The crucial question, therefore, is this: how can we guarantee
that AI decision-making is correct, safe, reliable and ethical?
For me, the answer is simple: we must apply our human
experience. As parents, we understand that it is our duty to
educate our children on what is right and wrong-to reinforce
societal norms, values and behaviors so that the next
generation is equipped to function as productive members of
society. The same approach needs to be taken with AI. What
is needed is a robust framework for teaching and training AI
applications-a machine learning curriculum that will help
ensure AI applications are aligned with our human values. Key
to this will be building a test framework, bespoke to AI
applications, to guarantee decision-making is transparent,
explainable, fair and non-discriminatory.
TESTING TIMES
Let us be clear: building such a test regime will be no easy feat. AI
applications comprise a large number of components, many of which
change over time. They are also fueled by vast amounts of structured
and unstructured data and processing this data can be challenging. The
algorithms themselves also need to come under scrutiny: software
engineers must carefully evaluate the accuracy and performance of the
learning models to ensure ethical and unbiased decisioning and
regulatory compliance. Engineers also need to build new test and
monitoring processes that account for the data-dependent nature of AI
systems. In the face of this complexity sits a clear requirement to
simplify design and validation processes to make the development of
ethical AI applications easier.