Summary

The UNMASKED project aimed to develop and apply a synergy of devised theatrical performances and scientific methods rooted in computer science (human-computer interaction, cybersecurity, AI) and psychology to provide a novel approach to tackle the issue of inauthenticity in cyberspace.

The main idea underlying the UNMASKED project is that we can develop theatre as a transdisciplinary and socio-technical research space – a lab (The Theatre of Inauthenticity Lab) – and use this as a mechanism to identify the issues related to inauthenticity, unmask (or expose) them, elicit an appreciation of inauthenticity, understand how inauthenticity is disseminated, and combat inauthenticity when used to harm, deceive and create conflict or tension. The Lab will also provide a means to reason about how to create and consume reliable inauthenticity.

To that end, we aimed to develop and stage a devised theatre performance, collaborating with theatre-makers (with expertise in devised theatre) to engage audiences and have a platform to explore inauthenticity in cyberspace.

Objectives

The main objective of the UNMASKED project was to investigate whether and how the fleeting inauthenticity of theatrical performances can help mitigate the problem of persistent online inauthenticity. To that end, UNMASKED aimed to tackle the following overall objectives:

1. Provide evidence that theatre can be used as an intervention in the context of inauthenticity.

2. Provide an initial set of methods and resources that can be shared and built upon, by the UNMASKED team and by other researchers.

3. Engage with different stakeholders to establish a transdisciplinary network to build the path to a larger project.

The first two objectives were achieved, as evidenced by the discussions below and by the papers that we are currently writing up, along with a follow-up project proposal. The only change that we made concerns objective 3, since it became evident early on, during the initial meetings of the project, that it would not be feasible to involve different stakeholders in the devising activity, which we instead carried out in collaboration with the theatre-makers only. We thus decided to involve different stakeholders from academia and industry at a later stage.

More specifically, we decided to structure the project in two phases: phase one consisted of performing the devised piece to the “general public” and evaluating its impact and potential, which is what we discuss in this report; phase two is currently underway, even after the official end of the project, with the aim to engage with experts in related fields to reflect on our approach and practice, get their feedback on the results of phase one, and inform the writing of the follow-up proposal.

Activities

The main activity of our project work was the devising and performance of the play “UNMASKED” in front of a lay audience. From November 2024 to January 2025, the coinvestigators met about a dozen times to prepare the devising, discussing different definitions of inauthenticity as well as different scenarios to be considered in the play. We had also three preparatory meetings and two workshops with the theatre-makers to brief them on our vision and objectives, and to prepare the devising by brainstorming on different possible scenarios.

The 5 co-investigators and the 4 theatre-makers devised the play in the workshop spaces of the Science Gallery London from the morning of February 25 to midday February 28, and the performance took place in the auditorium of the Science Gallery London in the afternoon of February 28.

The audience comprised of 66 members of the general public, who had replied to a call for participants issued through various channels, including standard channels of King’s College London and Science Gallery London, as well as channels of various organisations that advertise free theatrical events. It is interesting to note that 74 people had registered their interest and signed the consent form and only 8 withdrew their participation or did not show up on the day of the performance.

Participants were asked take part into a quantitative study and a qualitative study. They were asked to fill out a questionnaire twice, before and after the performance to assess how the performance had changed their views. 65 valid questionnaires were returned. Participants were then split into 3 groups to participate in focus group discussions. A summary of the results of these two studies is given at the end of this section.

A few weeks after the performance, we held a debrief meeting with the theatre-makers to collect their feedback on what worked in the process and what could be improved in the future. This feedback will inform the methodology that we will describe in the publications that we are preparing and that we will employ in the follow-up project proposal.

In the course of the summer, we will carry out a second study with expert stakeholders working in cybersecurity and AI. We will show them a recording of the performance and gauge their views on how our approach could be used extensively to address online inauthenticity and other issues related to cybersecurity and AI.

QUANTITATIVE STUDY

Overall, 65 participants took place in this study by completing a paper-based questionnaire twice (before and after the performance). The questionnaire was split into two parts. Participants were asked to fill out the first part only once, before the performance. Thisfirst part included a question on the frequency which they typically go to the theatre to watch a play (cf. Table 1) and 4 questions about the participants’ characteristics, such as gender, age, education level and ethnicity. Table 2 shows the gender distribution of the audience members.

The minimum age of the participants was 18 years old, while the maximum age was 74, with a mean of 37.1 and standard deviation 16.56. 49.2% of the participants were 31 years old and lower (born after 1994), and the rest were over 31 years old (born earlier than 1994). We have chosen to consider 1994 as it was a pivotal year for the Internet, marking the beginning of its widespread adoption and commercialization. The majority of the participants had an undergraduate degree (46.2%), followed by those who had a master’s degree (32.3%).

Table 1. The participant's frequency of going to the theatre to watch a play:

Table 2. Descriptive Statistics for participant gender:

The next questions asked participants about their AI expertise: all the participants heard about AI, with the majority (67.7% of them) being able to explain it somehow, while 21.5% considered themselves experts. Participants were then asked to use a 7-point Likert scale to rate the influence and the confidence in the reliability of the information provided that influenced their understanding of AI. They were presented 8 different potential sources, such as friends and family, news (e.g. BBC News, Newspapers), celebrities/social media influencers (e.g. people with over 30k followers), politicians and politically connected people, Experts/ Scientists/ Academics, theatre, or cinema.

Participants were asked to fill out the second part of the questionnaire twice, before and after the performance. This part included the validated self-authenticity scale developed by Cartwright et al. (Measuring authentic living from internal and external perspectives: A novel measure of self-authenticity. Social Sciences & Humanities Open, 8(1), 100698, 2023), and an emotional scale inspired by the validated scale designed by Watson et al. (Development and Validation of Brief Measures of Positive and Negative Affect: The PANAS Scales. Journal of Personality and Social Psychology, 54, 1063-1070, 1988).

More specifically, the self-authenticity scale included two sub-scales related to internal self-authenticity and external self-authenticity. Overall, the scale included 15 items in a 5-point Likert scale (1: Completely Disagree to 5: Completely Agree). To measure the internal self-authenticity, 8 items were used, including statements such as “I know how I want to live my life” or “I know what is important to me and what is not”. To measure the external self-authenticity, 7 items were used, including statements such as “Society stops me from being who I want to be” or “Other people greatly influence my actions”. The internal reliability consistency (i.e. Cronbach's Alpha) values based on the participants’ study before and after the intervention are presented in Table 3.

Table 3. Cronbach’s Alpha:

Results

Several research questions were explored in this study to investigate whether the performance impacted people’s emotional experience and self-authenticity, considering their age and their familiarity with AI and theatre. Specifically, the following are the four main research questions that we explored.

Research Question 1 (RQ1): Is there any significant difference between the emotional experience before and after the performance, considering the individuals’ age and their familiarity with AI and theatre?

A two-way MANOVA statistical analysis was conducted to test whether there was a significant difference in emotional experience before and after the performance, and address RQ1. Overall, there was no statistically significant interaction effect between age groups, theatre experience and AI familiarity and type of intervention on the combined dependent variables of emotional experience, F(4, 84) = 0.427, p = .789; n2=0.020, Wilks' Λ =0.961. Also, there was no significant difference between any combination of dependent variables: age groups (p=0.470), theatre experience (p=0.562), AI familiarity (p=0.719), age group and theatre experience (p=0.528), age group and AI familiarity (p=0.916), theatre experience and AI familiarity (p=0.789).

Research Question 2 (RQ2): Is there any significant difference between the internal and external self-authenticity before and after the performance, considering the individuals’ age and their familiarity with AI and theatre?

A two-way MANOVA statistical analysis was conducted to test whether there was a significant difference in internal self-authenticity before and after the performance. Overall, there was no statistically significant interaction effect between age groups, theatre experience and AI familiarity and type of intervention on the combined dependent variables of emotional experience, F(4, 92) = 0.491, p = .742; n2=0.021, Wilks' Λ =0.959. Also, there was no significant difference between any combination of dependent variables: age groups (p=0.757), theatre experience (p=0.486), AI familiarity (p=0.465), age group and theatre experience (p=0.234), age group and AI familiarity (p=0.426), theatre experience and AI familiarity (p=0.607).

A two-way MANOVA statistical analysis was conducted to test whether there was a significant difference in external self-authenticity before and after the performance. Overall, there was no statistically significant interaction effect between age groups, theatre experience and AI familiarity and type of intervention on the combined dependent variables of emotional experience, F(4, 90) = 0.360, p = .836; n2=0.016, Wilks' Λ = 0.969. There was no significant difference between any combination of dependent variables: age groups (p=0.722), theatre experience (p=0.286), age group and theatre experience (p=0.062), age group and AI familiarity (p=0.164), theatre experience and AI familiarity (p=0.873). However, only AI familiarity has a significant difference in external self-authenticity, indicating that individuals have not changed their views on external authenticity based on the theatre performance and the other people’s reactions to the performance, F(4, 116) = 2.749, p = .032; n2=0.087, Wilks' Λ = 0.834.

Table 4. Descriptive statistics for external self-authenticity before and after the theatre performance:

External Self-Authenticity before the theatre performance:

External Self-Authenticity after the theatre performance:

Research Question 3 (RQ3): Is there any significant difference between (a) influence and (b) confidence, considering the individuals’ age and their familiarity with AI and theatre?

Two separate multiple regression analyses were conducted to explore RQ3. For influence (M= 27.2, SD= 6.94), the regression model does not significantly predict 10.5% of the variance, Adjusted R2 = 10.50, F(3, 61) = 2.271, p = 0.090. There was a significant negative association with age group (b = -4.038, p =0.033) and a significant positive association with theatre experience (b = 1.565, p = 0.042). However, there was no significant association with AI familiarity (b =-0.595, p = 0.702).

For confidence (M= 27.1, SD= 8.34), the regression model does not significantly predict 1.5% of the variance, Adjusted R2 = 1.5, F(3, 59) = 0.275, p = 0.843. There was no significant association with age group (b = -1.847, p = 0.440), theatre experience (b = 0.701, p = 0.470) and AI familiarity (b = -0.020, p = 0.992).

Research Question 4 (RQ4): Is there a difference between the level of theatre experience with the emotional status before and after the theatre performance?

A one-way ANOVA was conducted to explore RQ4. Overall, there was no statistically significant difference between the level of the theatre experience with the emotional status, F(2, 58) = 0.810, p = .521; n2=0.029, Wilks' Λ = 0.943.

Overall findings

The quantitative analysis revealed a difference in AI familiarity between the external self-authenticity scores before and after the performance. There was a significant negative association with age group and a significant positive association with theatre experience.

QUALITATIVE STUDY

Data collection: After the performance, the audience (N=66) was invited to attend roundtable discussions about the performance. A large majority of the audience (about 60) opted to participate and were guided to separate rooms where discussions took place. Data was collected from three tables, in a roundtable discussion style, where a researcher moderated the discussion: each table consisted of about 20 participants from the audience plus a moderator. Discussions were guided by a semi-structured interview protocol, which we had prepared in advance. The discussions were audio recorded for internal purposes.

Data analysis: Audio recordings from three roundtable discussions were transcribed using Whisper AI, an automatic transcription tool installed locally on a secure institutional laptop so that at no stage data left the local machine.

Two researchers independently analysed the recordings and the transcripts. This process consisted of listening to the recordings and using the transcripts to help with the analysis. Each researcher started by familiarising themselves with the content, making notes of interesting points of discussion, summarising these points and labelling them. The two researchers then met and discussed the process in detail. Additionally, they discussed the findings regarding one of the recordings to ensure mutual understanding and check the implementation of the process and findings. The two researchers iterated over the three recordings independently. This resulted in two documents: one per researcher, with themes, snippets from the transcript related to each theme, and a researchers’ summary of each snippet. The following is a preliminary summary of the analysis.

One researcher then used the content analysis tool QCAMAP to go over these anonymised documents. Where necessary, themes were split into more than one, or multiple themes were merged, labels were amended to capture the changes. The final codebook (themes, subthemes, their definition, and example quotes) were discussed with the second researcher in a meeting. Disagreements were resolved via discussion and adjustments were made to the codebook. Subsequently, the codebook was shared with the rest of the team (consisting of the three other co-investigators). In a subsequent meeting, all team members provided input and the codebook was amended to capture the agreed upon codebook.

Overview of findings: We are currently performing a detailed analysis and will publish the final codebook in an upcoming peer-reviewed submission. The following are some interesting insights that emerged from the roundtable discussions:

• Themes that were presented in the play provoked debate on human issues and experiences related to social media and the influence of AL.

• The play showed the main protagonist adopting assumed identities in social media groups. Some of the participants interpreted this negatively, as hiding from reality, with some suggesting that this behaviour is inherently inauthentic. Counter-opinions referred to the use of assumed ID as potentially liberating, allowing for authentic expression that may be suppressed in day-to-day life.

• A further debate was provoked by the main character’s use of an AI therapist. Reactions to this were largely negative. Participants cited several concerns. These included that absence of lived experience and the danger of logic-based responses without emotional intelligence. Some participants sourced news stories involving people influenced by AI to self-harm as evidence supporting this view. Others also pointed to AI’s limited range of sensory capabilities.

• Several of the scenes provoked concerns about mental health implications. The main character’s obsessive consumption of online news was linked to diminished mental health with several participants describing concerns about becoming emotionally overwhelmed. Some participants also raised concerns about possible negative mental health effects of adopting different personas in online worlds.

• There was also a debate emanating from a scene in which characters converse through text messages with heavy use of emojis. An emerging view was that the simplicity and immediacy of the media, particularly the use of emojis, subtracted authenticity. Some participants saw the use of emojis as feigned empathy, allowing the sender to avoid a commitment to meaningful discourse.

Outputs

The main output of the project is the performance, which took place on February 28 at the Science Gallery London, with an audience comprising 66 members of the general public (who responded to our calls for participation).

Two other outputs are already available:

• Script of the performance

• Recording of the performance

These are, however, not publicly accessible. We will include the script in our publications (see below), but the recording will only be for internal use (e.g. for the second study that we will carry out in the summer) although we might publish some short clips or screenshots.

The following additional outputs are in preparation:

• A paper on the user studies that we carried out (both quantitative and qualitative), likely to be submitted to the CHI conference in the course of the summer.

• A paper on the methodology that we followed, discussing our whole approach and the co-creation process and what worked and not, and discussing the resulting script and which scenes we picked for development and which not (and how they worked). This will likely be submitted to an inter-disciplinary journal.

• Blog posts on the co-creation experience. We had originally planned to publish blog posts in the course of the project but we decided to postpone their publication to after having analysed the results of the user studies and submitted the papers so not to compromise their acceptance. We will publish a first, “neutral”, blog post in July.

• An article highlighting the innovative use of the theatre in Research and Development. This will likely be submitted to the New Scientist or the Conversation.

We are also working on a follow-up project proposal based on the positive experience of this pilot project. We will report on the future outputs, in particular the publications, as soon as they are available.

Impact

The actual impact is exemplified by the user studies that we carried out based on the performance we devised, which do indeed provide evidence that theatre can be used as an intervention in the context of inauthenticity (objective 1). They are a very good instrument to open up a discussion with the general public, help them to understand, identify and react to inauthentic online content, and thereby inform them about the dangers but also the opportunities that the Internet and AI offer.

The work that we have carried out (and the publications that we are currently writing up) provides evidence for objective 2, as we have indeed developed an initial set of methods and resources that can be shared and built upon, by the UNMASKED team and by other researchers.

Objective 3, namely engage with different stakeholders to establish a transdisciplinary network to build the path to a larger project, will be tackled by the additional study that we have planned for the next weeks, as described below.

The potential impact that we will explore in the future is that of theatre and performance as a means to involve different stakeholders in a discourse about technology.

Future work

In addition to the outputs that we are working on, and the follow-up proposal, we are planning to carry out a second user study involving stakeholders from academia and industry expert in cybersecurity and AI.

This will allow us to get feedback from experts on how to position our work but also to extend it from online inauthenticity to other cybersecurity and AI issues. It will thus allow us to achieve our initial objectives in full by providing an initial set of methods and resources that can be shared and built upon, by the UNMASKED team and by other researchers, and by engaging with different stakeholders to establish a transdisciplinary network to build the path to a larger project, in which we plan to involve also experts in performance studies and narratology, as well as policy and lawmakers to promote further discussion in and around inauthentic content generation and consumption.