Brianna White

Staff member
Jul 30, 2019
One of the most important aspects of data science is building trust. This is especially true when you're working with machine learning and AI technologies, which are new and unfamiliar to many people. When something goes wrong, what do you tell your customer? What do they say? What do they think will happen next? With explainable AI, you can provide answers that prove your product's legitimacy.
Explanation is a key part of building trust in any technology application, but it's even more important for machine learning applications where:
• You don't know how the system works (e.g., image classification).
• There isn't a clear causal relationship between inputs/outputs (e.g., recommendation systems).
Most models are "black box" models, and often when these models are trained, ML scientists are unable to understand how the model made a prediction or why it predicts what it does. The inability of the model or ML scientists to explain predictions to stakeholders, as well as the difficulty in interpreting the model training behavior, leads to a lack of stakeholder trust in the model and its predictions.
Explainable AI helps build trust in AI by providing continuous visibility into training and production models. ML scientists and stakeholders can understand why predictions are made and derive actionable insights for teams to fine tune and retrain the models.
Continue reading:
  • Like
Reactions: Brianna White