ADS Drinks & Data: Data Science on Human Interaction

On Thursday 26th September 2019, ADS hosted a meetup on Data Science on Human Interaction at CWI. Jacco van Ossenbruggen, head of Information Access at CWI and associate professor at VU Amsterdam, chaired the meetup. Our speakers explored the different ways humans interact with technology, how to track these interactions and how to ensure transparency when classifying these interactions.

Our first speaker, Pablo Cesar, leads the Distributed and Interactive Systems group at CWI. Pablo’s research focuses on human-centered multimedia systems and ubiquitous computing. He is also the primary investigator (PI) from CWI on several H2020 and Public-Private Partnership projects, like VRTogether.

Pablo’s talk centered on the sensor technology he and his team have been working on to quantify a user’s experience. This involves wearable technology and tracking people’s reactions to events including theater productions and dance events.

To see Pablo Cesar’s slides, click here.

Lars Lischke, the second speaker, is assistant professor for Human-Computer Interaction at VU Amsterdam. In his research, he focuses on the interaction with large high-resolution displays and interactive visualizations for complex data.

With advances in computing technology, all parts of society take decisions based on complex and processed data. Critical for the decision-making process is the communication of relevant information itself as well as an understanding of the data sources and applied processing techniques. This requires novel interaction techniques for exploring and visualizing data beyond exploring data at the desktop display.

Lars provided an overview on his research on interacting with visual information in human-computer interaction (HCI). He gave examples of how to explore data on multiple mobile devices, how to interact with large high-resolution displays and how information can be projected into industrial workplaces.

To see Lars Lischke’s slides, click here for Part I and Part II.

Our final talk came from Emma Beauxis-Aussalet, senior track associate of the Data-Driven Transformation track at the Digital Society School at HvA, where she works on applying AI for the best interest of society. Her current projects include chatbots for addressing online bullies, and educational materials to develop AI literacy in the general public. Emma particularly focuses on enabling transparent and accountable AI where bias and errors are explicit and understandable by all stakeholders, including those with limited AI literacy.

Emma discussed the uncertainty metrics and visualizations that concern both audiences i.e., stakeholders with extensive or limited AI expertise, and demonstrated two shortcomings of existing practices. First, end-user information requirements are not fully addressed by expert-oriented uncertainty assessments. Second, AI systems may not be tested sufficiently, e.g., with test sets that are too small, or not representative of the end-usage conditions in which classifiers are applied. In particular, the issue of error variance is largely under looked but is crucial to assess classification biases and test set reliability.

To see Emma Beauxis-Aussalet’s slides, click here.

To attend our upcoming Meetups, please check out our Meetup Page.