Hi Machine Learning, meet UX
In the years to come, the fertile fields of Machine Learning and UX will collide. As ML finds its way into more and more of the products and services we use on a daily basis, the focus now shifts to impacting the user experience in a purposeful way. Welcome to the new world of MLxUX.
As AI matures and becomes democratised, we are seeing an exciting shift towards a more user-centric usage of data science and Machine Learning. AI, as a design component, shows real value in creating better products and empowering users by influencing the user experience in a meaningful way. Having users interact more directly with AI systems does come with challenges though. The newly emerging field of ML/UX teaches us how we can make AI and data science more inclusive and transparent for end-users, so they can trust and use it to its full potential.
ML For Business
At their core, data science, AI and ML transform raw data into information, decisions and actions. They excel at automating data-driven processes and help us to make more informed decisions. This chain from data to action and their incredible flexibility makes them much-loved tools for optimising business value in all types of industries.
Picture Facebook's content feed, YouTube's recommender system, Instagram’s profanity detection or Uber’s demand forecasting. What’s their common denominator? They all use AI technologies to tackle business-critical challenges. Facebook and YouTube optimise for engagement, Instagram wants to create an advertising-friendly space and Uber wants to optimise resources and predict price elasticity. The fact that AI is at the core of such widely-used products is a testament to the maturity of AI and its business value.
Nevertheless, we tend to ignore the impact on the experience of the end-user when choosing to optimise for business reasons. Untransparent personalised feeds and recommendations create a false sense of reality and echo chambers. Biased or poorly communicated filters and content censorship lead to frustration and distrust in end-users. Engagement is mistaken for an enjoyable or meaningful experience, often leading to an adverse impact on mental health and the proliferation of fake news.
We tend to ignore the impact on the experience of the end-user when choosing to optimise for business reasons.
All too often we blame products and companies for using AI in order to make money, but we think too little about the unintended consequences AI has on the user experience side. So, instead of cursing the past, we should focus on what we’ve already learned and improved in recent years and start looking to the future.
ML For Users
When we think about how ML in products affects the user, two considerations take centre stage: ‘Expectation Management’ and ‘User Control.’
In traditional digital product design, we assume deterministic behaviour from the components we use. When we press ‘send’, we expect our mail to travel the web. We learned over the years that pressing ‘control+z’ will undo our last action the same way that flicking a switch turns on a light. And when it doesn’t, we freak out. The point is that we have built up trust and expectations around particular digital objects, and breaking these expectations causes a lot of friction for the user.
Machine Learning is the opposite of deterministic. The predictions are probabilistic, and their quality depends both on the data the model learns from and the context we use the model in. These properties lead to opinionated results, and when designing interactions, we have to frame the results along with their uncertainty. The results should be presented not as an objective fact or magical solution, but as a prediction that can be wrong and should be challenged.
To empower the user, instead of infuriating them with a system that will never be perfect, we have to identify what the user is willing to offload to an ML system and what the user wants to control. Repetitive, boring, data-intensive or time-intensive tasks, producing only intermediary results, are often good candidates to offload to ML. Tasks or actions with a high impact, a high cost of being wrong, or nuanced results are better off staying in the user’s control.
To empower the user we have to identify what the user is willing to offload to an ML system and what the user wants to control.
An excellent example of this balance is Google Maps. After entering the start and end-point, the user offloads the task of finding the best route to the ML system. The system responds by offering multiple possible paths, information about the trajectories and the possibility of adding more waypoints to keep the user in control of the final result.
ML For Product Teams
Creating ML-enabled digital products that integrate ML in a meaningful way for the end-user is a team effort. It involves both the engineers, the designers and the team’s product manager.
The product manager needs to ensure that user needs are captured correctly and know what parts of the process are worth automating or augmenting using ML.
The engineering team needs to master a wide range of technologies within and outside the realm of ML, understanding the powers and drawbacks of these technologies in the context of the specific user problem. The resulting design space enables designers to create a wide range of interactions that solve real user needs and give them the freedom to frame the AI predictions and capture feedback correctly. Ultimately, this feedback updates the product manager's view on user needs and influences the next technology update from the engineering team.
Purpose beats function
So, what is the takeaway from all this? The power of data science, Machine Learning and AI means something totally different to a business than it does to the end-user, and this should affect your development process. Including user needs and feedback in the development cycle means adding UX design and user research in the data science mix to go from functional AI to purposeful AI.
It is important to note that this is not the first time we have carried out this kind of exercise. Two decades ago, we started creating a common language to help businesses and users use, understand and navigate the web. Later, we did the same with mobile apps, shaping a digital space aiming to be a meaningful part of people's life. Data, AI and ML are now at that same stage: having a clear impact on people's lives but in dire need of a common language so they can be actually meaningful to people.