Results

Final report – by the Workshop organizers

The brainstorming activity during the workshop was organised into two distinct panels, whose panelists were the authors that discussed their positions. The first panel focused on AI systems and their impact on human beings; it discussed examples of typical problems that AI systems pose to the users (such as ambiguities that confuse users, lack of control, lack of trust, etc.) and what are design, development, and user testing methods and practices currently adopted in AI.

The second panel addressed the HCI competences that AI specialists should have to design AI systems that are beneficial to human beings. It tried to highlight the interaction paradigms/modalities/metaphors for AI systems that best support the interaction with users, the HCI theories, principles, and methodologies that can prevent failures of AI systems, the design, development and user testing methods available in HCI that could be adopted in AI and that can empower people.

In the following, we distill the main thematic areas and topics that were developed from the presentations and the discussion in the two panels. A more detailed set of aspects can be found in the digital workspace that was collaboratively created during the workshop by taking note of the main points covered by the different participants’ positions.

Fostering the mindset. A first challenge to be addressed is the creation of the right “mindset” toward the HCI discipline which is often neglected by engineering and technically-minded students – (and faculties as well) (Ardito and Di Noia), but can instead solve several problems that might occur in non-properly designed AI systems: 

• Differences between the nature of ‘intelligence’, as meant in technical AI terminology, and the way we think like humans (Navigli)
• Differences between the system model and the user mental model of an artefact (Jeon)
• Algorithmic biases and failures in automation due to the probabilistic behaviour of AI systems
• Filter bubble (Ardito and di Noia)
• Lack of intuitiveness of AI algorithm results

AI as a paradigm change. AI requires a paradigm change in designing interaction: from designing rules (algorithms) to designing tools that learn rules from data (Rizzo et al.). Designers/developers thus need to design the meta-systems that then shape up at runtime the system’s interaction (Jeon). However, due to the problems that AI systems bring with them, the design should focus on human-machine cooperation without delegating AI to create the human world (Cabitza; Väänänen and Olsson). The lack of a user perspective in design and evaluation might lead to non-effective systems, regardless of their algorithmic performance (Andolina and Rocchesso). Topics to be covered could be:

• Substitution vs augmentation
• User control vs system autonomy 
• Unconscious biases and the need for fairness 
• Algorithmic opacity and the need for computational transparency (Wong)
• Explainability  (Cau and Spano)
• Interpretability (Oviatt)

Core HCI. The core concept of HCI can allow designers to consider the user perspective, as they focus on the users’ involvement in every aspect of the design. Although it is possible that students took an introductory course of HCI in their bachelor, it is crucial to devote some amount of time to reinforce this aspect (Väänänen and Olsson) and to remind (or teach) basic methods and techniques of user-centered design and design paradigms: 

• User-centered design, Usability, and User eXperience
• Methods for user involvement: questionnaires, interviews, wizard of oz (Andolina and Rocchesso)
• User-based evaluation: real vs proxy tasks, experimental vs in-the-wild evaluation.
• Problem-based thinking (who’s the user) (Gennari et al.) 
• Design thinking (Rizzo et al.)
• Design fiction (Maliza and Carta)
• Design toolkits (Gennari et al., Malizia and Carta)

Interaction paradigms. The notion of interaction paradigm should be adequately understood by students to be able to design the interaction in an AI system properly. An accurate reflection on new interaction paradigms should be fostered (although technical and implementation aspects should not necessarily be part of an HCI course): 

• Conversational User Interfaces 
• Direct manipulation
• Natural gestures
• Multimodality (Oviatt)

Advanced HCI topics. Although this part could be adapted to the specific requirements of the program, it would be essential to address at least some advance of the following topics: 

• adaptive vs adaptable interfaces 
• user modeling (as the basis of personalization and adaptive interfaces) 
• Intelligent user interfaces
• agency and proactivity
• human-robot interaction and its peculiarities as, for instance, autonomy vs user control, embodied interaction, etc.  (Nardi; Sciutti and Sandini) 
• End-Users Development as an approach for user trust and control (Wong) 

Interdisciplinary approach. HCI is a genuinely interdisciplinary approach. The interaction with AI systems, especially when it is based on novel natural interaction paradigms, requires notions from other disciplines, to name but a few:

• cognitive architectures and social cognition basics (Sciutti and Sandini) 
• shared perception (in particular for programs focused on robotics

Ethics. Finally, a relevant space should be devoted to discussing ethical aspects related to the design and the broad deployment of autonomous systems (Malizia and Carta). Ethics should be considered in the double aspect of (a) the possibility (and the risks) of embedding ethical decisions in autonomous systems and (b) the ethical implication of designing intelligent systems, that is helping students understand the impact of AI (Wong).