Is Legal AI Ethical AI?

In their talk, Jessica Fjeld, Adam Nagy and Mason Kortz of Harvard University posed the question “Is legal AI ethical AI?” They clearly showed that this question is difficult to answer because it is hard to convey what is ethical and moral in meaningful legal language. AI algorithms will soon be required to meet key ethical principles such as non-discrimination and fairness. Typically, these principles need to be written in unambiguous legal language which can then be used to create accountability. These principles would then need to be incorporated into algorithms. As a result, there are four layers of abstraction: moral (one’s personal judgements of what is good and bad), legal (what the community agrees to be right or wrong), heuristics (the assumption or information an algorithm is based upon), and algorithms (how the heuristics are implemented). This process of abstraction comes with a huge challenge: algorithms end up becoming a proxy of a proxy of a proxy of our beloved moral principle! There are many ways this can go wrong. For instance, consider the idea of fair housing in the U.S. As per the fair housing law, when awarding mortgages one cannot discriminate based on race, sex, nationality or any other protected attributes. This is why to reach the so-called “fairness through unawareness,” mortgage lending may choose not to use protected attributes such as race. Yet, in so doing, they may engage in a different, more subtle, form of discrimination. For example, not having any information about race makes it impossible to know whether a dataset is well-balanced in terms of race or not, and certain variables the algorithm does use to deny a mortgage may correlate with race (e.g., income).

To find out more, see the Nokia Bell Labs Responsible AI hub: https://www.bell-labs.com/research-innovation/ai-software-systems/responsible-ai/

Видео Is Legal AI Ethical AI? канала Nokia Bell Labs