Last updated 3 weeks ago

White-box Architecture

We've often heard applications of artificial intelligence be described as "AI magic". While this description seems endearing, it's also somewhat problematic: how can we understand and optimally use a system if it's surrounded by mystery? That's why the Skill Engine API was made to explain itself and be understood, thanks to its white-box architecture. This system is supported by the following design aspects:

  • Decoupled architecture: the Skill Engine is not a single end-to-end AI model, but rather a collection of models working in union. Each of these has a clear task with interpretable in- and outputs, which can be evaluated and controlled separately. This allows us to control for bias and fairness, while maximising performance.
  • Explainability: each match result produced by the Skill Engine can be linked all the way back to the original input. This way, everything can be explained not just from the system's perspective, but from your perspective as well: our system builds its reasoning on skills, experience and education, and it's always ready to explain how. An accurate explanation of suggestions not only reinforces trust, but also provides you with insight into the inner workings of the Skill Engine.
  • Human-on-the-loop: while explainability goes a long way, truly responsible AI gives its users the option to get involved when they want to. Through our skill profile feedback system, this voice is taken into account.