You may have heard the term Black Box spoken in relation to Facebook or Google. Black Box is unclear in how it works, and it is not open as to how it is actually using your data.
ProPublica have published a number of articles about various companies’ black boxes. These articles have touched on how black boxes work and what is being used to determine the results.
The use of Black boxes can lead to machine bias. Machine bias is the effect of erroneous assumptions in machine learning processes. Bias reflects problems related to the gathering or use of data, where systems draw improper conclusions about data sets, either because of human intervention or as a result of a lack of cognitive assessment of data.
Facebook has an internal project tool called Fairness Flow, which can allegedly can determine whether a machine learning algorithm is biased; meaning it systematically provides certain groups worse results along the lines of race, gender, or age.
Microsoft has also been working on developing a tool to automatically catch bias in machine learning.
A number of the algorithms use math concepts that have been used for a number of years. One of those is the regression model. In the case of Cambridge Analytica (CA), they took their data from users on Facebook. CA took the number of inputs from the users’ data (which would be likes), with each being an input, then used the data to fit into a regression model.
David Sumpter covers a number of the black box algorithms that exist as well as what companies do with your data in his book, Outnumbered: Exploring the Algorithms That Control Our Lives.
Algorithms are only as smart as the companies that have developed them. The issue with companies using one termed as Black Box algorithms is that the system is closed, and you do not know how those algorithms work, or how they might impact the sites or services that you use.
Machine learning is only as smart as the AI that has been programmed, and the data that can be used to learn from. Where bias comes in machine learning is that humans have a certain built-in bias, and by nature, some of that bias will exist in machine learning unless it is looked for. When the bias is found, internal/open tools work to understand how to correct it.
