Here’s what we found.

How to Fuel a Design and Innovation Project with AI

Sundberg-Ferar is actively exploring the ever-multiplying possibilities presented by AI and its power to impact our workflows and outputs as a creative team and the way we collaborate with our partners. To prospectively experiment with AI tools in our creative process, we’ve embarked on a “mock project” based on the National Parks Michigan Mobility Challenge to serve as the brief in which to test aspects of what AI can offer.

This is the third blog in a series documenting our project journey with AI and highlighting tips and tricks we’re learning along the way, plus insights on where AI could make the biggest difference in our process.

To learn more about the National Parks Michigan Mobility Challenge and to catch up previous epiphanies in our journey, check out our previous blogs.

Stop #3: The Picnic Area

What is a hypothesis workshop?

One of the ways that we lay a strong foundation for meaningful concept ideation later in a project is through a Hypothesis Workshop. Hypothesis workshops are a highly structured and collaborative activity where the team collaboratively generates hypotheses stating where the team thinks opportunities might exist for a new product, new business model or other type of innovation. Along with hypothesis statements, the team generates knowledge gap statements citing what we would need to investigate in discovery research to either prove or disprove the hypothesis.
Hypothesis workshops provide the focus for subsequent discovery research and hypotheses which are later proven as valid in research become the basis for concept ideation.

What happens when we bring the power of AI to a hypothesis workshop?

After carefully crafting the inputs for this workshop including personas of stakeholder groups involved in the National Parks, user journey maps for those personas, and relevant design, tech, and industry trends, we experimented with bringing AI tools into our live workshop workflow in two ways:

One was by designating specific team members to interface with ChatGPT 4.0 and Miro (business subscription) to expand on the team’s ideas while the others contributed sans AI use. The reason for this was based on the experience of other groups who found that when everyone in a workshop was given freedom to use AI, it led to siloed trains of thought and hindered the process of sharing hypothesis ideas and building off one-another’s work that creates the true magic of human collaboration in these workshops.

We also experimented with allowing everyone to use ChatGPT 4.0 and Miro at-will to help generate hypothesis statements and knowledge gaps. This would tell us if our own experience agreed with the experience of other groups who have tried this.

Here’s what we found:

1. Prepping the AI as a “participant” in the Workshop:

The point of hypothesis workshops is to come up with specific and novel hypotheses for completely new innovations. The strength of LLMs is in predicting likely outcomes. It’s less good at predicting the unlikely or novel outcomes and getting it to “think” this way requires priming the query thread with lots of contextual information before the AI has a hope of generating a truly creative, novel and contextualized hypothesis statement from scratch.

While we found that overall AI wasn’t quite able to generate vivid and novel hypotheses with the ease and dexterity of human workshop participants, prepping it with as much context as possible beforehand prevents wasting precious workshop time and enables better results during the workshop.

Foundationally, this experiment highlighted the importance of not just prepping workshop participants, materials, inputs, and activities in advance, but preparing a conversational AI tool like ChatGPT4.0 in advance as well. Almost like prepping a participant for a workshop with introductory materials, research, and information, we quickly found that in order for ChatGPT to be a contributing member to the group trains of thought which can pivot suddenly and aren’t necessarily linear, we need to prepare it with a wealth of contextual information. It’s not news that optimizing performance of an AI like ChatGPT requires providing it with the maximum amount of contextual information. However, for a live workshop, this step of making sure ChatGPT is up to speed and on the right like of reasoning before attempting to work it alongside a live team is key. It can take time and multiple redirects to get ChatGPT on the same page and able to be an additive voice in the context of a live workshop.

2. If not hypotheses, what was AI good at?

While human participants may have been more adept at quickly connecting information in novel and unlikely ways to come up with initial hypothesis statements, populating all the knowledge gaps that need to be investigated to prove or disprove a hypothesis during the workshop can be time consuming and brain draining.
Here, AI could give us a huge leg up.
Beginning with the starting point of an already crafted hypothesis statement, both ChatGPT and miro were great at expanding with variations on that hypothesis which helped us refine and  generate more nuanced final statements. ChatGPT was also exceptional at populating knowledge gaps for each hypothesis statement where investigation would most likely be needed to prove or disprove the hypothesis.
These aspects of the workshop would normally take up a significant chunk of the live session or have to be backfilled afterwards due to time constraints. Having one or two team members work with AI on these tasks saved us tons of time and enabled the other human participants to spend their valuable time on the more creative tasks. In workshop scenarios with our partners, quickly generating knowledge gaps like this can enable a higher-level conversation with our partner right then and there to collaboratively draw out the most urgent knowledge gaps on which to focus discovery research – a conversation that would normally be separate.

3. Designated AI users or

free for all?

After trying it both ways, we found that it depends.
At least in the context of this experiment, giving everyone access to AI during the workshop was a benefit with these strong caveats. This only works if:

• All AI users have a certain level of proficiency with the tool. Otherwise, it becomes a distraction to the purpose of the workshop and participants spend more time getting the hang of it than generating valuable outputs.
• AI is not relied on as the source for hypothesis statements. Participants should be encouraged in a workflow to craft a hypothesis themselves and then use that as the starting point for an AI query. Otherwise time consuming work is needed to train AI on the contextual premise.
• One way to do this might be to structure a workshops so that participants are encouraged to generate as many novel ideas as possible in a pre-AI segment, and then empowered to elaborate on those ideas with AI in a later segment. With the capabilities of ChatGPT 4.0 today, this structure enables human participants to focus on what they do best, creatively connecting inputs in unlikely ways to generate novel ideas, and allows AI to step in as an aid where it excels best, expanding upon existing ideas with likely predictions.

The Journey Continues

As we wrap up this task, the journey is far from over. We’ve scratched the surface, but the vast landscape of AI tools and their applications lies ahead. We hope you’ll join us on this expedition and together we’ll navigate the evolving landscape of AI, unlocking its potential for innovation and efficiency. Stay tuned for the next chapter in our AI adventure!

We invite you to join us on this adventure by subscribing to our newsletter and following us on Linkedin where we’re posting regular updates on what we’re learning, what shocks us disappoints us, delights us, and will change how we approach innovation. 

Discover more articles on Innovation Strategy and Prototyping