Assessing and evaluating evidence

[chrisdburr]

The CDEI have a list of techniques for assuring AI systems:

https://cdeiuk.github.io/ai-assurance-guide/techniques

What do each of these techniques provide evidence for in the context of an assurance case?

For example, what would an algorithmic impact assessment be reliable and valid evidence for?

[chrisdburr]

Another technique is the bias audit. This one is easier to link to claims. For instance, the results of a bias audit would provide evidence that identification and mitigation of possible biases (e.g. statistical, social, and cognitive biases) has been undertaken during the course of a project’s lifecycle.

This could feed into an assurance case that has a goal related to fairness (e.g. non-discrimination, equitable impact).