As artificial intelligence and machine learning tools become more pervasive and accessible, product and engineering teams from all types of organizations are developing innovative products and features based on artificial intelligence. Artificial intelligence is particularly suitable for model recognition, prediction and prediction and personalization of user experience, all common elements in organizations that deal with data.
A precursor in the application of artificial intelligence is given by the data, many and many! Large data sets are generally required to form an artificial intelligence model, and any organization that has large data sets will undoubtedly have to face the challenges that AI can help solve. Alternatively, data collection may be "phase one" of AI product development if the data sets do not yet exist.
Whatever the dataset you plan to use, it is very likely that people have been involved in acquiring that data or are engaging in some way with your AI functionality. Principles for UX the design and visualization of data should be an early consideration during data acquisition and / or in the presentation of data to users.
1. Consider the user experience in advance
Understanding how users will interact with your AI product at the start of model development can help put useful protections on your AI project and ensure that the team is focused on a shared end goal.
If we take the "Recommended for you" section of a movie streaming service, for example, outlining what the user will see in this function before starting the data analysis will allow the team to focus only on the model outputs that will add value. Therefore, if the user's research determines the title, image, actors and length of the film will be valuable information for the user to be displayed in the recommendation, the engineering team would have an important context in deciding which data sets should form the model. Data on the length of the actor and the film seem crucial to ensure that the advice is accurate.
The user experience can be divided into three parts:
- First: what is the user trying to achieve? How does the user get to this experience? Where are they going? What should they expect?
- During - What should they see to find their way? Is it clear what to do next? How are they guided through mistakes?
- After: Has the user achieved his goal? Is there a clear "end" to the experience? What are the follow-up steps (if any)?
Knowing what a user should see before, during and after interacting with the model will ensure that the engineering team is training the AI model on precise data from the start, as well as providing more useful output for users.
2. Be transparent about how you are using the data
Will your users know what's going on with the data you're collecting from them and why do you need it? Should your users read the pages of your Terms and Conditions for a suggestion? Think about adding logic to the product itself. A simple "this data will allow us to recommend better content" could remove friction points from the user experience and add a level of transparency to the experience.
When users ask for support from a Trevor Project consultant, we make it clear that the information we ask for before connecting them with a consultant will be used to provide them with better support.
If your model presents output to users, take another step and explain how your model has come to an end. "Why this announcement?" Google's option gives you an idea of what drives the search results you see. It also allows you to completely disable ad personalization, allowing you to control how your personal information is used. Explaining how your model works or its level of accuracy can increase trust in your user base and allow users to decide for themselves whether to commit to the result. Low accuracy levels could also be used as prompts to gather further insights from users to improve your model.
3. Gather users' opinions about the model's performance
The request to users to provide feedback on their experience allows the Product team to make continuous improvements to the user experience over time. When thinking about collecting feedback, consider how the AI engineering team could also benefit from continuous user feedback. Sometimes humans can spot obvious errors that AI wouldn't make and your user base is made up of humans only!
An example of collecting feedback from users in action is when Google identifies an email as dangerous, but allows the user to use their own logic to mark the email as "safe". This manual and ongoing correction to the user allows the model to continuously learn how dangerous messaging looks over time.
If your user base also has contextual knowledge to explain why AI is incorrect, this context could be crucial for improving the model. If a user notices an anomaly in the results returned by artificial intelligence, think about how you could include a way for the user to easily report the anomaly. What questions could you ask a user to collect key information for the design team and to provide useful signals to improve the model? UX engineering teams and designers can work together during model development to plan feedback collection in advance and set the model for ongoing iterative improvement.
4. Evaluate accessibility when collecting user data
Accessibility problems involve distorted data collection and artificial intelligence trained on exclusive datasets can create AI distortions. For example, facial recognition algorithms that have been trained on a dataset consisting primarily of white male faces will perform poorly for anyone other than white or male. For organizations like The Trevor Project that directly support LGBTQ youth, considerations of sexual orientation and gender identity are also extremely important. Searching for externally inclusive datasets is just as important as ensuring that the data you bring to the table, or that you intend to collect, is inclusive.
When collecting user data, consider the platform that your users will be able to use to interact with your AI and how to make it more accessible. If your platform requires payment, does not meet accessibility guidelines or has a particularly cumbersome user experience, you will receive less signals from those who cannot afford the subscription, have accessibility needs or are less tech savvy.
Every product leader and artificial intelligence engineer has the ability to ensure that marginalized and underrepresented groups in society can access the products they are building. Understanding those who are unconsciously excluding themselves from their dataset is the first step to creating more inclusive artificial intelligence products.
5. Consider how to measure equity early in model development
Fairness goes hand in hand with ensuring that your training data is inclusive. To measure equity in a model it is necessary to understand how the model might be less fair in some use cases. For models that use people data, observing how the model performs on different demographics can be a good start. However, if the dataset does not include demographic information, this type of equity analysis may be impossible.
When designing the model, think about how the output could be distorted from your data or how it might underestimate certain people. Make sure that the datasets you use for training and the data you are collecting from users are rich enough to measure fairness. Consider how to monitor equity as part of normal model maintenance. Set an equity threshold and create a plan on how to adapt or retrain the model if it becomes less fair over time.
As a new technology worker or expert in developing AI-based tools, it's never too early or too late to consider how your tools are perceived and affect your users. AI technology has the potential to reach millions of users on a large scale and can be applied in high-risk use cases. Holistically considering the user experience, including the impact of AI output on people, is not only good practice, but it can be an ethical necessity.
