UX Design Teams and AI
UX design teams and Machine Learning are getting acquainted. But the relationship is still in its early stages.
In Dove et al. 2017, the contributions of 27 UX design teams to novel machine learning projects is investigated. How did the UX team impact the development of an AI product and what was their role in the process?
The involvement of the design team was broadly categorised as:
Gave an interactive form to a machine learning idea that came from others (e.g. engineers or data scientists)
Generated a novel design concept utilising machine learning, which was presented and then selected for integration into a new product or service.
The UX design team collaborated with engineers, product managers or others, and jointly developed an idea for a new product or service that utilised machine learning.
Broadly speaking, this amounts to 1. UXers coming in after the fact to smooth things over and make the product usable, 2. design teams understanding ML well enough to come up with novel and feasible ideas, 3. design teams requiring partnership with data scientists.
To me, this represents a surprisingly good return for design teams. The ‘desirable’ approaches tally at 19, or 70% of the teams surveyed.
Even the typically lamented category 1 which could be pragmatically translated as ‘the design team was pulled in late to make the AI product usable’ remains a non-ideal but somewhat inevitable result of a world where UX is not completely embedded in development and business teams.
The paper goes into detail about the challenges that designers experience when working with AI. Especially, the constraints of working with this new, poorly-understood design material.
That designers working in AI should have a deep familiarity with the subject is a no-brainer; albeit non-trivial to realise.
To me, what’s more interesting and unfortunately not covered in the paper, is how the designers experiences play out in the joint sessions. What happens within those sessions and why is this collaborative method not as rosy as it might first appear.
Knowledge distributions
It’s not controversial to say that collaborative product development is desirable. But it’s not as simple as getting everyone in the same room. Or even of getting the right people in the room.
Where the knowledge of AI sits has a profound effect on the quality of the session. Does everybody in the room have a good understanding of AI? Or is expertise cleanly split between those who ‘get’ the algorithms and those who ‘get’ the user. A room with individuals with a healthy overlap of shared knowledge far more likely to generate plausible, attractive ideas than one where knowledge sits in discrete groups.
The more that UX of AI knowledge sits within individual brains, the better.
The best thinking is done between synapses rather than in the air between two minds. To generate the spark of collaborative genius requires the individuals to understand each other. With too much friction, too much translation between the participants, ideas cannot flow and collaboration turns to committee meeting.
Trust
Shared understanding has benefits beyond the quality of individual idea generation. Shared understanding builds trust.
Creativity demands that individuals feel safe. That perception of safety can quickly evaporate when participants feel judged, or isolated from others in the room. One way that this can fester is when participants have very different points of view. I’ve seen creative confidence dissipate before my eyes when seemingly ‘dumb’ suggestions have been put forward by those unfamiliar with AI or when a team of engineers betrays a disregard for the experience of the user.
Getting UX of AI knowledge into more individual brains boosts the shared understanding in the group and creates an environment where good ideas are more likely to flow.
— — —
And so the challenge is to transport that intimate knowledge of AI into the minds of designers. Not so that they can work alone, but rather so that design sessions can capitalise on the combinatorial power of Yes! And, and the friction coefficient of ideation sessions goes to zero.
— — —
Sources & inspiration