Current students


PUCCI EMANUELECycle: XXXVII

Section: Computer Science and Engineering
Advisor: MATERA MARISTELLA
Tutor: CAPPIELLO CINZIA

Major Research topic:
Explaining technology through conversation: can we redesign explainable AI putting humans at its center?

Abstract:
Summary
This research wants to define a Human-centered Artificial Intelligence (HAI) framework in which new participatory design methods can help envision future scenarios and disruptive uses of interactive technologies for the adoption of reliable, safe and trustworthy intelligent systems. In the general context of a more transparent and comprehensible AI, the research aims to define a new methodological framework for technologies that will help domain experts and the final users better understand and control the AI-based systems they work with.

Introduction
Artificial Intelligence (AI) has the potential to transform industries and societies. Much progress has been made since the 1950 Turing Test. Yet today AI algorithms are still conceived as a sort of a black-box, with implications for people such as ethics and wellbeing [13]. It is important to develop additional frameworks and tools that can facilitate the development of AI-powered systems that are intrinsically explainable and, as such, can be controlled, especially in sensitive environments.

Research focus
This research aims to contribute to the development of HAI by exploring how this framework can improve the design of intelligent systems. The COVID-19 crisis has amplified the importance of such systems. The imposed lockdown has forced institutions into the exploration of new digital approaches: in this “new normality”, people with different backgrounds and peculiar vulnerabilities are less likely to receive the needed support and are therefore the ones who might lose the most. The research will focus on studying how domain experts and final users interact with AI-based technology, and how this technology can better solve their needs. I will focus on the intersection of Human-Computer Interaction (HCI) and Artificial Intelligence (AI): how does AI revolve around society, and how can we work to design better systems, with a value for the final users, inspired by themes such as inclusivity and underrepresentation?HCI is a point of convergence for psychologists, designers, engineers, data scientists, sociologists, and economists. Artificial Intelligence is itself a very technical field. Creating connections and points of communication among those two apparently different worlds is not easy, yet during the last years something has changed. People and researchers have been recognizing the importance of broadening the debate on AI and bringing it out of its traditional engineering sphere to investigate the impact it has on individuals and, more in general, on society. Due to its newness, the sector has no single definition yet. Different terms have been used depending on who is addressing the matter. Currently, there are mainly two streams of research: ;
    ;
  • Human-centered Artificial Intelligence (HCAI): the research community applies HCI concepts, in particular the human-in-the-loop framework, to design artificial intelligence systems capable of considering the users, giving them the capability to intervene during the process of training, deployment, and evaluation of the systems themselves.
  • ;
  • Interactive Machine Learning (IML): it focuses on the same issues of HCAI, yet from a more technical and engineering perspective. Here the focus is on the explanation of single features of automated systems, in order to let the final user or the domain expert understand why the system has come to a certain conclusion.
  • ;
; Both schools fall under the umbrella of Explainable AI (XAI), whose goal is to extend the “black-box” model of AI by explaining why an intelligent system has come to a specific outcome [15]. Although a lot of work has been done during the last decade, the recurring pattern is to explain decisions and their impact from technician to technician. I see it as a big obstacle toward a society which can actually benefit from intelligent systems, and so do many HCAI practitioners and researchers. The fundamental element in HAI is indeed the human component. Therefore, I will address the definition of robust interactive paradigms, focusing on adequate UIs but also on “negotiation” processes that can empower the end users to understand first and then control and customize the AI system behavior [10]. User involvement will be paramount in the design and evaluation of the methods and tools proposed by the research.  

Methodology
The research will need a comprehensive study of the existing literature paired with user-based research, followed by a definition of new models, methods and frameworks to be then evaluated. The outcomes will be refined with a cyclical agile approach. A multidisciplinary approach will be encouraged. Therefore, while my focus will be on HCI aspects, I will try to collaborate with AI specialists, and keep refining the outcome of the study following a human-centered approach. The three years of research will be organized as follow: ;
    ;
  • A thorough analysis of the state of the art will be conducted in the first year. I will examine prior works to better comprehend AI technologies, how they are used to support the design of AI systems, and the recent advancements in HAI. I will in the meanwhile select a reference domain and carry on co-design studies to understand what the final users want to be explained, and how: this is a paramount step in order to offer meaningful explanations. Advanced co-desig
  • ;
  • In the second year I will focus on designing the technological framework to accomplish the goal of the research. I will identify the aspects, and the related modeling abstractions, that can extend current AI algorithms to make them transparent and This activity will allow me to identify the foundations for an HAI-oriented methodology, supported by design guidelines focusing on the two sides of HAI: adequate interaction paradigms, to let the users understand and control AI systems, and AI models purposely extended to improve their transparency.
  • ;
  • In the third year the impact of the methodology and the related tools will be evaluated through on-the-field analysis. The results achieved for the selected reference domain will be abstracted and generalized though a reflection process that will elicit the lesson learned and extend the HAI framework. One major outcome will be a set of design patterns integrated into a technological framework supporting the design of interactive AI systems.
  • ;
; Research contribution
The research wants to contribute to the growing debate on automation and HAI focusing on the importance of an explanation that can be digested by non-technical users as well. Enhancing explainability and control on AI technologies in those contexts will lead to a more adequate support for the end user through the design of systems that are reliable, safe and trustworthy. Overall, the research question will be: ;
    ;
  1. Can we define frameworks to understand user’s expectation about explainable systems?
  2. ;
  3. Can we then apply the frameworks to improve the relationships between human and intelligent systems?
  4. ;
  5. Are the new explanation actually useful to the end-users and domain experts? Can we define an evaluation framework?
  6. ;
;  Answering this question could also have implications on related issues, as: ;
    ;
  • The fairness of artificial intelligence and deep learning algorithms;
  • ;
  • How to deal with eventual social biases introduced by the technology;
  • ;
  • How to match technology with relevant social norms.
  • ;
; Bibliography  ;
    ;
  1. Amershi, S., Weld, D., Vorvoreanu, M., Fourney, A., Nushi, B., Collisson, P., ... & Horvitz, E. (2019). Guidelines for human-AI interaction. Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems (Paper 3, 1-13). ACM.
  2. ;
  3. Burnett, M., Stumpf, S., Makri, S., Macbeth, J., Beckwith, L., Kwan, I., Peters, A. & Jernigan, W. (2016). GenderMag: A Method for Evaluating Software’s Gender Inclusiveness. Interacting with Computers, doi: 10.1093/iwc/iwv046
  4. ;
  5. Burr, C., Cristianini, N. and Ladyman, J. 2018. An Analysis of the Interaction Between Intelligent Software Agents and Human Users. Springer, Minds and Machines (2018) 28: 735-774.
  6. ;
  7. Calvo RA, Peters D., Vold V, Ryan, RM (2020). Supporting human autonomy in AI systems: A framework for ethical enquiry. In Burr, C. & Floridi, L. (Eds.) Ethics of Digital Well-Being: A Multidisciplinary Approach. Springer Open.
  8. ;
  9. Chen, H., Park, H. W., & Breazeal, C. (2020). Teaching and learning with children: Impact of reciprocal peer learning with a social robot on children’s learning and emotive engagement. Computers & Education, 150, 103836. https://doi.org/10. 1016/j.compedu.2020.103836
  10. ;
  11. Faiella F., Ricciardi M. (2015), Gamification and learning: a review of issues and research, Journal of e-Learning and Knowledge Society, v.11, n.3, 13-21. ISSN: 1826-6223, e-ISSN:1971-8829
  12. ;
  13. Farooq, U. and Grudin, J. Human-computer integration. ACM Interactions 23, 6 (2016), 27–32.
  14. ;
  15. Gordon, G., Spaulding, S., Kory Westlund, J., Lee, J. J., Plummer, L., Martinez, M., Das, M., & Breazeal, C. (2016). Affective Personalization of a Social Robot Tutor for Children’s Second Language Skills. Proceedings of the AAAI Conference on Artificial Intelligence30(1). Retrieved from https://ojs.aaai.org/index.php/AAAI/article/view/9914
  16. ;
;
    ;
  1. J.M.Kory,“Storytellingwithrobots:Effectsofrobotlanguagelevel on children’s language learning,” Ph.D. dissertation, Massachusetts Institute of Technology, 2014.
  2. ;
  3. Piro, L., Desolda, G., Maristella, M., Lanzilotti, R., Mosca, S., Pucci, E. 2021 - An Interactive Paradigm for the End-User Development of Chatbots for Data Exploration. Of Interact 2021, the 18th International Conference promoted by the IFIP Technical Committee 13 on Human–Computer Interaction, Bari, Italy, September 2021.
  4. ;
  5. Shneiderman, B. 2020. Human-Centered Artificial Intelligence: Reliable, Safe & Trustworthy. International Journal of Human-Computer Interaction, 36:6, 495-504, DOI: 10.1080/10447318.2020.1741118
  6. ;
;
    ;
  1. Spitale, M., Gelsomini, M., Beccaluva, E., Viola, L., Garzotto, F.: Meeting the needs of people with neuro-developmental disorder through a phygital approach. In: Proceedings of the 13th Biannual Conference of the Italian SIGCHI Chapter: Designing the next interaction (CHItaly 2019), 10 p. ACM, New York (2019). Article 22.  https://doi.org/10.1145/3351995.3352055
  2. ;
  3. Xu, W. Toward human-centered AI: a perspective from human-computer interaction. ACM Interactions 26, 4 (2019), 42-46.
  4. ;
;
    ;
  1. Riccardo Guidotti, Anna Monreale, Salvatore Ruggieri, Franco Turini, Fosca Giannotti, Dino Pedreschi: A Survey of Methods for Explaining Black Box Models. ACM Comput. Surv. 51(5): 93:1-93:42 (2019)
  2. ;