How we used a little fun to get serious about AI at Color

Sarah Turrin
Color
Published in
4 min readMar 18, 2024

I may be biased, but I believe our product and engineering teams here at Color are among the most conscientious and capable in the digital health space today. Our team members deeply value doing right by our participants and patients, respecting their privacy and security, and building experiences (digital, telephonic, or IRL) that are accessible to all.

So when generative AI made its way into the tech zeitgeist in 2023, our team was understandably cautious. We were (and are) optimistic and excited about the incredible possibilities that this technology brings, but in a sensitive and highly regulated environment, we were hesitant to jump in. After fits and starts of conversation, but never really moving beyond discussion, we knew we needed a different approach.

As anyone with a toddler can tell you, the antidote we found was time-tested and parent-approved: we decided to play. All things are easier to learn through play, and all problems are easier to solve with a little levity. We kicked things off in December 2023 with our holiday themed AI hackathon: HolidAI Hacks. The agenda was absurd and awesome, the perfect vibe for the week before our company-wide holiday break.

  • Winter Wonderland of AI Tools: an overview of new AI tools on the market, and which ones we were making available to our teams.
  • Merry Prompt-Making: A context to prompt engineer the jolliest futuristic AI art. The challenge: Create a piece of art that shows your ideal holiday tradition 50 years in the future.

A few of my favorites:

Prompt: 4 different families crowding around a table with a tablecloth with the artwork of a wong kar wai film and plated with a panoply of dishes like roast duck with crispy skin flaking off, the kids are floating above on a magic carpet, there is a robot in the kitchen preparing the next dish, and there is a 400 year old lady who has a glass ball for an eye.

Prompt: A futuristic Christmas scene in Southern Los Angeles, set in 2070, with several Pomeranians enjoying the holiday

  • Candlelit Compliance: Using AI Responsibly: A fireside chat with our legal team to ensure that everyone knew the guardrails around using PHI, proprietary information and other sensitive data in AI tools.
  • AI-generated holiday music written specifically for Color, including the would-be hits “Tech and Tinsel, a Merry Blend” and “In Color’s Halls Where The Bright Screens Shine”
  • An AI-generated video deepfake of our Hackathon Architect and MC, Arfa Rehman

Real Arfa with Deepfake Arfa

  • Mandatory pun-ny names for every hack (e.g., Sleighin the Patient Summary)

But where we ended up when the hacking was over and the judging was complete wasn’t absurd at all — our teams had a much deeper understanding of what AI could do for them and for our patients (we know this — we did a poll!). And most important, we walked away with the start of some real life investments in our product that have the potential to improve our internal operations and the experience of our patients.

A few that we’ll discuss in more detail in posts to come:

  • An AI-generated patient summary to prepare our providers with all the context needed for their visit, a huge win for both clinical efficiency and patient care.
  • An AI tool to assist clinicians in identifying potential drug interactions using pharmacogenetic data.
  • A slack-based GPT assistant trained on Color Medical’s unique clinical protocols
  • A prototype of an AI calling assistant to help our Care Advocacy schedule cancer screenings for our patients
  • A custom GPT trained on behavioral design best practices to assist help us apply behavioral economics to product design
  • A release risk assessment tool to identify potential issues with soon-to-be released code

What follows here will be a map of our AI playground: the hacks we tried and failed to get off the ground, the ones we nailed and worked to productionalize, and the ones we did just for fun. All of them are instrumental to our understanding of generative AI, building our intuition and judgment about the most valuable ways to leverage AI in our business, and how to do it well when safety, privacy and ethics are paramount.

--

--