Welcome to an article series — ‘Machine Centred Design’ where I will be sharing my experience as a product designer working exclusively on machine learning driven features and products for the last couple of years. In the series I will be explaining how user experience design differs when working with machine learning, how to collaborate with data scientists and what skills do you need to get started. More about the series and what to expect can be found in the introduction here.
In this part, I will provide an overview of the core skills required to begin designing experiences for machine-learning-driven products and features. Through my experience, I’ve identified three core skills:
- Understanding machine learning, including the basics of how it works and the different types of it.
- Collaborating with data scientists, understanding their process, and facilitating collaboration as a product designer.
- Translating algorithm outputs to create user experiences, and the different considerations and tradeoffs to be aware of.
In this article, I will go over all three skills in greater detail, explaining what they mean and why they are important. The series’ next two articles will delve even deeper into the process of collaborating with data scientists and different experience considerations when working with machine learning. However, I will not be writing a separate article on how machine learning works because there are far better and more detailed resources available online — I will include links to some of them below.
Understanding machine learning
As a product or experience designer, you do not need to be a data scientist to provide valuable input on machine learning-driven features and products. However, as with any other discipline, fundamental knowledge and understanding of the language can go a long way. It should make it easier for you to collaborate with data scientists and understand the different experience design considerations when working with machine learning.
First off and I guess the most obvious thing to learn is the basics of how machine learning works. I highly recommend a course by Andrew Ng founder of deeplearning.ai called ‘AI for everyone’ — It’s very comprehensive, easy to watch, and will give you all the context you need to get started. Also, it’s worth keeping this glossary at hand in case any of the terms don’t make sense.
The general premise of machine learning is based on making models that use algorithms to make predictions based on input data.
These models can then make fresh predictions based on new data. Now it’s worth noting that there are different kinds of machine learning — supervised, unsupervised, and reinforcement. The difference between them is useful to know, even if the majority of commercial machine learning models use supervised learning. This will be covered in the ‘AI for Everyone’ course, but you can also learn more about it here. Once you have the basics down there’s a couple of things that will help you build a really strong foundation as a product designer.
As you can see from above, a key part of any machine learning model is the data that powers it. As a designer, it’s important to understand what types of data are out there and how data gets sourced and used. This should give you enough context to be involved in conversations around data and watch out for bias in the model. Bias in AI is a hot topic and I’m sure there are many articles about it, however, one thing to call out is that typically bias happens from poor or incomplete data that amplifies biases in the real world.
Finally it’s worth learning about concepts related to machine learning model outputs as you will be using these outputs to create experiences for users. Particularly concepts like confusion matrix, accuracy, explainability, precision, and recall. The impact of these concepts will be covered in the last part of this series, but it’s good to familiarise beforehand.
Now that you have a general understanding of what machine learning is, let me give you a quick rundown of what you don’t need to know about it. This will keep you from going down the rabbit hole, unless part of you is interested in becoming a data scientist.
One of the most obvious things you don’t need to know is how to build machine learning models, as well as related frameworks and programming languages like PyTorch and Python. You also don’t need to be familiar with all of the different types of machine learning models, though some general knowledge about model differences, such as the difference between a decision tree and a neural network, may be useful. Finally, while it is important to understand the general process of data science projects as well as the life cycle of machine learning models, you do not need to understand all the detailed steps of the process.
Collaborating with data scientist
Having foundation knowledge of how machine learning works will definitely propel you towards success, however in my opinion knowing how to effectively collaborate with data scientists is one of the most important skills for a product designer when working on machine learning-driven products or features.
You may wonder why to have a product designer involved in a machine learning project in the first place. The answer is fairly simple — product and experience designers are advocates for users, their needs, and their problems. Working on a project aimed at delivering a machine learning-driven product or feature is no exception. That is why it is important for product designers to be involved in all stages of the project lifecycle, collaborating with data scientists to ensure that the right user problem is solved and that the user problem is solved in the right way.
Being a designer naturally involves lots of collaboration and working with a variety of stakeholders, both cross-disciplinary and cross-functional. Data scientists should also be a part of the stakeholder group, but unlike stakeholders who are limited to just providing feedback, they should be a part of the core product team in the same way that engineers and product managers are. This unlocks easier collaboration by providing more opportunities for input on data and model decisions, as well as the ability to co-create user experiences.
Due to limited data science resources, this may be difficult to achieve in some organisations — in these cases, the focus should shift to strong project kick-offs and co-creating the end-user experience. A good project kick-off will help everyone get on the same page when it comes to user problems and project objectives. Co-creating the end-user experience later in the project lifecycle will help to ensure that the machine learning model has solved the user problem and that the proposed design is compatible with the model that data science has developed.
As mentioned earlier, data science should be a part of your product team, or at the very least a key stakeholder. And, as with any stakeholder, there are different things to consider when collaborating with them. Data scientists, first and foremost, use different technology, complete with its own frameworks, tools, and methods. That also means their process is quite different from a more familiar software engineering process. Finally, that process comes with a slightly different approach to problem-solving. I won’t delve deeper into the differences in this article, but make sure to keep an eye out for Part 2 of the series where I will be solely focusing on collaboration with data scientists and will be expanding more on the detailed process, how it relates to the design process and other things to keep in mind.
Translating algorithm output to create experiences
Final skill to highlight is the ability to translate machine learning algorithm output into user experiences. Translate is a useful word here because, just like languages, understanding one is necessary in order to construct a sentence in another. Similarly, in order to provide a seamless experience for users, you need to understand how to use the outputs of machine learning algorithms. Understanding the fundamentals of how machine learning works, as well as working collaboratively with data scientists, will definitely help in understanding the output — however, translating it into something intuitive and easy to understand for users is a skill on its own.
The entire translation process stems from a single concept: creating an experience based on a prediction, which, like any prediction, can be incorrect or difficult to understand at times. Similarly, not all predictions have the same weight — some may be critical, while others may turn out to be trivial. These considerations will play an important part when designing machine learning-driven experiences and products as they often add a layer of complexity that is not always immediately visible. If these considerations are overlooked, not only can they cause friction and confusion in the experience, but they can also harm brand perceptions or worst case cross ethical boundaries.
The considerations mentioned above are just a few high-level examples — each has more nuance and detail. To illustrate, let’s use the example of predictions having different weight, some being more important than others. Imagine we’re creating a recommendation feature for an online candle store, predicting which candles users would like based on their previous purchases. Sometimes we might not have enough data to make a very accurate prediction, therefore our recommendation experience might show some candles that are not very relevant. Now, imagine we’re creating the same recommendation feature for an online stock broker, predicting which stocks users should invest in based on their past performance.
That last example immediately feels tenser, and for obvious reasons, predicting price and providing investment advice based on it will irritate many lawyers and regulators. On the other hand, recommending a cherry-scented candle instead of a vanilla one might just be okay. There’s still a lot to unpack here, but these examples begin to show how accuracy of the machine learning model and the type of predictions being made directly impact user experience and the product itself. I will be expanding further on these considerations, as well as providing actual examples of how they play out in user experience, in the final part of this article series.
This is it for Part 1, hope it was helpful to better understand what skills you will need to start designing experiences for machine learning-driven products or features. There’s a lot of material already covered here, but do stick around for Part 2 and 3, where I will be diving further into the process of collaborating with data scientists and different experience considerations when working with machine learning. Finally, if you’re a fan of the intersection between user experience and AI, follow me for updates and make sure to like the article if you found it interesting. See you in Part 2!