The majority of data scientists are working for humans. Their data is often arising from human behavior, and their data engineering and modeling result in algorithms and tools, visualizations, recommendations or decision support systems, nowadays often broadly construed as Artificial Intelligence (AI): systems smart enough to support or sometimes replace human decision making. We all use and benefit from these technologies in our daily lives, for example when we rely on Netflix or Spotify to give as good recommendations to watch or listen to. However, anyone who sometimes gets annoyed by the somewhatoff-putting recommendation for a show on Netflix or the obscure music recommended by Spotify, will understand the need to better understand the recommendation output as well as the need to interact with the system to improve these recommendations in some way…|
To develop AI systems that adequately support humans, a solid understanding of human cognitive capabilities and decision making skills is needed. It also requires knowledge on how such tools can be designed to be understandable, interpretable and explainable to their users as well as being sufficiently interactive and controllable whenever the users feel they need to provide feedback to the tool when it is wrong, biased or not sufficiently tuned to their needs. If not, in the end systems like this might be abandoned by their very users and will not reach their (business) goals.
For this purpose the course will introduce the students to relevant topics from cognitive science as well as more recent developments in interactive machine learning and explainable AI and then will apply this knowledge to develop and test such systems in a interative user-centric design process similar to agile/lean start-up approaches.
In the course we distinguish between data science tools and methods that support regular (consumer) users in their daily needs and decisions (recommender systems, web shops, apps) as well as tools that support expert decision makers in their data-driven decision making processes, including data scientists themselves. To support the first type of user, it is challenging to determine how to use and present data science tools in order to create good user experiences and support users’ choices that are in line with their – often unknown – values, needs and motivations and to what extent these tools need to be interactive and explainable. To support the professional users to improve their (data-driven) decision processes, interactivity and explainability might be even more important to improve trust and to increase the likelihood of these tools to be implemented into the flow of business decision making. Interestingly, recent research shows that even data scientists themselves are in need of good tools to better understand and interactive with their models, as they often fail to adequately interpret their own modeling efforts.
- To understand cognitive mechanisms related to human behavior, decision making, and human experiences and their application to the data science domain and applications such as recommender systems and decision support tools.
- To understand recent techniques and methods in interactive Machine Learning and explainable AI
- To conduct research on existing data sets from a cognitive science perspective as input for innovative design of novel data science / AI tools that offer interactivity and/or explainability
- To understand and apply design thinking as a process towards innovative products & services (e.g. human-centered design process) based on user needs and improving user experiences
Students will work in groups over the entire course. These groups will work together during the different activities of this course. The course consists of lectures and discussions to cover learning objectives 1 and 2, and design workshop with feedback sessions to over learning objectives 3 and 4. |
1. Lectures focus on theoretical background of cognitive science, including a clear understanding of human cognition, such as attention, memory, language, human problem solving and decision making, as their applications to data science in terms of decision support systems, recommender systems, interactive machine learning and explainable AI tools. Students are expected to come prepared to the lectures by reading relevant chapters or papers. Lectures are followed with discussion sessions on relevant academic papers.
2. Discussions focus on the application of cognitive science theories and interactive and explainable AI onto topics relevant for data science, by studying and discussing relevant papers. Students prepare by reading a paper and post discussion points prior to the discussion meeting. In their discussion points, students try to integrate knowledge from the lectures in discussing the paper. One group will chair and prepare the meeting using the discussion topics as posted by the students.
3. Design Workshops: During the semester the groups will work towards an innovative design of an interactive / explainable data science tool. As part of the design process, students will analyze existing datasets from a specific cognitive science perspective. These analyses will help the ideation face in the design thinking part of the course. We will provide several topics related to current research projects at JADS (for example sensor data, decision support tools or (group) recommender systems). Data and theory in cognition will act as inspiration for this process and the focus will be on tools that provide some level of interactivity/explainability. Further steps will follow the user-centered design process in which the prototype is developed based on users’ and stakeholders’ requirements and validated with users. The outcome is not expected to be fully functioning software, but rather a prototype which can demonstrate the main activities with good quality and something which e.g. could be presented to investors or other stakeholders. These workshops go through the entire design cycle including user needs elicitation, ideation and concept design, prototyping and testing.
4. Feedback sessions: as students work on the design project, we will have several opportunities for students to discuss their progress and get feedback from their peers and the teachers. These will typically involve a short presentation from each group followed by time for questions & feedback.
5. Professional skills: as part of the professional skills in the master, you will be able to follow a workshop on ideation, outside the regular lecture hours. These skills will be assessed separately.