A creative approach to workshops on AI-powered design

A case study of running a workshop on AI and user-centric design with imaginative scenarios

by Maira Ribelles

A non-traditional approach for non-traditional technology

Imagine an owl with narcolepsy that needs someone to keep waking him up so he can go find food during the night. As bizarre as it sounds, that's exactly the kind of challenge our workshop participants tackled using a Large Language Model—in this case a custom GPT. This article is a case study that presents one of our initiatives to share knowledge on designing products powered by Large Language Models (LLMs).

More specifically, apart from having a fun time, these were our three primary objectives:

1. Enable team members unfamiliar with designing LLM-powered products to understand the capabilities and limitations of this technology in a practical way
2. Demonstrate the simplicity of prototyping and iterating using custom GPTs
3. Let participants understand how a user-centered design approach fits in the process of designing LLM-powered products

Our Vision and Practical Approach as a Design Studio

At C°F we are researching, exploring and experimenting with AI techniques, to better understand the impact AI has on our work as a data visualizations studio—and equally importantly, the exciting opportunities it offers to improve our process. We want to develop a responsible approach to designing with and integrating Artificial Intelligence into our products. To achieve this we need to create awareness and understanding within our team, clients and users about AI technology and the impact it has on our work.

One of the initiatives we undertook with the goal of creating awareness within this domain was to conduct workshops: first, together with Developer Philo van Kemenade, we gave a workshop to industry professionals at the “AI & Erfgoed Conferentie ” (AI & Heritage Conference) in March, and a month later we gave a similar workshop internally at our studio. Keeping in mind what we learned from the first one, the second iteration encouraged us to approach the session with an open and creative mindset.

Workshop setup

We began with a one-hour theoretical presentation where we explained the basics of LLMs, their capabilities, limitations, and best practices for designing AI-powered products, using examples from previous projects. After the theory session and a break, we continued with a hands-on workshop that lasted one and a half hours.

The participants worked in pairs and each group was provided with a ChatGPT user account. Creating groups of two was ideal for such activity, since both participants could discuss and work effectively using one single laptop during the creation of the custom GPT (Generative Pre-Trained Transformer). A custom GPT is a version of ChatGPT that you can create for a specific purpose. GPTs allow users to introduce rules or instructions with natural language to define the behavior of the GPT. That way, anyone without coding skills can create a customized version of ChatGPT to be more helpful in specific tasks or scenarios.

An unconventional approach

When organizing workshops, we strive to make them engaging and relevant to our audience. This time, knowing our audience leans towards the creative side, we adopted a more experimental approach that came from the following premise: Designing with LLMs requires both an open mindset and a strong focus on the user's context and needs rather than relying solely on the capabilities of the technology.

Keeping this in mind, I drew inspiration from the board game Dixit, which features cards with imaginative illustrations of various characters and situations. When playing Dixit, I always empathize with the characters and try to understand their contexts—similar to the process of designing a concept. In a real-world scenario I would conduct thorough user research, but in a workshop with limited time, quickly observing a Dixit card and coming up with the characters’ context,challenges and needs sounded like a great idea to convey user centricity in a straightforward way.

Another advantage of this approach is that we all have preconceived notions about what AI products can do, which can be limiting when trying to come up with creative ideas. Using Dixit cards proved to be an excellent tool for prompting participants to come up with unconventional ideas, breaking free from their usual thought patterns.

A user-centered approach

Let’s dive deeper into the details on how the Dixit cards encouraged the participants to go through a user-centric thought process to eventually come up with ideas and prototype them into a custom GPT. We guided the participants to go through the five different stages most designers in the UX field should be familiar with: empathize, ideate, define, prototype and test:

Empathize: The first thing we do in user-centric design is to understand the people we are going to design for and the context around them. The workshop participants defined their character’s situation and the problem it was facing by answering very simple questions such as: “What’s the character’s story?”, “What are its dreams and wishes?”, “What are its frustrations?”, and “How is the character feeling?”.

Ideate: The second step was to come up with potential solutions to help the character overcome their challenges. This activity had specific rules to make sure that later on the participants would be able translate the chosen solution into GPT instructions. We asked the participants to think of themselves as remote helpers for their assigned character, so that they couldn’t come up with more “service-like” solutions rather than physical ones. The questions that helped them come up with the ideas were: “What is your goal?”, “What knowledge or external information would you need to help the character?”, and “What actions would you take to help the character?”.

Define: In this phase, participants developed a detailed, step-by-step plan to assist the character. This included outlining specific actions and resources needed (Refer to image number X).

Prototype & Test: The last two steps consisted of putting all the thinking and concept into practice by creating a custom GPT. The participants wrote the instructions in natural language and tested the prototype (GPT) themselves.

Main learnings from facilitating the workshop

Guide your participants through the process: If I had to choose a key takeaway from planning and facilitating this workshop, it would be the importance of guiding participants effectively through the process. When aiming to convey user-centricity in a workshop with limited time, Dixit cards proved to be an effective tool for facilitating key stages: Empathize, Ideate, and Define.

Provide examples for clarity: To help participants translate their plans into instructions for the GPT, we thoroughly explained each template and provided clear examples. Participants could refer to these examples, displayed on projected slides, while filling in their templates and creating their custom GPTs. This hands-on approach ensured that everyone understood how to apply the concepts discussed.

Wrap up with presentations: Ending the workshop with presentations allowed participants to showcase their prototypes in a fun and engaging way. Each group demonstrated their prototype, which led to a lot of laughter and enjoyment.

Feedback and reflections: Most participants were pleasantly surprised by how well their prototypes worked. The scenarios demonstrated not only the versatility of the technology but also highlighted the phenomenon of GPT hallucinations, where AI adapts creatively to a given scenario. This sparked valuable reflections on the importance of validating, testing further, and iterating on prompt engineering to achieve the highest possible control and accuracy.

Workshop final discussion

At the end of the workshop, we briefly presented some slides to encourage participants to think beyond their custom GPT prototypes. We emphasized that the real value in a digital product like this often emerges when the GPT flow of instructions is broken down into different components that integrate into a single product, achieved through the use of a dedicated UI. Additionally, we discussed the importance of further testing and assessing the degree of validation each product would need.

The final word

The task of crafting fictional stories around characters from Dixit cards to later translate them into custom GPTs turned out to be a fun experience for the participants, while also bringing some unique challenges to light. The creative freedom around the activity meant that the scenarios ranged broadly in complexity, from straightforward to more complex or abstract.

Although every group successfully completed the exercise, those who opted for more complex narratives took longer to come up with a solution.

We wonder how a different crowd of participants would react to such a creative task. Perhaps it would be interesting to reflect about ways to take a less fictional approach and one more connected to real life.

We would like to invite everyone that is planning to conduct a similar workshop to get inspired in this approach. We can’t wait to hear your learnings, anecdotes and feedback if you try it out!

If you're interested in the workshop slides, you can find them here .