Suggested time: 2 minutes
Like with all technical implementations, they need to be maintained and monitored throughout their use. This is particularly true for predictive algorithms, which are subject to a phenomena known as concept drift, wherein the accuracy of predictions tend to worsen over time due to attributes becoming less correlated with whatever is being predicted. As a consequence of concept drift, explanations may shift or become less meaningful over time.
Furthermore, the interest in algorithmic transparency is growing rapidly, and new methods for algorithmic transparency are being created on a yearly basis. A new breakthrough in transparency could come at any moment – and it’s critical that your organization stays up to date with best practices!
With these considerations in mind, we highly recommend you consult with relevant stakeholders, particularly those building algorithmic systems and those using them, to design a plan for maintaining, updating, and repairing the transparency features of algorithmic systems.
DELIVERABLE: A process for continually maintaining algorithmic transparency within your organization.