Contemporary Issues in Computer Science

Algorithm Bias

 

We often idealize technology as a way to completely eliminate human error and bias.  But, unless we are brutally self-aware in the design stage (and probably even if we are), our technology will mimic our own weaknesses.  

 

So, read:

 

http://www.nytimes.com/2016/06/26/opinion/sunday/artificial-intelligences-white-guy-problem.html?_r=1 

 

https://www.technologyreview.com/s/601775/why-we-should-expect-algorithms-to-be-biased/ 

 

http://www.theverge.com/2016/5/25/11773108/research-method-measure-algorithm-bias 

 

http://spectrum.ieee.org/tech-talk/computing/software/computer-scientists-find-bias-in-algorithms 

 

Answer these two questions.  Focus on the connection between technical issues and societal/ ethical impact.

 

1.      Choose one specific problem domain in which sophisticated, data-driven, and probably biased algorithms are already making important decisions.  Then answer these questions:

 

a)     Very briefly describe the problem and the algorithm(s). 

 

b)     What data do they use as training sets? 

 

c)      What is the evidence for bias?  What features (for example, race or gender or socioeconomic status or ethnicity) may be involved?  Can you suggest reasons why available data could lead to bias?

 

d)     Briefly analyze the likely impact of this bias.  On whom?

 

e)     Is anyone trying to solve the problem?  

 

[Write about 500 words, more if you really get into the issue.  If possible, find an example different from the ones mentioned in these articles.  Or choose one of these and tell us more about it.]

 

2.      What techniques can or should we as a society use to reduce algorithmic bias as much as possible without giving up the positive features of the systems that rely on sophisticated algorithms and the machine learning on which they’re based?  Make an argument in favor of or against one or more of these solutions.  [Write about 300 words.]

 

Be sure to cite your sources.