Hey design manager: Do you know what AI means for you and your team yet?
As your design team explores how artificial intelligence (AI) will enhance your products and services, there are a few things to consider. Read on for five tips on how to fast-track your design team to be better at designing AI-infused services and experiences.
Ever since the transition to the experience economy, the connection between service platform and user experience has pushed organizations to actively focus on their customers’ complete service experience beyond concrete and tangible product and interface design. To do this, designers create design tools that help them visualize, ideate, communicate, and validate their designs and the relation to user experience.
However, data-driven and AI-infused services is a new territory for designers. The tools that make up the typical toolset of UX and Service Designers — such as Service Blueprints (Shostack, 1984), Customer Journey Maps (Nenonen et al., 2008; Stickdorn et al., 2018), Business Model Canvas (Osterwalder & Pigneur, 2010), Personas and Scenarios (Cooper et al., 2014), and Ecosystem Maps (Polaine et al., 2013) — are not explicitly tailored to describe user experience, impact, and value for services that rely on AI.
Even though decades of work in the space of AI-powered services have produced several sets of guidelines (e.g. Horvitz, 1999; Yang et al., 2018; Amershi et al., 2019), it is clear that more specific tools are still needed. Current wireframing and prototyping tools do not support user interfaces where personalization and other evolutionary adaptations are at the core of the user experience and develop over time. New design materials often require new tools and the characteristics of AI technology and computation in general, and Machine Learning in particular, might well be considered a “design material” of its own (Dove et al., 2017).
So what’s new, and what’s different?
First of all, AI is tricky to define. It’s a misconception that AI is imitating human intelligence — at least for practical purposes. What we usually mean when we talk about practical AI today, is typically Machine Learning (i.e. the ability to predict or classify previously unseen data based on a model generated from training data input). Machine Learning (ML) is, so to speak, becoming its own thing as it gets more mature. This is typical for AI research. One of the pioneers of modern AI research, John McCarthy, said: ”As soon as it works, no-one calls it AI anymore.”
Machine Learning (ML) is different from traditional, deterministic systems because it takes us to new territory.
Here are some interesting trajectories for your design team to think about when it comes to how your products and services will work:
From rules to models
An ML system builds its own internal model for prediction based on batches of training data. This means that rules are not explicitly coded into a ”program”. The “rules” become part of the model that processes new input, but is not inspectable by humans. This is what the field of Explainable AI deals with. What this means for us is that a designer cannot simply walk up to a programmer and discuss the logical rules of the application in the traditional sense. Because there are no explicit rules to talk about, and the programmer can usually not hand-code exceptions without decreasing the power of the evolving ML model.
From fail by error to fail by design
An ML output will always consist of false positives and false negatives. Eliminating errors completely is impossible. (If you understand the domain well enough to be able to do that, you actually won’t need ML!) Therefore, designers need to explicitly design for failure of these models. That requires risk mitigation strategies and domain-specific reasoning about what different consequences of these errors can be for the user. In some cases a false negative can have a very small impact (such as recommending a song that the user will not like). In other cases it could be fatal (such as in medical diagnosis). Discovering the effects of such erroneous output should be explored and planned for as early as possible.
From consistency to adaptivity over time
As a hallmark of great tool design, we have been taught that consistent and predictable behavior is key. This changes in ML services. It follows from the items above, that certain ”golden rules” of usability and UX will need to be modified when we enter the realm of complex adaptive services powered by ML models. Heuristics for evaluating positive impact of AI-infused interactions need updating.
These are just some characteristics that allow us to start re-framing what AI-powered digital services are, and how we build and deploy them.
From “tool” to “butler”
One way to re-frame what an AI-infused service is, is to think of it as a system capable of doing things on its own, according to our preferences. This is what Chris Noessel calls the ”agentive” perspective. We are in effect re-framing a digital tool as more of a butler. This is very compatible with a Service Design approach, where we tend to put a ”use over time”-perspective on digital services, and view service as a network of different actors (sometimes called an ecosystem). With the advent of agentive services, they can actually rise from being seen as ”a channel” to ”an actor”. Tools that model existing holistic and strategic Service Design tools to accommodate AI actors as part of the overall user or customer experience — and relate them to the backstage operations of the service provider organization — is lacking.
Prototyping and Evaluation
However, not all AI-powered services need to be semi-autonomous butlers. There are low-hanging fruits even in simpler AI functionality: such as recommending new songs in a music streaming service, or suggesting energy-saving activities for your home in an electricity-monitoring app, etc. But even in those cases, current tools fall short. This is challenging if you’re running a tight agile iterative development cycle. Since the service capabilities evolve over time and changes as soon as users provide new input it becomes hard to wireframe, prototype and evaluate such personalized functionality in an authentic way.
Strategies for data collection
In the case of evolving adaptive systems, we need to continuously feed it with data for re-training. This collection is not trivial. To name a few aspects, it needs to be collected at use-time, and therefore be designed in a way that does not disturb the user’s workflow and experience. It needs to be ethical so that the user is informed about what data is collected and how it is used. And the fact that new data can give rise to unanticipated system behavior needs to be taken into account as well. Lastly, data is collected and maintained from several different sources, so the design effort needs to be in sync with the overall data management strategy.
So if you want to harness the benefits of Machine Learning and strengthen your ability to design thoughtful and valuable AI-powered services, you need to make sure your designers know how to navigate this new territory.
What can you do?
Given this, what are the challenges for UX and Service Designers who want to shape AI-powered services (Yang et al., 2020), and how can you work with them systematically? Help your designers and developers to:
1. Understand AI capabilities: Take courses and allow time for training and learning. Allow designers to get experience and play with tools like TensorFlow or IBM Watson API:s, etc. And better yet, go beyond these tools and have them learn basic data analysis and machine learning. (I know, we’re getting back to the age-old discussion “should designers be able to code or not?” Let’s just say that if you know how an ML model is generated and how it works, you will probably be slightly better off than if you don’t.)
2. Envision novel feasible AI for UX problems: Knowing and understanding AI is not the same thing as being able to ideate successfully. Learning and creating new design patterns is very valuable. Otherwise, we run the risk of defaulting to well-known, but basic, patterns such as product recommendations. We have only scratched the surface of what can be done with data-driven prediction and deep learning, so keep on exploring and learning from others.
3. Prototype and evaluate: Get up to speed with the latest approaches and methods for prototyping and evaluating evolving adaptive systems, and how ”butlers” rather than ”tools” are used and experienced.
4. Understand long-term impact: Strategic UX and Service Design is key here. Your strategists need to talk to Data Scientists and think carefully and hard about long-term unintended (and intended) consequences. AI is not only about adaptive user interface design. It’s also about how AI can affect and completely disrupt your frontstage operations and backstage processes, as well as your entire business model.
5. Enhance interdisciplinary communication: Learn how to collaborate with data scientists, AI engineers, and developers. Understand that each profession use their own language, and has been schooled in different traditions. We have, as a field, developed ways to communicate between designers and coders during the last decades. But with the advent of ML and Data Science jargon, we’re back at the stage where we need to learn how to communicate again. It’s not enough for a designer to know basic developer terminology anymore. You have to quickly understand why and how supervised learning and unsupervised learning will affect the user experience. You will also need to develop a ”sense of” whether or not the currently available data is enough or of a good enough quality.
In order to do this you need both internal training, and possibly hire new competencies (such as data scientists, or one of the few human-centered designers that understands AI), as well as accumulating experience by playing and prototyping with AI as a design material. As a field, design needs collections of great design patterns and best-practices for UX and Service Design of AI-infused services. Be relentless in your learning journey, and accumulate and share as much of what you learn as possible. By growing the shared knowledge in our community, we can develop better services and experiences for all.
Go play. Happy learning.
