I contributed to an article titled “Machine Learning, Social Welfare and Fairness” in the October 2020 edition of the online Pure AI web site. See https://pureai.com/articles/2020/10/05/ml-fairness.aspx.
The article is intended to give an introduction to a classical economics topic to people who know a bit about machine learning. Nash Social Welfare (NSW) is a way to allocate items among people when the people involved place different values on the items.
I explain the idea using an example where there are three brothers, Adam, Brad, Carl, who must divide five items of jewelry among themselves. The NSW principle says that the items should be allocated so that the geometric mean of the utility values of the items is maximized.
The NSW principle is called envy-free (EF) because it turns out that after dividing up the items, no person will prefer another person’s bundle of items to their own.
In the article, I explain that machine learning enters the picture in two ways. First, some machine learning techniques can be used to compute the NSW allocation (it’s a very difficult problem). Second, for machine learning algorithms that produce some sort of allocation, the allocation can be analyzed to see if it is envy-free or not.
The moral of the story here is that very powerful systems can be created by combining knowledge from different areas. In some cases the whole is greater than the sum of the parts.
Artist Daniel Arrhakis combines several types of art media to give very nice images that, to me, are nicer than images created using just one type of media.