Ethical Challenges for Human–Agent Interaction in Virtual Collaboration at Work

Figures & data

Table 1. Description of the interview partners.

ID	Professional position or expertise	Country	Expert in
			Computer science	Collaboration	Ethics
E1	Associate Professor in information systems and researcher for communication technologies for team collaboration	USA	X	X
E2	Researcher in the field of digitalization in companies and virtual teams	Germany		X
E3	Professor in the field of innovation in society	Australia	X
E4	Professor for business information systems	Germany	X
E5	Researcher in human–computer interaction and computer science	Norway	X
E6	Professor in the fields of management and technology	Austria	X	X
E7	Professor for business information systems	Germany	X
E8	Professor for information science and information ethics	Germany	X		X
E9	Professor in the field of philosophy, business ethics and digitalization	Germany	X		X
E10	Researcher at the intersection of artificial intelligence and human behavior	Germany	X	X
E11	Professor for digital transformation and business information systems	Germany		X
E12	Associate Professor in the fields of media and information	USA	X
E13	Professor in the fields of leadership and organization	Germany		X
E14	Professor for the field of software development and collaboration engineering	USA	X	X
E15	Researcher in the field of AI ethics and Information and media ethics	Germany	X		X

Table 2. Key themes and illustrative passages from the interviews.

Ethical principle	Themes	Illustrative passages from the interviews
Beneficence	Social risks	“I can see that that would happen if people work with AI instead of with other humans. So, if I have another counterpart that I work with all the time, but that counterpart is replaced by an AI (…).” (E14)
	Prevent social risks	“No, I don’t, because where there is technology to be abused, it will be abused.” (E3)
Justice, solidarity, and workspace awareness	Threats to solidarity	“So, if performance goals remain team goals, then it’s much more likely that they’re also will develop a level of team coherence and identification with the team versus if your objectives are purely individual, then you stop caring more about the team. So that is a design choice that organizational designers have to make.” (E14)
	Information deficits	"But we know that the functions, i.e., the functional component of cooperation, is always strengthened where the personal, the social component is also strong, and there is of course a clear connection. The more distributed, the more difficult." (E13)
	Workspace awareness	"So, I don’t think we should put up more technical barriers between people." (E2)
Non-maleficence and privacy	Types of abuse	"So I think there’s, as with all software and as with all data that we collect, there’s an infinite range of possibilities for mischief." (E4)
Explicability	Burdens	“So, I think transparency and performance evaluation are the biggest fears or areas of discomfort that people will experience.” (E14)
Accountability/autonomy	Responsibility	"When systems make decisions, then those who have made the decision to use these systems are responsible. So, I don’t think that it is, except in the case of blatant programming errors or deliberate programming errors, but I think it is not the one who programs, because he is not responsible for a system being used somewhere in the way it is used. It’s the person who makes the decision to use it." (E7)
	Avoid dependencies	“So explainability in AI will be important to avoid dependency.” (E5) "I think it’s a bit of give and take. So, you take away a bit of autonomy when it comes to performing a certain task, but then you get additional autonomy again through another task because you can now make a better decision or because you now have more information, you have better security, things like that." (E6) "And I don’t know if it should be avoided, quite honestly, this dependence of technology. Or if it’s just the next step in what we’re doing anyway (…) how we’re just evolving." (E6) "I think technology, and especially technology that works, always makes us dependent in a certain way." (E9) "This complaining about the loss of true and correct and good and canonical knowledge is as old as humanity itself. So, I think that’s pointless. I think we have to look at what exactly is being lost and then consider: Is that something that affects our autonomy in a very special way? And then we have to see whether we can perhaps substitute it in some other way, or possibly find another solution somehow.” (E15)

Table 3. Summary of key future research.

Areas/ethical principles	Possible research questions	Research directions
Beneficence	RQ1: How can the maintenance of social interaction, trust, and well-being within a virtual team using a CA be ensured?	Investigating the impact of different communication modalities (text-based, voice-based, avatar-based) facilitated by CAs on social interaction within virtual teams. Designing interventions or guidelines to promote trust-building activities and foster team cohesion in CA-mediated virtual teamwork. Exploring the influence of team diversity (cultural, linguistic, etc.) on social interaction and well-being in virtual teams utilizing CAs
Non-maleficence and privacy	RQ2: What adjustments to existing privacy policies need to be made to counteract data misuse through the use of a CA in virtual teamwork?	Examining the legal and ethical implications of data privacy in the context of CAs used in virtual teamwork. Designing robust security measures and encryption techniques to safeguard personal data transmitted and stored during virtual team collaboration. Investigating the impact of privacy policy transparency and user control mechanisms on users’ trust and willingness to share data with CAs.
Accountability/autonomy	RQ3: In which areas of virtual team collaboration is it necessary to limit the independence of the CA from the team to avoid endangering the autonomy of the team and individual team members?	Investigating the optimal balance between autonomous decision-making by CAs and team decision-making in different stages of virtual team collaboration. Designing adaptive CA interfaces that allow team members to customize and influence the decision-making processes of CAs. Exploring the impact of CA transparency and explainability on team members’ acceptance and trust in CA-generated decisions.
	RQ4: Which entity can be held accountable for the actions, decisions, and consequences resulting from a CA or with the help of a CA in virtual team collaboration?	Examining the legal frameworks and liability regulations surrounding CA usage in virtual team collaboration. Investigating the roles and responsibilities of stakeholders (organizations, developers, team members) in assuming accountability for CA actions and decisions. Designing governance models and contractual agreements to allocate responsibility and mitigate potential risks in CA-mediated virtual teamwork.
Explicability	RQ5: How can the building of trust within virtual teams using a CA be strengthened?	Investigating the impact of transparency, explainability, and accountability mechanisms in CAs on trust formation within virtual teams. Designing interventions and training programs to enhance team members’ trust in CAs and foster positive perceptions of their reliability and competence. Exploring the influence of team communication strategies and team-building activities on trust development in CA-mediated virtual teamwork.
	RQ6: How can the processing of personal data of team members be explained to users to increase transparency?	Developing clear and user-friendly interfaces that provide detailed information about the data processing practices of CAs in virtual team collaboration. Investigating the effectiveness of different communication methods (visualizations, notifications, privacy dashboards) in conveying data processing information to users. Designing educational materials and privacy literacy programs to enhance users’ understanding and awareness of data privacy in CA-mediated virtual teamwork.
Justice, solidarity, and workspace awareness	RQ7: How regularly and by what means should users of CAs in virtual collaboration be informed about the use of their data to ensure comprehensive awareness?	Investigating the optimal frequency and timing of data usage notifications and updates to keep virtual team members informed and aware. Designing user-centered communication channels (email, in-app notifications, interactive interfaces) that effectively convey data usage information to users. Exploring the impact of personalized and customizable data usage notifications on users’ comprehension and trust in CA-mediated virtual teamwork.

Data availability statement

A pseudonymized form of the transcripts can be made available to interested parties upon request. Since even after pseudonymization it cannot be guaranteed that the interview partners cannot be identified on the basis of their statements and their job title, we refrain from making the transcripts openly accessible. All other materials, such as the interview guide, are freely accessible in the appendix.