“Letting Users Control AI” Recap by Marie-Luise Schlander
Meet-up 10.01.19, Thursday, January 10, 2019
For the first Demystifying the Smart City meet-up of 2019, we had invited three speakers to discuss with us the challenges and solutions to get a better understanding on the implications of Artificial Intelligence (AI). Paul van Beek and Joep Elderman [Bureau Moeilijke Dingen], Mark Mooij [AI-Applied] and Wouter Moraal presented a variety of projects in order to address a series of questions linked to AI/ML and end users.
Some leading questions of the evening were: To what extent can we let the user control AI? To what extent can a user understand what AI does? What role does design have in communicating these topics and how do we design for A.I.?
Paul van Beek and Joep Elderman of Bureau Moeilijke Dingen, a design agency that focuses on the development of smart products, started the evening by showcasing AI projects designed for the exhibition artificial intelligence training centre during the Dutch Design Week in Eindhoven. Their exhibition made the inner-workings of AI and machine learning accessible to non-expert users and proposed a future vision about how users could interact with their own AI models on an everyday basis. For the exhibition, the team of Bureau Moeilijke Dingen created several prototypes whose algorithms visitors could train themselves. One of the prototypes was a TV whose algorithm can be trained so that it proposes you TV shows according to your sex, age and personal taste. Another creative example for an everyday use of AI was a spam filtering mailbox which users can train depending on their own definition of spam mail. Additionally, joint monitors in the exhibition informed the public about the inner-workings of the algorithm and contributed to a better understanding of AI and machine learning while enabling the users with a wink to discover creative advantages of AI for their everyday life.
Moreover, Bureau Moeilijke Dingen is currently building an AI kit which they’ve also presented during the design week in Eindhoven. Following the creators, the AI kit would allow ‘normal people’ to approach AI and its inner-workings as well as the process of training and building algorithms as a non-expert user. By training people how to approach AI, Bureau Moeilijke Dingen wants to democratize AI which is still mainly in the hands of big companies. The ultimate outcome of the project should thus be a software free to use for everyone.
As a second speaker, Mark Mooij [AI-Applied] presented three projects he has realised in collaboration with different partners. Specialised in Artificial Intelligence, Mark co-founded Ai Applied together with Bruno Jakic in 2007. In 2011 he collaborated with CLEVER°FRANKE to realise a Weather map which visualises the comparison of the actual weather data and how it was perceived by the people. This example shows the difficulties AI can have when dealing with subjective comments of people and how AI has to be trained how to cope with language.
TrendViz, Mark Mooij’s second example showcases how data analysis of online news articles and social media messages can be visualized in an intuitive way presenting patterns and relations and tone of voice. For the example, websites of pharmaceutical companies were used. However, if TrendViz’s algorithm could be trained to compare general news articles online and analyse how those news agencies report differently on a certain topic, TrendViz would present a way to democratize the news.
Lastly, KeenCorp, an analyse tool based on company-internal information on employee satisfaction, raised more critical questions. Based on in-house communication (daily communication via e-mail or chat), the tool allows to anonymously measure employee satisfaction and reveal possible issues within a company by displaying a daily metric overview. To ensure anonymity, the tool does not store text data, only numerical value. Following Mark’s presentation, the audience discussed advantages and disadvantages of the tool and critically questioned if they would want to use it in their respective companies. Questions concerned the privacy of employees, if the application of the tool could change their behaviour and if algorithms might be (mis)used to judge employees’ performance. To ensure transparency concerning such a sensible topic, the audience rated it important to explain the inner-workings of the algorithm with in the respective company and to give employees full access to the daily results. Nonetheless, one question remained unanswered: “How can users control the profiles that others create of them?”.
This question was then picked up by Wouter Moraal (MSc) who creates interactive art installations to deepen the public — and his own — understanding about the role of new technologies in society. He presented his latest development Artificial Impact, an educational board game. The game is about understanding the workings of deep learning algorithms, including the data collection, the algorithm itself and how to use it. During the game, players can train an algorithm which then makes predictions on the players’ characters based on the information they put in before. Those predictions can either be true or biased, depending on the accuracy of information given. By letting players experience how the algorithm can be biased and (mis)judge them, Wouter Moraal wants to raise awareness for the dark sides of the use of algorithms by institutions.
“I would like people to understand what the limitations of DLA could be, concerning both the data and the outcome, I would like people to understand the whole process.” Wouter Moraal
If you’re interested in getting your own educational board game, you can contact him online.
Concluding discussion
Moderator Gert Franke opened the discussion with the question who of those present knew how to explain to end users / their clients how AI functions. Following the speakers and the audience, it is important to give short but illustrative explanations to let end users better understand the inner workings of AI which could be best explained with regards to the specific context of the user. In this scope, it is equally important to manage clients’ expectations in order to only give ground for right expectations.
“What should we educate the end users, those being affected by AI systems? What should we tell them?” Gert Franke
As an answer, especially Wouter Moraal raised the concern that people affected by the implementation of AI in decision making processes often do not have the power to change something about the systems. He thus asks for more transparency concerning the communication about the inner-working of algorithms. Mark Mooij agreed when saying that in the end, it is most important to explain properly to the end user concerned how a certain algorithm works / affects the human in question. As it can be frustrating for humans to be ‘in control by machines’ without having any insight in their decision-making processes, it is crucial to be able to reconstruct their artificially intelligent reasoning.
Maybe then, sooner or later, we will speak differently about AI and algorithms. When Gert asked if the speakers reckoned if they would still have to explain the same about AI in five to ten years, different answers were given. One person in the audience told about classes that are given to better educate people about AI. Also, Wouter Moraal appeared optimistic indicating school education could allow for a broader knowledge on AI/ML in the upcoming years. Paul van Beek and Joep Elderman suggested that the current hype on AI will help for people to better understand. However, they also criticized that the general public often tends to have discussions about possible future uses of AI. As those uses are not feasible yet, the discussions are too early which prevents other, more meaningful discussions at present.
— -
INFORMATION: DEMYSTIFYING THE SMART CITY RESEARCH PROGRAM
With this research program, Sensor Lab aims to stay true to its principals of bringing together the smart technology community, to collaborate on Proof of Concepts that help demystify the Smart City. Together we will provide tangible examples of what is occurring in the smart city, and how we can inform residents, in addition to establishing a standard and protocols for future developments.
Each of the research tracks consists of field work, meet-ups, and roundtable discussions with experts, artists, designers, stakeholders and citizens. Each outcome is accompanied with documentation material that outlines the specific methods, and insights used and gained in the process. The projects outcomes will result in public events, be shared with the press, and decision and policy makers.
https://www.sensorlab.nl/research-program/
Marie-Luise Schlander
Master’s student in Human Geography at the University of Bonn and Utrecht. Specializing in Urban Geography with interest in political and urban structures and developments, civic participation and Smart Cities.
