⚡ Challenges UX designers face when designing AI experiences
_________________________________________________________________
PSA: It is a key value for us at AIxDESIGN to open-source our work and research. The forced paywalls here have led us to stop using Medium so while you can still read the article below, future writings & resources will be published on other platforms. Learn more at aixdesign.co or come hang with us on any of our other channels. Hope to see you there 👋
The Empowerment of Designers in Software
Artificial Intelligence (AI) and its promising sub-field of deep learning are taking over the world by storm. However, why care about designers and AI? Isn’t AI a technical discipline for machine learning engineers? To answer these questions, let’s look at software development for a moment.
Software development used to be in the hands of engineers. Engineering managers would define a set of technical and functional requirements. A group of engineers would take these requirements and turn them into a fully functional and scalable software product. This “waterfall” approach came with a cost: Many resources were spent but products failed to reach product market fit. Consequently, industry and research started to derive better methodologies for software development — concepts such as “lean”, “agile”, and “user-centricity” came to light. In 1993, Scrum formalized an agile development process that encourages quick launches and continuous testing. In the 2000s, design thinking had its breakthrough. Historically, designers tended to be involved only in later parts of the product development process, focusing their attention on aesthetics (“Can you make this pretty?”). However, many businesses realized the enormous value of embedding design at the center of their innovation process. The “double diamond“ was born — today’s best practice for how we discover, define, develop and deliver innovative software solutions in a user-centered fashion.
Design Challenges for Human-AI Interactions
So what’s different for AI-powered products? And is it really harder for designers to work out pleasurable and useful Human-AI interactions?
Prior to AI, software products used to be fully deterministic. Platforms such as Netflix for example allow you to watch movies and TV shows in just a few clicks. Each “click” brings you to a new page — there are little surprises. AI transforms the user experience toward a probabilistic, dynamic, and evolving nature. In some cases, Netflix’s recommendation system may help you in finding the content you like based on your previous interactions. In other cases, the AI-powered system may also suggest movies with high confidence that don’t interest you at all. The probabilistic nature of AI makes errors an inevitable byproduct designers need to deal with. Designers are encouraged to anticipate AI failures in early design phases and provide users with a path forward from failures. However, because product designers commonly don’t have a strong technical background, they may have difficulties understanding the capabilities and limitations of machine learning.
“We designers do not understand the limits of machine learning and what it can/can’t do. Machine learning experts often complain to me that designers act like you can just sprinkle some data science onto a design and it will become automatically magical.” [2]
In addition, the system behaves differently for each and every user. A prototype would need to account for various user behaviors and scenarios if we wanted to test a dynamically behaving recommendation system for example. To capture the true AI experience in a prototype we would need to include real input data and the actual model somehow. Prototyping tools such as Figma, InVision, or Sketch don’t allow this as of today. To address the dynamic nature of AI, many designers today rely on Wizard of Oz (WoZ) techniques for testing AI experiences with end-users. The WoZ technique has a user interacting with an interface without knowing that the responses are being generated by a human, not an AI. In other words, you trick the user into believing there is an AI but in reality, someone behind-the-scenes is pulling the levers and flipping the switches. This works well in some cases, but WoZ can be overly optimistic and easily overlook important details of AI implementation. Furthermore, designers often avoid making the same mistakes that an AI would make and thus fail to achieve a realistic error representation during testing.
“Making interactive prototypes that incorporates machine learning is hard (haven’t found a way to do that yet in an easy fashion)” [2]
Lastly, in contrast to traditional software products, AI-powered user experiences are evolving over time. Netflix’s recommendation system was far from perfect after its first launch. However, feeding the system with more data made the algorithm better and better. It can be difficult to communicate this evolving nature of AI to end-users, especially in high or medium-stake situations. Imagine you are a designer of a computer vision system that supports radiologists in differentiating between malignant and healthy breast scans. Such breast cancer screening software will never be perfect especially not in its early days. But imagine you having to explain this to a doctor when life is literally at stake. You may also wonder how much trust domain experts should put into the system. In theory, there are several design guidelines such as Google’s People + AI Research Guidebook or Microsoft’s Guidelines on Human-AI Interaction that can help designers answer these tricky questions. However, in practice, these recommendations only provide high-level guidance but lack actionable details.
“If [the AI] makes a (grievous) error, who is held accountable? ….can it be trusted to make decisions or take actions on its own?” [2]
A human-centered Design Process in AI
A promising approach is a human-centered design approach to AI — a process that puts the human at the center in all steps of the development process. This includes our values, ethics and emotions. Designers, UX researchers and product managers tend to represent the voice of the user. Hence, for us to achieve a human-centered approach to AI it will be of extreme importance to empower designers to play a more proactive and dominant role in the AI development process.
“While machine learning as an enabling technology is still in the early stages, we’re likely to be one step behind the engineers that create it… changing this relationship to being one that is design-led, or at least an equal partnership will be important — we need to shift the conversation from technology to people — we’ll need to bring the ethical and human-centered voice to the algorithms that make it all a reality.” [2]
Closing remarks
UX researchers and designers face several challenges when designing AI-powered solutions: It can be difficult to express AI design ideas, foresee potential effects of the AI or make sure the AI is not creepy. Let’s empower UX designers with tools and methodologies to improve the AI experience design process.
Credits 👏
[1] Qian Yang, Aaron Steinfeld, Carolyn Rosé, and John Zimmerman. 2020. Re-examining Whether, Why, and How Human-AI Interaction Is Uniquely Difficult to Design. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (CHI ‘20). Association for Computing Machinery, New York, NY, USA, 1–13. https://doi.org/10.1145/3313831.3376301
[2] Graham Dove, Kim Halskov, Jodi Forlizzi, and John Zimmerman. 2017. UX Design Innovation: Challenges for Working with Machine Learning as a Design Material. In Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems (CHI ‘17). Association for Computing Machinery, New York, NY, USA, 278–288. https://doi.org/10.1145/3025453.3025739
