Why don’t we test machine learning like we test software?

The sheer complexity of problems we are trying to solve with AI means that more often than not we start hitting limitations of the systems. Having the ability to proactively examine how well the system covers the expected operational domain is invaluable in knowing the risk an organisation takes with that system.

You can read more about our strategy on using software testing ideology, expanding and applying them to Machine Learning systems on the Industry features article on Towards Data Science.