Meetup recap: building trustworthy AI

On September 3rd, we hosted an engaging meetup focused on the critical topic of "Building trustworthy AI." As the field of artificial intelligence continues to evolve at a rapid pace, the importance of fostering trust and transparency in AI systems has never been more clear. We invited three speakers to give presentations on their perspectives regarding this topic: Kars Alfrink gave us an introduction to the topic of contestable AI, Seowoo Nam provided us with examples of specific UX practices to make AI more transparent, and Brecht De Man talked about applications of AI technologies in the creative industries, specifically in music production. We had an audience from diverse backgrounds gathered to discuss key challenges, share insights, and explore best practices for developing AI technologies that prioritize ethical considerations and societal impact. This recap highlights the key takeaways and stimulating conversations that emerged during the event.

Our first speaker was Kars Alfink, a postdoctoral researcher at Delft University of Technology. Kars discussed his PhD thesis on contestable AI, highlighting the need for AI systems to respond to disagreements and allow for control throughout their lifecycle. He shared examples from his fieldwork in Amsterdam, such as EV charging stations, to illustrate the challenges and benefits of transparency and contestability. He proposed mechanisms like interactive control, transparency, and means for speaking out to enhance public trust and acceptance of AI systems.

The second presentation was given by C°F UX designer Seowoo Nam, who has a special interest in how AI affects UX practices. Seowoo’s talk focused on making AI more tangible and transparent through design patterns. First, she emphasized the difficulty of being critical about AI due to abstract values surrounding AI use and its hidden inner workings. She then highlighted how most AI tools are limited in their capabilities or in the accuracy of their results. To remedy this issue, her suggestions are to: add disclaimers so users know what to expect from the tool; and to open up the process of the tool so users can see how the AI reaches the results that it does.

Seowoo followed this up with a thought experiment involving one of our own projects at C°F, Voices from the Frontlines. She reimagined the project to include AI features and proposed using AI to generate narratives about climate disruptions in Africa. With this idea she presented the design patterns she had been working on, which are different ways that users could interact with and modify the AI tool at hand, involving them in the generation process and providing options to exclude unreliable sources. Seowoo concluded her talk by leaving us with four main points of advice and questions to ask when designing for transparency with AI:

• Show the process, not only the outputs. — Which step of the process matters to users?
• Explain where the information came from. — How should the source be communicated to users?
• Let users override, don't just explain. — To what extent should we allow overriding?
• Let users have say in what should be highlighted. — What aspects would users care most to adjust?

Lastly, as an audio engineer and an R&D researcher, Brecht De Man explored the integration of AI in the music industry, highlighting both its opportunities and challenges. AI is transforming music production by enabling tasks like source separation and noise reduction, but it also raises ethical concerns about data usage and the potential for homogenized music. Brecht emphasized the unique nostalgia in music, contrasting it with other industries, and argued for a co-creative approach between humans and AI. A big part of this co-creation is the democratization of music production allowed by new AI technologies: now even people without technical skill and knowledge can have a go at creating music. That said, those with the technical skill and know-how required for music production will still be greatly valued in the industry. Brecht ultimately urged musicians and other creatives to think along together with AI tools and adapt to working with them.

The panel discussion, guided by our gracious host Philo van Kemenade , generated a lot of thought-provoking points. Highlights included:

• the need for prototyping and hands on experience of AI tools for creatives
• the responsibility of designers to learn and understand the properties of AI models
• the importance of human control and presence in the loop when designing with AI tools

Ultimately, it was noted that AI technologies will not go away — rather they will mature and we will find more sensible applications for them.

It was a lovely evening full of lively conversation and discussions. We’re very grateful to everyone who joined us, and we’re looking forward to next editions of C°F meetups early next year!