COLOUR – workshop Bias in AI
Details
Photography project
COLOUR is a project exploring bias in AI. How do our biases influence AI and its outputs?
The foundation of the project is a large database of close-up photographs of people, created by Johan Nieuwenhuize. The images focus on skin, but also include elements such as clothing, jewellery, scars, and tattoos. The database represents a diverse range of skin colours, ages, and cultural backgrounds.
Diversity and inclusion workshop
This database forms the basis of a workshop designed to raise awareness of bias in AI. During the workshop, participants complete several rounds of labelling a dataset of photographs. In the first round, they label images under time pressure using a set of pre-made labels. In the second round, they create their own labels and take as much time as they need.
In the final round, participants assign photos to pre-made labels displayed on walls or tables.
Group dialogue
At the end of the workshop, the group engages in an open dialogue to discuss:
- Where they recognise bias in AI
- How they labelled the images and the biases they encountered
- The role of their identities and biases in the labelling process
- The differences between the labelling rounds
The discussion also explores whether participants’ perspectives shifted during the rounds, as well as the principles and applications of AI and its inherent biases. Ultimately, the group examines how bias in AI impacts its outputs and the implications for diversity and inclusion within organisations.
What do participants get out of it? What is the take away?
This workshop focuses on raising awareness and offers several key takeaways.
Participants will:
- Become aware of the exclusionary mechanisms and biases in AI
- Confront their own prejudices
- Experience firsthand how data labelling works
- Understand the role their identity plays in labelling data
- Learn about AI technology and “clickwork”
- Gain insight into how AI impacts diversity and inclusion in organisations
- Engage with data labelling in a playful, hands-on way
- Collaborate to create a temporary improvised exhibition
- Receive a follow-up email with links to articles on AI bias for further reading
This workshop was developed by Johan Nieuwenhuize in collaboration with the lectorate Philosophy and Professional Practice of The Hague University of Applied Sciences.
Background
An algorithm is a step-by-step plan consisting of rules in a fixed order to solve a problem or achieve a goal. Advanced self-learning algorithms are a core component of artificial intelligence (AI). AI is a powerful tool for addressing both simple and complex challenges, from smartphone camera apps to chatbots.
On social media, AI tailors content to personal preferences by analysing interactions—such as liked posts, followed accounts, and time spent on videos. While convenient, this personalisation creates a “filter bubble,” narrowing the diversity of content users encounter.
However, AI also has a darker side. Its use can lead to far-reaching consequences, as demonstrated by the Dutch childcare benefits scandal (de toeslagenaffaire). The Dutch Tax Authority used an AI-driven risk selection system to flag benefit applications for additional checks. Allegedly, the system incorporated legally irrelevant factors—such as gender, religion, ethnicity, and address—into its decision-making.
As a result, applicants with dual nationality were disproportionately targeted, leading to many being wrongfully forced to repay their benefits in full.
“The Dutch childcare benefits scandal (de toeslagenaffaire) demonstrates that we need explainable AI rules.”
Source:De toeslagenaffaire toont aan dat we uitlegbare AI-regels nodig hebben by Błażej Kuźniacki, www.uva.nl, 13 Februari 2023, retrieved May 2024. https://www.uva.nl/shared-content/faculteiten/nl/faculteit-der-rechtsgeleerdheid/nieuws/2023/02/de-onthulling-van-het-kinderopvangtoeslagschandaal-kan-betekenen-dat-nederland-vooroploopt.html (1.)
“Johan’s workshop made me realize that I unconsciously bring biases to categorizing. An dhat how I think about something is by definition coloured. Very interesting!
The workshop actively encourages to reflect on the power and possibilities of AI.
A must for anyone who is curious about how AI has a large impact on our lives.”
– workshop participant, June 15th, 2024
“To me the workshop clarified arbitrariness and in particular the non-arbitrariness behind the omnipresent algorithms and AI we encounter in our daily lives.”
– workshop participant, June 15th, 2024
Workshop Structure
The workshop begins with a sensitiser activity, asking participants: “Which labels would you use to describe a human?
It then proceeds through:
- Labelling round #1 (time-pressured)
- Labelling round #2 (self-directed)
- Coffee break
- Labelling round #3 (pre-made labels)
- Group dialogue
The workshop lasts 1–2 hours, depending on the group and the depth of the discussion. Customised workshops are available upon request.
Participants in the workshop will:
- Be confronted with their own prejudices
- Experience firsthand how data labelling works
- Explore the role their identity plays in labelling data
- Learn about AI technology and “clickwork”
- Become aware of AI’s exclusionary mechanisms and biases
- Gain insight into how AI impacts diversity and inclusion within organisations
- Engage with AI in an interactive, playful, and hands-on way
- Collaborate to create a temporary improvised exhibition
- Receive a follow-up email with links to articles on AI bias for further reading
Investment
The cost for the Diversity and Inclusion Workshop: Algorithmic Thinking with Photography is €95 per person, with a minimum of eight participants. This includes a preparatory meeting, materials, and a follow-up email. Prices exclude VAT and travel costs.
For a quotation and to check availability, contact Johan at johan@johannieuwenhuize.nl.
