AI Guidelines in the Creative Process

How we’re putting the Human-AI Guidelines into practice at Microsoft

By Penny Collisson, Saleema Amershi, Mihaela Vorvoreanu. Illustrations by Michaelvincent Santos.

We recently published the Guidelines for Human-AI Interaction. These eighteen guidelines represent best practices for AI design. They provide guidance on how to create meaningful AI-infused experiences that leave users feeling in control, and respect their values, goals, and attention. For example, they prompt us to make clear how well the system can do what it can do (guideline 02) and make clear why the system did what it did (guideline 11) to help users navigate AI-infused experiences that may be inaccurate.

In the Microsoft Office product group, we’ve started to build these guidelines into our work. For example, our teams perform heuristic evaluations using a handy deck of AI guideline cards (with examples) we created for the guidelines. We shape usability tasks and discussion guides with the guidelines in mind. We use the guidelines to bolster our case when sharing back research findings.

We recently started to explore how the guidelines can play a role throughout the entire design process. That is, can the guidelines inspire creative and human-centered thinking, in addition to being a tool for product evaluation and refinement? If so, how? Today we’ll share a few ideas.

#1. Decide on a few guidelines to rally around.

Focusing on eighteen guidelines at once may seem like a daunting task. It also might not invoke the creative mindset you’re after in early design stages. So, consider ways to bring focus:

• Agree on a few guidelines that are most important for your experience. Time services based on context (guideline 3), for example, might be especially relevant for experiences that include notifications.
• Pick one of the four guideline categories (i.e., Initially, During Interaction, When Wrong, or Over Time) to ideate around first.
• Assign sponsors to different guidelines. Ask those people to be on the lookout for new opportunities or violations, and to track what they observe in the product backlog.

The key is that you can create a different mindset simply by managing attention in creative ways.

#2. Explore (don’t avoid) violations.

It’s natural to want to brush aside guidelines you aren’t intending to follow. After all, things that might go wrong are stressful, and stress is the enemy of creativity. Instead, explore violations. Doing so can help your team think outside the box. Try the following:

• Decide on a few guidelines to set aside or even deviate from and discuss why you will do that.
• Outline the range of feedback you might hear. For example, what’s the worst and best feedback you might expect if your AI doesn’t remember recent interactions (guideline 12). How would you optimize the design in response to that feedback?
• Do research with products on the market that violate different guidelines. For example, identify products that don’t mitigate social biases (guideline 06). Watch what people do, understand how they feel, and listen to what they ask for.

Actively exploring violations might unlock new and creative ideas.

#3. Mash up different approaches in design jams.

Design jams can be an awesome way to get a breadth of creative ideas for solving human-centered problems. Why not leverage the AI guidelines as a stimulus in your next jam? Here are some approaches:

• Have each of your team members optimize the experience for a single guideline. What are all the ways you could scope services when in doubt (guideline 10), when you aren’t clear on the user’s intent, for example?
• Next, push people in the opposite direction — what does a terrible experience look like for this guideline in your AI? The juxtaposition of great and terrible design ideas may reveal new creative opportunities and pitfalls.
• Ask people to create different mash-ups of the concepts produced. Integrating ideas — even seemingly opposing ones — can unlock new ways of thinking about your AI’s design.

The act of bringing together different perspectives for approaching the guidelines can help your team get to truly innovative solutions.

#4. Embrace tensions between guidelines.

It’s easy to learn a set of “rules” and see them as that, rules. But these aren’t AI rules, they’re AI guidelines. That means you break them, play with them, and even intermix them. Don’t avoid places where the AI guidelines conflict with each other, or even with classic interaction guidelines. Dive headfirst into any tensions you can find and use them to create energy. For example:

• How do you update and adapt cautiously (guideline 14) the AI system’s behaviors, while also creating an experience that is predictable?
• How do you encourage granular feedback (guideline 15) during regular interaction, while also achieving a minimalist design?

Creative power often comes from having to consider two ideas that — at least at face value — seem to be in conflict.

#5. Use prototypes to gather new perspectives.

Prototypes can be used to build empathy and even generate new ideas. Consider ways to use the guidelines, and concepts you draw to solve for them, to focus your next prototype research. Try this:

• Do a co-creation activity using your early concepts as a stimulus. For example, take concepts that make clear why the system did what it did (guideline 11), and ask people to “redraw” it in a way that leaves them feeling more in control.
• Explore reactions to a breadth of different concepts, and get people talking about why they like what they do. Your team can ideate on new concepts that optimize for all the positive themes. For example, take concepts that convey the consequences of user actions (guideline 16) into research, and evaluate what elements help most with comprehension.

Your initial ways of solving for the guidelines can generate conversations with customers, and even more ideas.

Talk with us.

These are things we’ve been recommending to other people innovating with AI experiences at Microsoft, but we’re sure there are other ideas out there. What do you think? Are there other tactics you’re trying out with the guidelines?

If you’re up for it, share this article with your teams and start a conversation. Or with your network and start a debate. We’d love to hear about your experiences in the comments. You can also catch Penny, Saleema, Mihaela, and Michaelvincent on LinkedIn.

Authors.

Saleema Amershi is a researcher working on human-AI interaction at Microsoft Research AI.

Mihaela Vorvoreanu is a program manager working on human-AI interaction at Microsoft Research AI.

Michaelvincent Santos is a designer working on collaboration in Office.

Penny Collisson is a user research manager working on AI in Office.

With special thanks to Gwenyth Hardiman, Ruth Kikin-Gil, Trish Miner, Ken Milne, Meghan Stockdale, and Benjamin Aleshire for their teamwork and contributions.

Also with thanks to our team who developed The Guidelines for Human-AI Interaction: Saleema Amershi, Dan Weld, Mihaela Vorvoreanu, Adam Fourney, Besmira Nushi, Penny Collisson, Jina Suh, Shamsi Iqbal, Paul Bennett, Kori Inkpen, Jaime Teevan, Ruth Kikin-Gil, and Eric Horvitz.