A person reading the news.

Reducing Information Overload with Artificial Intelligence

year

2018

type of work

UX/UI, Co-Creation, IA Interfaces

premise

Artificial Intelligence is slowly finding more and more areas of application. This raises the importance of how we design respective interfaces. Due to a non-disclosure agreement I have to be a bit abstract when talking about this project. Let’s say the client which I was working for developed an algorithm to evaluate the positivity/negativity of news articles. Based on its evaluation the algorithm will assign each article a score. The scores are summed to an overall score for different themes.

design challenge

To improve the confidence of the algorithm, it required user feedback on the presented scores. The first version of the product was, however, built without any design involvement and yielded only limited feedback possibilities. The project’s challenge was consequently to unite the requirements of data scientists with user behavior and needs. My role included the collection of these requirements, the moderation of both parties to agree on needed features as well as the design of an adequate interface allowing the algorithm to learn and draw the right conclusions from human behavior.

Heuristic Analysis

I started the project with a heuristic evaluation of the existing interface. Such analysis allowed me to identify obvious design flaws in usability and gestalt. I ended up recording a list of more than 50 instances, including minor UI issues like inconsistent link styling as well as major flaws such as missing friction for crucial actions. Together with the development team, I further analyzed the current back- and frontend restrictions, to increase the awareness of those during the design process.

Heuristic analysis of the existing interface (altered interface).
— Heuristic analysis of the existing interface (altered interface).

Understanding User and Business Requirements

My research phase was split into two parts. On the one side I had to understand the feedback required by the algorithm to improve itself. On the other hand, it was crucial to understand the daily workflows of the users and how the product integrated into these. I therefore conducted qualitative, exploratory interviews with data scientist as well as existing users. During the user interviews, I focused on identifying their direct and indirect ways of feedback for the algorithm. At the same time the interviews allowed me to establish a personal relationship and to earn the interviewee’s trust, which became major success factors throughout the project.

Interface Design for AI

Based on the learnings from research and the heuristic analysis, I created an initial set of mockups combined in a click-dummy. Unfortunately I’m not allowed to present any details of the interface I designed at this point in time. Be aware that the designs presented on this page as well as the information on such has therefore been purposely modified or omitted.

Wireframes of the redesigned interface (altered interface).
— Wireframes of the redesigned interface (altered interface).

Co-Creation Workshop

The users were then asked to test and evaluate the click-dummy within a workshop held at their premises. Within different tasks the group prioritized the presented features based on their needs. The further design and development roadmap was derived from this prioritization. The design drafts were reworked and evolved into a high-fidelity mockups using the corporate identity design system. Finally, I prepared the specifications and required assets before handing over to the development team.