Position – Lead Designer

Contributions – User Experience, Design direction, and Vendor Management

 

PURPOSE

AI Learning Doors is a data collection initiative focused on facial recognition capture. A handful of data collection access points where installed across the Microsoft campus to analyze the impacts of ambient lighting, human velocity relative to capture rate, and experiential impact.

This Data was then used to provide “Frictionless” access points across the campus for the employees


SCENARIOS

There are 3 different use cases we evaluated when creating the A.I. Learning door. We needed to ensure employees had access and they knew what they were opting into or out of. Then we also needed to share with visitors that their data was not being used and it was immediately discarded during these moments of capture.

Employee Access

  • Upon entry through select campus doors, badge reader data was tagged to facial capture

Opt-Out

  • Alternative access doors were provided in all buildings

  • A formal opt-out process was available to those who wished not to participate

Visitor Access

  • Facial capture without benefit of tagged badge data

CORE COMPONENTS

When approaching the creation of this product the following where the core drivers in how decisions where made:


Intelligent UX

  • Facial Recognition

  • Correlated Badge Data

 

Features and Controls

  • Badge Reader

  • Opt-out

  • Signage & corporate communication

 

Ethical Implications

  • Fairness

    • Need to identify all key personal attributes that lead to higher error rates

  • Transparency

    • Employees need to understand the solution purpose and function

 

UI PATTERNS

  • Facial Capture

  • Entry Threshold

  • Navigation

 

RESULTS

After capturing employee data for over 3 years we have now been able to create a reliable entry system that allows employees smoother access to buildings and spaces that they require access to. This has allowed for much fewer lock outs of the buildings and increased safety and awareness in highly controlled areas. Through the ML process and improved cameras we have a dramatically lowered rates of mis-identification for all skin types.