Privacy-first principles: Get $10 credit today with Fathom Analytics. Find out more.

Principles of Design with Watson

Basic tenets that can be applied when working with IBM Watson — or any AI system and how to think through using them.

  1. Know your data.

    Become best friends with the dev and data scientists on the team. Engineers know the APIs like designers know grids. Build culture and trust. Keep teams lean and consistent. Put your confidence in them, and they will put theirs in you. Together you’ll guide your team to narrow your focus on achievable realities based on your current data — or a hypothesis (in data science lingo wink)

  2. Design for honesty, not omnipotence.

    We need to know what Watson is telling our user — don’t be confident and wrong. AI projects don’t have “normal” bugs to solve. We have to design to solve for the user’s expectations. If Watson’s confidence is low, we need to make the user aware of that. The answer will likely not correlate with what the user expects. We need to take a step back as a designer — how would I answer a question to a friend about something I don’t know: “I’m not sure how right this is but…maybe x, y, and z?” , or point the person to someone who would know. We need to assume and design a fail state or a pivot point for to happen.

  3. Show — don’t just do.

    We have to open the door, let users peek into the black box of AI and let them know how Watson made that decision.
    The idea is similar to the concept of progressive disclosure, but we are applying it using evidence and data points. There is a cooperative aspect to these experiences we need to enable. Showing what Watson knows will help users feel confident in Watson’s decisions but also allows them to course correct or improve Watson’s knowledge at any time. Watson’s confidence, relevance, and correlation directly impact the user’s view of the system — by improving domain knowledge we can build trust.

  4. Deliver insights, not visualizations.

    Visualizations within the experience should address misconceptions, add to the end user outcome, and make the data easier to comprehend — visualizations are never the deliverable. The deliverable is an insight within it — but never a visualization alone. Watson is amazing when it is finding patterns and teasing those out in massive amounts of data — what we have to keep in mind is what our users need to take away — we are leading them to the needle in the noisy data haystack.

  5. Always in service of people never in lieu of them.

    The user and Watson are partners working towards mutual goals through an experience. We have to solve the user’s need which is central to success but we also have to enable human and computer cooperation — not only interaction. Building empathy, trust, and cooperation with our users is the goal. We should be focused on removing limitations and addressing pain points. Watson never makes a decision for a person — it finds, augment and educates.