Three challenges AI creates for UX designers

Following the recent development of machine learning capabilities, both high-profiles individuals (Elon Musk, Stephen Hawking) and research center (Pew Research Center) have warned about the threat AI (Artificial Intelligence) represents for more and more job categories. I’ve read a number of articles on this likely disruption, mostly trying to explain how humans can stay relevant in a future AI-dominated world. Some of these were related to how UX will still be essential. In particular, this article from Caroline Sinders advocates the need for UX designers to help users understand AI. I also believe designers should take on a much larger role in shaping the AI revolution.

#1 Design operational mental models for machine learning

The machine learning field is still in its infancy, even if the pace of adoption in industries is accelerating. As UX designers, we have a role to play so machine learning systems become accessible for users. By accessibility, I mean that users are not only able to widely access the results of machine learning algorithms, but have a degree of autonomy in their construction and usage. For this, you need to have a basic understanding of how things work, and a way to interact meaningfully with the algorithms.

What kind of interactions could a user have with a machine learning algorithms once they understand how it works? They might choose the initial training set or the tuning set, select a specific algorithm based on the success criteria or reward system it was based on. They might choose to grant access to only a part of their own data. All of this certainly looks very complex, but it’s essential if we want users to retain a degree of agency, a capacity to act and think independently, so they are truly empowered thanks to AI, not just manipulated. It’s the job of UX designers to make it look simple and intuitive.

Designing such a mental model is not a simple task. The most common interface currently used is the virtual assistant. This interface hides the complexity very well, but the mental model it suggests (a person) is inadequate. Machine learning algorithms might mimic certain aspects of “intelligence” but are nothing like human intelligence. Worse, the virtual assistant metaphor doesn’t provide an operational mental model, because you can’t program / define a person, whereas you can do that with machine learning algorithms. There is no way for the user to understand the specificity of this algorithm, nor to engage meaningfully with the way it learns or react.

We need a new mental model, similar to how the desktop, windows and files have been the operational metaphor of the GUI and have enabled everyone to use a computer without having to write a single line of code. This new model should make clear what kind of interactions and choices are possible, what the limitations of the algorithms are.

#2 Discover new use cases and invent new services

So far, machine learning is dominated by an elite minority of data scientist and computer science engineers, who either are researchers in university labs or employed by large tech companies. The use cases they come up with for machine learning are therefore often limited to their areas of competence, market interests and the availability of data.

UX designers have another, larger mission: the mission to imagine new use cases, the ones that might not be the basis of an existing industry, the ones where data might not exist yet. To imagine these new scenarios, we need to find the problems we don’t even see as such in a world without machine learning. Where to start? Revisit user research, conduct new studies and identify pain points that might have seemed unsolvable before.

#3 Create the data to empower users, not (only) to serve business interests

Solving these new problems will likely mean collecting new kinds of data, or imagining new ways of collecting data. What? Don’t we have already enough data? Aren’t we overwhelmed by data? True, humans are easily overwhelmed by vast amount of data, but not so machine learning. Too much data for our small brain blinds us to the infinite possibility of the other metrics learning algorithms could use.

Data is always partial, always reflects a specific point of view. Right now, systems are built based on the data available: companies, people are “data driven” (Interesting read about data and creativity). The data available is usually what the provider of the services is most interested in, it might not reflect all the aspects of the user experience. Focusing solely on maximizing business metrics without critical insights can result in serious collateral damages (see Facebook model that emphasizes clicks and sharing, whatever the content is, which helped spread fake news).

So adopting a human-centered approach to machine learning means that designers should ask themselves questions such as: how could this user experience be best reflected in data? What new kind of data we could measure or collect? And how can users have control over their data? See Livable Tech for inspiration.

A new mission for UX research and design

Designers should reclaim the initiative: invent the machine learning mental model that will empower people, find their needs, imagine the solutions, decide which data is necessary to implement it, and how to collect this data if it’s not readily available.

First published on

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.