Why I love UX/UI as an ML engineer.

“There’s a truth, universally accepted, that an AI startup in posession of funding must be in search of good UX designers”

As I have said before in things I’ve written, I think we’re coming out of the machine learning hype cycle and into the “plateau of realism” (or the plateau of broken dreams!). It can be a tough time in a technology development cycle, and it’s undoubtedly the period when those who weren’t serious about the promise of ML/AI will break for the exits, or quantum computing, or blockchain, or some other technology. For those of us who choose to stay, though, there are many reasons to be cheerful.

Firstly, and possibly most importantly, tools have come a long way since the heady days of 2016 when nothing was real and everything was possible. We have seen the original R/python data science battles become the Tensorflow/Pytorch battle, in which competition has made both much better (and more similar…). The other reality is that we have seen, from some companies at least, very clear signals of how to build a data-driven business. In particular, it’s noticeable that companies like Alphabet, Amazon and Meta in particular have managed to harness the “data virtuous cycle”:?

using user data…

…to build tools that improve the user experience…

…so that more users join, and you have more user data…

…to build better tools…

It’s not the easiest cycle to get into (cold start problem anyone?), and there are many organisations out there who start down this road only to discover that they don’t have sufficient, or sufficient quality data to support even taking the first step. Which is great, but what does this have to do with UX and UI? The importance of both these fields is twofold -?

1. ?to deal with the fact that machine learning models can make dumb mistakes in an elegant way
2. ?to keep users on the happy path where they use our services as we expect, and so provide data we can use to make them better

The difference in user experience between e.g.?

• ?A chatbot that answers a completely different question from the one that you asked and
• ? A chatbot that prefaces that answer with “I’m not sure what you mean, but I think the answer might be …” based on an awareness of it’s uncertainty

Is clearly vast!?

“Traditional” software team operational structures usually define a linear process whereby teams communicate almost exclusively with their nearest neighbours. All watched over by the machines of loving grace that are project and product managers. The strongest interface of this whole structure is between backend and frontend teams, and is traditionally enforced by APIs. Somewhere on the backend side of that structure sit machine learning models, very far from the people who design the interfaces that serve those models. The strongest path of communication the two groups is via the product manager.

Wouldn’t the world be better with symbiosis? Why not put the people who generate the data (the frontend designers) and the people who consume it (the machine learning folks like me) closer together? That way the producers and consumers can negotiate more effectively for what they need, and support each other into the mix.

Recently, I have been thinking a lot about how to incorporate machine learning solutions into products, and I have come to strongly believe that cross-functional product teams are key. We have had a pretty successful time incorporating a data analyst (to bring understanding of what we have), a data engineer (to clean and present it), a data scientist (to use the data) and a machine learning engineer (to deploy the model that comes out of the other end) into teams from the beginning.?

But there has always been something missing. We have always needed the buy in from frontend engineers to get data delivered to us in a way that works - and that’s hard to do when they’re “just building the frontend”, so bring them in early. My advice to anyone embarking on a machine-learning powered project are two + n rules:

1. Don’t use ML unless you absolutely have to.
2. Put UX front and centre from the very beginning.
3. {A bunch of other can’t miss rules that I’m sure I would add if I were talking to you!}