of organisation and governance of integrated care' and the Chief Executive Officer of Vilans.She has extensive experience in the field of health care organisation, integrated care, collaboration and networks and the innovation of governance.Except for interview 3.Only include Respondent 3.

Yvonne de
• 3.2 When an answer is given by the (in)formal carer: 3 When no answer is given: Another response to a question (not an answer) Person with dementia The person with dementia answers a question with "I don't know"; which can be either a lack of understanding or just a valid answer.
Would you like having such a system in your home?I don't know.

PwD does not know
Person with dementia Checks with (in)formal carer if their answer is correct or specifically asks for support in answering

Figure 1 .
Figure 1.Overview of the three different analyses.

Figure 2 .
Figure 2. Information exchange -Different answer categories and length of answers.

Figure 3 .
Figure 3. Information exchange -Different sequence of questions and length of answers.

Figure 4 .
Figure 4. Information exchange -Different nature of questions and questions and length of answers.

Figure 5 .
Figure 5. Percentages of positive, negative, and neutral responses of the person with dementia per interview.

Figure 6 .
Figure 6.Action types, question types and related responses.

Table 1 .
Information about participants.

Table 2 .
Overview of contribution to the interview and in answering the questions including time in minutes and word count in percentages.

you like the cognitive training aspect, or is it the gymnastics that you like the most
Example question including the aspects: person-directed (='you'), closed question with multiple options (=cognitive training or gymnastics), reflective (=reflection on an activity) and positive wording (='like'): [in the activity at the care organization]?Person with dementia:No, it is the physical training ánd the cognitive exercises.

Table 3 .
Quantitative overview of information exchange -seven categories of questions and answers 1 .
1Total numbers per question type are given.The percentages relate to the type of question (e.g., 41,1% of the one-word questions were answered by the PwD).The highest percentages per column are outlined per category of questions.*One-word questions were not further coded.**Caregiver/Interviewer directed questions were not further coded.

Table 4 .
Information exchange -nature of questions and nature of answers 1 .

Table 5 .
Percentages of the actions per interview 1 .
Kort is a full professor of environmental psychology in the Human-Technology Interaction group and vice dean of the Department of Industrial Engineering and Innovation Sciences at the Eindhoven University of Technology, The Netherlands.She obtained her Ph.D. in Environmental Psychology from TU/e.Yvonne manages the Mental Health program in TU/e's Center for Humans and Technology.
1Except for interview 2. There are 2 people with dementia (Respondent 1 and 2).Respondent 3 is informal carer. 2 )formal carer extends an answer given by the person with dementia.The interviewer is prompted with a new topic.
HUMAN-COMPUTER INTERACTION