We put forward an architecture for the next generation of AI-mediated intra-organizational communication (IOC), towards enhanced team productivity and satisfaction. Our proposal rests on three key principles: hybrid human-AI collaboration via natural language interactions; diversity-aware dissemination of queries within the organization; incremental and participatory development of the IOC policy. We briefly discuss our ongoing work towards realizing the proposed IOC architecture.
(Principle 3) Incremental improvement of the AI-mediation policy in a participatory manner that involves the employees themselves, as the domain experts.
In the remainder of this paper, we first overview past work that relates to our proposed IOC platform (Section 2), we discuss the platform’s constituent modules and how they relate to the three key principles discussed above (Section 3), and we briefly report on our ongoing work for implementing the platform (Section 4).
The problem of supporting online collaboration by identifying appropriate members from a
group to work on a given project or task has received attention in recent years [
u-Help is a further relevant example of a distributed collaboration platform [
Closer to the knowledge-sharing aspect of our IOC platform is a mobile application by
Mourtzis et al. [
With the exception of the last reviewed work above, the remaining works do not focus on the IOC problem, but rather on broad day-to-day tasks or particular communities. More importantly however, none of the reviewed works attempt to consider diversity in their recommendations, nor do they propose a self-sustainable process for the improvement of the communication and recommendation policies.
Among the commercially-available IOC products that we were able to identify, closer to
satisfying our proposed principles is Starmind’s expert identification tool [
Some considerations surrounding the role of diversity in an online social platform are touched upon by Helm et al. [14]. When viewed under a descriptive lens, diversity can help capture diferences between users across their skills, practices, and personality traits. When viewed under a normative lens, it acts as an instrumental value to be promoted in support of certain fundamental values (e.g., productivity and inclusion), while being mindful of its ramifications on other fundamental values (e.g., protection and privacy). Both views are readily applicable to our IOC platform, towards ensuring that like-minded employees can work on tasks of joint interest, while making diverse employees feel comfortable to contribute and complement other employees within the frame of a dificult task [15].
policy, to a subset of the user’s colleagues, taking into account each employee’s organizational profile, including their duties and responsibilities (and other deep-level characteristics [ 15]), as determined, for instance, either by their job scope and mandate, or by their supervisor’s evaluation of their business know-how. In alignment with Principle 2, the matching above is counterbalanced with diversity-awareness measures according to legal, social, or ethical norms, which might typically relate to surface-level characteristics [15] like gender and age.
In the simplest scenario, employees can self-report whether a query they have received was appropriately forwarded to them, or whether a query they had posed was adequately addressed by their colleagues. Beyond that, however, the architecture provides for an escalation mechanism for unhandled queries to higher levels of the organizational hierarchy. Supervisors are assumed to be aware of whether their subordinates are (or should be) able to respond to a particular query, and can thus provide direct feedback for the improvement of the matching policy.
Queries that go completely unhandled are forwarded to a Quality Assurance Manager, who can investigate the existence of an out-of-scope query, the presence of a gap in organizational expertise, or other eventualities. A more mundane, but perhaps more likely, eventuality is the inappropriate parsing of a query or of some previously ofered piece of feedback. In such cases, the Quality Assurance Manager can (instruct a technical expert to) intervene and fine-tune the platform’s policies.
The provision for explicit translation and matching policies comes in support of the alignment of the architecture with Principle 3. The feedback coming from the employees and the Quality Assurance Manager can be much richer and direct compared to a labeled data-point, as typically used in supervised learning. Consequently, a policy can improve much more eficiently, and be explainable by design, by having the feedback become an integral part of its updated version. Quality-wise, this feedback leads to a less arbitrary update of the policy, compared to the often circumstantial generalizations that come from few-shot supervised learning.
Our proposed architecture does not commit to any particular implementation. In our ongoing work, however, we have identified certain tools and techniques that we believe are good fit, and which we are using to realize our envisioned platform.
For the incremental and participatory improvement of the matching policy we adopt machine coaching [16, 17, 18], a paradigm that dictates how a machine can elicit policies in a principled and eficient manner via an argumentative dialogue with humans. The translation from natural language to the logic-based arguments required for machine coaching leverages elements from NESTOR [19], a system that itself adopts the machine coaching paradigm, at a meta level, to incrementally elicit (from the Quality Assurance Manager) the translation policy that it utilizes.
Central in our implementation plan is the use of the WeNet infrastructure [20] for the eficient development of machine-mediated social interaction applications, while protecting the users’ privacy from potential surveillance of their communications, and balancing the promotion of diversity against other matters [21]. Once developed, we hope to evaluate our IOC platform within our host organizations.
This work was supported by funding from the EU’s Horizon 2020 Research and Innovation Programme under grant agreements no. 739578 and no. 823783, and from the Government of the Republic of Cyprus through the Deputy Ministry of Research, Innovation, and Digital Policy. [13] A. Krishnakumar, Assessing the Fairness of AI Recruitment Systems, 2019. URL: http:// resolver.tudelft.nl/uuid:1ce06e89-72a7-47fe-bdbd-93775732a30c. [14] P. Helm, L. Michael, L. Schelenz, Diversity by Design? Balancing the Inclusion and Protection of Users in an Online Social Platform, in: Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society (AIES), 2022, pp. 324–334. URL: https://doi.org/10. 1145/3514094.3534149. [15] L. Schelenz, I. Bison, M. Busso, A. de Götzen, D. Gatica-Perez, F. Giunchiglia, L. Meegahapola, S. Ruiz-Correa, The Theory, Practice, and Ethical Challenges of Designing a DiversityAware Platform for Social Relations, in: Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society (AIES), 2021, pp. 905–915. URL: https://doi.org/10.1145/3461702.3462595. [16] L. Michael, Machine Coaching, in: Proceedings of the IJCAI Workshop on Explainable
Artificial Intelligence (XAI), 2019, pp. 80–86. URL: https://doi.org/10.5281/zenodo.3931266. [17] V. Markos, M. Thoma, L. Michael, Machine Coaching with Proxy Coaches, in: Proceedings of the 1st COMMA Workshop on Argumentation and Machine Learning (ArgML), 2022, pp. 45–64. URL: https://ceur-ws.org/Vol-3208/paper4.pdf. [18] V. Markos, L. Michael, Prudens: An Argumentation-Based Language for Cognitive Assistants, in: Proceedings of the 6th International Joint Conference on Rules and Reasoning (RuleML+RR), 2022, pp. 296–304. URL: https://doi.org/10.1007/978-3-031-21541-4_19. [19] C. Ioannou, L. Michael, Knowledge-Based Translation of Natural Language into Symbolic Form, in: Proceedings of the 7th Linguistic and Cognitive Approaches To Dialog Agents Workshop (LaCATODA), 2021, pp. 24–32. URL: https://ceur-ws.org/Vol-2935/paper3.pdf. [20] WeNet: The Internet of Us, https://www.internetofus.eu/, ???? Accessed: 2023-04-30. [21] V. Markos, L. Michael, Post-hoc Diversity-Aware Curation of Rankings, in: Proceedings of the 14th International Conference on Agents and Artificial Intelligence (ICAART), 2022, pp. 323–334. URL: https://www.scitepress.org/Link.aspx?doi=10.5220/0010839600003116.