The debut of ChatGPT has been a watershed moment for generative AI. ChatGPT has seen record growth and is gaining widespread adoption. On the flip side, recent fiascos involving Google’s Bard and Microsoft’s Bing highlight the limitations of the technology and raise questions about the model’s cognitive processes. Still, the vast potential and practicality of this model are undeniable.
Not one to sit on our hands during such a pivotal moment, our team at Free Association has been actively exploring the potential applications of this technology. We used our North Star envisioning process to uncover the possibilities of how Large Language Models (LLM) can be integrated into the products we use daily to unlock new user features and benefits.
Rather than philosophize about it, we did what we always do — learn through making.
To kickoff, we conducted a round of user research to think through interesting ways LLM could affect a product experience. We came to realize that any product with a search bar is an immediate candidate for LLM integration. Taking this into consideration, we narrowed our focus to products where search was the primary entry point. However, we didn’t want to limit our thinking to traditional search functions, so we decided to think of a search field more broadly as a mechanism for conversational user reciprocity. This approach allowed us to explore the full potential of LLM and its ability to enhance and improvise on user input, regardless of its original purpose.
Here are some of our highlights —
Spotify: Closer to the Music
Spotify’s vision statement outlines the company’s ambition as becoming a cultural platform where “everyone can enjoy an immersive artistic experience that enables us to empathize with each other and to feel part of a greater whole.” Using this as a springboard for our experimentation, we began to envision a Spotify that transcends the boundaries of music streaming and, through the power of LLM, can bring users closer to the music, the artists and the community.
Spotify already has a sophisticated personal graph of a user’s musical tastes and preferences. With Spotify’s ambition to bring users closer to the music, we envision integrating LLM into the experience in a manner that provides an enriching music-centered experience that goes beyond streaming.
Picture waking up and visiting Spotify to find personalized content tailored specifically to you, providing insight into the music you stream. This could include details such as the origin of the band’s name, the backstory behind the album, the significance of the lyrics, the tumultuous relationship between band members, and the historical context surrounding the recording. Our goal is to provide you with a more profound understanding and appreciation of the music you love.
This Spotify LLM integration is a gateway to ever deeper discovery and connection to the music. For example here are some novel prompts we envision:
- Compare and contrast the two most popular covers of this song
- Teach me the history of this music genre
- Show me people to follow who also enjoy this music
- Build a playlist based off of this thread
How would you enjoy getting closer to the music?
Expedia: Embracing the Messy Middle of Travel Planning
If you’re anything like us, you would love a more organized and integrated tool to cut down on the endless research needed to plan a trip. Today’s online platforms are largely geared towards two things — booking your trip and upselling the add-ons. Building complex plans starts to prove extremely difficult and when people want to collaborate on these plans, forget about it!
We challenged ourselves, “With the integration of LLM features, might we be able to offer users a seamless experience where one could satisfy their discovery, planning and booking journey without ever having to open a second tab or spreadsheet?”
To provide us with sufficient complexity, we established the persona of a female solo traveler who is heading to Japan in May for a wedding. She enters this into the search bar along with the question “What would be a great travel itinerary for a solo female traveler after the wedding?”
Okay we have the blueprint! Now let’s hone in on Kyoto and hotels. A second panel appears with a list of highly rated hotels that would suit a solo traveler. It’s not bad, but “what about something with a bit more of an artistic vibe?” — the panel refreshes along with the map and she can begin to save suggestions as a working itinerary.
With winning ideas favorited, the traveler can now request they be turned into a bookable itinerary.
Throughout the planning process the LLM isn’t the star of the show, the travel is. Unlike today’s Expedia, this one operates like an unobtrusive concierge — letting users pick and choose how to incorporate their passions into their plans.
Where would you take your next adventure?
Waze: Enjoying the Time Between A and B
Waze takes a different approach to navigation, offering a personality-forward experience with the ability to theme a drive. (Full disclosure: Waze is a client partner, and we’ve worked on the UX of theming.)
Inspired by Waze, we realized that there’s a difference between efficient navigation and personalized navigation. Traditional navigation provides a straightforward, one-dimensional experience, while personalized navigation allows users to get there in a way that adds richness to the journey — whether that means taking a more scenic route, enjoying more entertainment, or learning about their surroundings as they travel.
According to data from the FHA, the average American driver logs approximately 225 hours per year. Yet we spend those hours as if in a dream, passing by the world outside but barely processing it. What if we could use this time to connect more deeply with our environment? Imagine you are on trip upstate. You’ve made the drive dozens of times but realize you know little about the history of the area. With our envisioned generative AI integration, all you need to do is ask —
With this integration, all forms of travel can be enriched with contextual information through Waze— walks, bike rides and beyond.
What cool history is hiding behind your commute?
Managing for Risk
Brands beware! There are inherent risks when venturing into generative content. For instance, when collaborating with Chat GPT to generate the Colchester content in the Waze concept, we discovered that even after several rounds, there were still instances of historical bias present in the output, no matter how nuanced. This is a natural limitation of the technology, and occasional inaccuracies or hallucinations may also occur.
The topic of managing risk when integrating LLM is a hot topic and there’s more to unpack here than can fit neatly into this article. However, one strategy is to establish a distinct identity for an LLM chat integration in order to help users easily recognize it and understand its specific capabilities and limitations. Additionally, we strongly recommend informing users about the technology’s characteristics so they can make informed decisions about their use of the technology.
All that said, we are enthusiastic about the potential advantages that LLM can bring in transforming our familiar products, making them more aligned with our evolving mindsets and language, and broadening the scope of how a brand can fulfill its core commitment.
- Sunil Manchikanti — Product Designer
- Chris Kaufman — Product Designer
- Alexa Liley — Design Director
- David Landa — Design Director
- Carlos Ancalmo — Storyboards