Day 62 — Future of Design series 3/7: “Designing AI”

Original Photo by Joseph Chan on Unsplash

With the rapid growing power of computer processing, and new ways to harnessing it with neural network, artificial intelligence becomes the new hot topic in Digital Transformation. According to Forbes Magazine, 91% of executives say AI will help them outpace their rivals, according to a Forbes Insights survey. As a designer, developer, or product manager, what do we need to know in designing AI services? How can we effectively adopt AI design? In today’s article, I’m going to share my knowledge in the following structure:

• Quick Intro: Types of AIs
• Designing AI vs. Designed by AI
• Designing AI service with HCD approach

Types of AIs

Demystify AI

According to Geoff Colont, the author of Disruptive Marketing, there are mainly 3 types of things people refer to when they are talking about AI:

1. Strong AI: This is usually referred to the AI has capabilities to build other machine and make conscious choice just like human being, rather than a simple simulation. An extreme example could be the scary Skynet in the movie Terminator 2, the AI who decided to destroy human being by trigger nuclear war between Russia & USA. Currently the challenge in technology world for it is that we don’t know enough about human mind as a benchmark to create a simulation that’s as close as human mind
2. Applied AI: It is the type of agentive technology, which process input in a smart way and map the appropriate tasks that need to be implemented. For example, Amazon’s Alexa, Google’s Hey Google, Apple’s Siri. In the financial industry, the algorithmic trading, in which it takes market or other data to decide how the trade/order should be made/filled.
3. Cognitive Simulation: Typically this type of AI utilize cognitive science to process large amount of data or automate certain tasks. For example, Apple’s Face ID, or the security verification system “Gotcha- (I’m not a Robot)” are some of those imagery processing AI.

Image source: Artificial Intelligence Magazine

Designing AI vs. Designed by AI

Although designing AI would be the center focus of this article, it’s probably better to make it clear about the difference between Designing AI vs. Designed by AI.

Designed by AI

What are the things are designed by AI? Computer code? Aerodynamic shapes of jets? Some might have the gut-instinct that AI is good at those engineering focus of design. In fact, more and more AIs are creating aesthetic-focus artifact. Just to name a few:

1. Website design service company WIX uses AI to create new website template for users
2. Alibaba used AI to create thousands of ad banners in a second
3. AI can generate art pieces and it’s competing with human artists

Design Process Powered by AI

With the growing capabilities of AI, we’re seeing more and more aesthetic driven AIs being created. Not just the impact on the deliverables, AI also is facilitating some design process to help designers’ job more efficient. For example, AI can help preparing related assets and content, generate moodboard, finding relevant templates and UI components, and automate export and specification process.

Some of the design activities/processes can be enhanced with AI, for example, Moodboard. Image source: Milanote

To take the concept of “designed by AI” to extreme, there’s even thoughts around AI replacing designers. That is a legitimate concern if we only define designers job as putting UI together in a layout. However, from my personal experience working with so many design talents, great designers are not only problem solvers, they know how to frame the problem to help stakeholder “solve the right problem”. Utilizing user/stakeholder research skills and modeling the data, great designers spend equal amount of time to understand the problem before solving it. As Einstein said, “if I had only one hour to save the world, I would spend fifty-five minutes defining the problem, and only five minutes finding the solution.” That’s the great value of having a great designer on our team, because we know that we can’t expect Machine learning to figure out what problems to solve.

“If I had only one hour to save the world, I would spend fifty-five minutes defining the problem, and only five minutes finding the solution.” — Albert Einstein

Now, back to designing AI related service, as a designer, can we take a proactive role to understand how we can incorporate some new knowledge in regards to AI Design? Why do we need it, and how to do it? There are some knowledge and experience I can share with you.

Image source: Data Driven Investor

Designing AI service with HCD approach

Even though AI design process sounds like a big black box that’s unimaginable to many, at the end of the day if it’s designed for serving human, it still shares the common challenges of bridging the gap between human and a system. For example:

• What’s users’ mental model? What’s the gap between the system process, users’ mental model, and designer/developer’s mental model?
• How would the system allow/react-to the user input? Can the system process those input effectively, also in a way that’s aligned with what users are expected?
• How does the system present the feedback in an intuitive way?
• What kind of actions can the system execute on behalf of users? Are they what users expected? How do we know if users are satisfied with the results?
• When there’s error and other “unhappy path” use case, how does the system handle it?

Compare ideas in the matrix: User impact vs. Machine Learning impact. Image source: Google Design

So, if we’re designing an AI for human, the fundamentals are not that much different. However, there are some AI specific scenarios that’s unique and worth addressing in different ways. Just to name a few:

1. Labeling: One of the key factors of how AI works is the labeling. Labeling means the AI will tag the content based on the input it received (e.g. I “thumb down” a movie on Netflix, or 70% of the show on my Netflix List is Sci-Fi). In order to help design the AI to be efficient solving user’s problem, UX designers can utilize our Information Architecture skills to map the correct label from user’s mental model to feed into AI system. For example,designers can run open cart sorting exercise to understand how users would label those content, and how they would structure the content; these are invaluable data for AI to be based on.
2. Defining Success: Everyone has different taste, different preferences; It would be hard for AI to “guess” if we don’t let them know and learn from experience, even we don’t tell the AI that it’s guessing it right or wrong. Having a way for users to provide input to the AI is imperative in AI’s machine learning process. Not only UX designers can help create a effective and non-intrusive way for users to provide input, but also the early research about how do users define success, and at what stage users feel comfortable giving that feedback, and in what fashion.
3. User Goal: There are many things an AI can process and provide output, but it’s hard for AI or engineer to know what kind of output users are looking for without deep understanding through user research. Using the voice UI paring with machine learning as an example , when users say “give me an update about Apple”, does the user want to know the stock price? or they are Apple product fan who want to know when Apple will release the new iPhone? Or they are asking about the apple farm they have” It’s important to have the design team to engage user early on to understand users intention, so that we can determine what kind of output is valuable and what not.
4. Multi Channels: In case of AI isn’t able to provide appropriate solutions to suit users need, always design a plan B that’s a sure-fire, reliable fall-back plan so that users don’t get stuck because of the incompetent AI. When we design this multi-channel solution, always reminding users which channel/content they are in, and help them understand what they can expect to happen next.
5. Change Update: Given the speed of update in machine learning is continuous and fast, it’s always a great idea to keep users in the loop of what’s going on, especially if we remove certain functions or pathway, so that we don’t confuse them with the new update (which is a common pain point whether it’s an AI app or not)

Human Centered AI Canvas. Image source: Artificial Intelligence

Resource

Introducing the Human-Centered AI Canvas

Guidelines for Human-AI Interaction

Human-Centered Machine Learning

Conclusion

1. Designing AI and designed by AI are two very different areas, and both of them are seeing strong potential growths;
2. The general principles of designing AI related service is not that different the designing any other complex system. The key difference is the specific nature of “AI’s rapid learning process and the required feedback loop”;
3. When designing AI related services, always think about a “Emergency Exit” or “Plan B” in case the AI fails its job.

Do you have any experience designing AI related applications? What was your approach, and what did you learn? I’m eager to learn from you.

ABC. Always be clappin’.

To see more

All Daily Agile UX tips

Day 61 — Future of Design series 2/7: “Voice UI”

Day 60 — Future of Design series 1/7: “Service Design & Innovation”

The opinions expressed in this article are those of the author. They do not represent current or previous client or employer views.