Design thinking supported by generative AI: How far can we go?

Stefan Agamanolis, associate director of strategic research initiatives, Weatherhead School of Management

In the rapidly evolving landscape of product and service development, generative AI tools are already having a transformative effect throughout the design process.  

In early process stages, LLMs (Large Language Models) like ChatGPT can reduce the time needed to create things like interview discussion guides and surveys with appropriately worded questions.  The same tools can help analyze data and create user personas and journey maps.  The value of generative AI in ideation exercises is already well recognized, and these models can also be applied in the tasks of pattern detection and cluster creation that can be jumping off points for further brainstorming.  Image generation systems like DALL-E can greatly speed the process of creating product sketches and interface wireframes, and indeed go far beyond these traditional steps with photorealistic renderings or polished storyboards that generate richer feedback when exposed to potential customers.  

The outcome of a design process is frequently intended for human use, aiming to significantly influence and enhance the quality of human lives.  And so we actively seek insights and contributions from humans at every stage of the design process, from initial discovery and research to gathering feedback from pilot programs and tests.  However, it is often difficult to achieve a desired volume of human input, or to achieve a suitable diversity and depth of perspectives, for a variety of reasons.  Time and resource constraints are common and it is all too easy to cut corners in the selection and involvement of human participants.  Could generative AI be helpful in filling these gaps?

For example, could you use ChatGPT to not only generate an interview guide, but also conduct the interview itself, leveraging ChatGPT’s ability to play roles, which can be based on detailed design personas presented as part of the prompt?  Could you have these personas vote on or otherwise react to individual ideas emerging from an ideation exercise, helping a designer arrive more quickly at a prioritized list of the most promising ideas?  Could you expose solution specifications, features, and visuals to these personas in a virtual product test and ask each persona, and variations thereof, to fill out your feedback survey?

Replacing real human input with generative AI in this fashion is provocative to say the least.  The most advanced LLMs display a breathtaking ability to channel the full breadth and depth of human experience, including human emotion and cultural nuances, which is no surprise considering the vast library of human creative products these systems have been trained on.  Nevertheless, several potential concerns exist:  Would an interview of a persona in ChatGPT result in stale insights because of a training date cutoff?  Will researchers miss unexpected anecdotes that ChatGPT can’t summon due to insufficiently detailed prompts or gaps and bias in training data (for example around the specifics of existing solutions or competing products)?  Could one of those missed anecdotes have been the spark of the next genius product feature?  

These are very active research questions, and any perceived deficiencies stand to be mitigated rapidly as generative AI systems improve.  And if today’s systems don’t suffice to support every aspect of the design process, they can certainly assist in accelerating it, addressing common gaps, and increasing confidence in outputs.  One thing is certain, the integration of generative AI into the design thinking process offers a frontier brimming with potential, promising to revolutionize how we approach virtually everything in product development.

Join me on May 29th for my one-day Weatherhead Executive Education course “Leveraging Generative AI in Product Development”, co-designed with Mike Fisher, former CTO of Etsy.