Best practices to draw Machine Learning Interfaces: 3 types of approaches

Intelligent systems surround us; they are no longer something of the future, but in the present. With each passing day, new solutions, software, and systems are equipped with Machine Learning to help with decision-making, operationalize complex tasks with users, and monitor user activities, among other objectives and intentions.

In order to help the user, the interfaces must take into account new realities. Principles that were once used merely for machines and complex operations (e.g., lifting an airplane), now have to be present in our daily lives as UX Designers and Usability specialists.

As usual, I want to share some of the ideas that emerged based on some research in scientific articles referenced in the bibliography.

1 → The Collector

The interface should display and give access to other sources of knowledge and information. The system pick and gather useful or useless information from external data sources and collect all in one single point. 

Hierarchy, segregation and, information management is a must to have in mind: help the users overcome data overload effect. Technically, the system needs to give the capability to search and navigate faster throughout the information. Several interactive features should be investigated and tested with the users to aid the user manual discovery patterns in data set.

And… what could the interface look like in practice?

---

2 → The Cooperative

The focus is on developing intelligent, semi-autonomous, machine agents that participate as cooperative members of distributed person-machine systems for accomplishing cognitive tasks. 

A central tenet of the cooperative system paradigm is the need to provide team members with shared knowledge of the current situation and the goals and actions of participants, both Humans and Machines. (System and User have to be in the same page, none know more than another).

Two approaches to implement this:

a) System as a critic develop a system that is tutor, a mentor to critique the human-generated solutions. The system must examine the user solution and offer an advice if the solution is deemed incomplete or incorrect.

And… what could the interface look like in practice? 

• Alerting the user to possible errors, oversights, suggesting relevant issues to consider;
• Passive or active mode, could be invoked by a user or continuously monitoring user performance and giving suggestions;
• Implement some influencers (guided tour), debiasers (error states), directors (wizards).

Advantages and benefits

• More efficient than a total automatic system;
• The collaborative manipulation from the user allows cooperation and detect when the strategy user by the computer was inappropriate to the situation;
• This enables the user to recognize situations where computer strategy was not good;
• Preserves the users accustomed role;
• Avoid loss of skills from the users.

b) Systems as a team player: system that act as a team player; In contrast to the critiquing approach the machine is allocated to a portion of the cognitive tasks, toward a common goal.  (Auto-pilot, power plant control, train control rooms). The user and the system share the responsibility for task performance, but the user remains as the team leader responsible for achieving task goals. 

And… what could the interface look like in practice? 

• Support the user as team leader: ask for advice, help and accept corrections from the user;
• Reliable and consistent performance: predictable actions and clear assessments of situation;
• Coordination and dynamic task sharing;
• Cooperative communication: the interface is a medium of new information for both.

Advantages and benefits

• The user has command authority
• The user has responsibility for supervision and correction of the intelligent system
• The system fulfills an aiding role, provide information to assist the User
• A team player is Reliable, good Coordinator, good Communicator. 
• The user must be able to take the control, disable the intelligent system.

---

3 → The Informer

The goal of this approach is to capitalize on advances in graphics and knowledge-based techniques to create computer-generated presentations that facilitate human problem-solving and decision-making. 

Special thanks to my colleagues in the area Maria João Leite & Luís Ferreira 👏

Bibliography

• Claus O. Wilke. Fundamentals of Data Visualizations. 2019.
• Design, User Experience, and Usability Behavior, 10th International Conference, DUXU 2021
• https://pair.withgoogle.com/guidebook