Responsible and Ethical Search in the Real World

When we are releasing search algorithms into the real world, that is when we need to be especially careful to ensure we have developed the solutions in an ethical and responsible way and that they do not cause harm!  In some sense, we covered the idea of real-world deployment of searches in the first module on ethics in search, when we discussed the unintentional side effects of route-navigation sending people through quiet neighborhoods.  For this module, we will focus on the nature of the sub-optimal solutions identified through local search.

Sub-optimal solutions

Real-world search is challenging, as you no doubt have learned in this module!  The methods we discussed in this module are not guaranteed to provide optimal solutions but they may provide sub-optimal solutions in a much faster time than a truly optimal search would take (if it was even possible, for the specific application).  This tradeoff between speed and optimality comes with some ethical implications.

In order to discuss sub-optimal solutions, I want to introduce one other concept that we haven’t discussed in class so far. It is called satisficing.  The term was introduced by one of the pioneers of AI: Herb Simon.  The idea is that you are making a decision that is “good enough” though not necessarily optimal.  You can read about it in the two links below.

• Wikipedia on Satisficing
• Investopedia on the satisficing in a business context

For this topic, we are going to look at a case study where sub-optimal solutions can have serious real-world implications.

Case study: Unintentional Racial Bias in Healthcare Algorithms

A recent (2019) algorithm was developed to help with healthcare management decisions.  Unfortunately it had unintended side effects!  To jump into this case study, read the following articles (they should be freely available both on and off campus since this is not a recent article, if you run into issues of it needing you to pay for the articles, let me know and I’ll post pdfs to canvas or slack).

• Dissecting racial bias in an algorithm used to manage the health of populations (Science)
• A Health Care Algorithm Offered Less Care to Black Patients (Wired)
• Racial bias in a medical algorithm favors white patients over sicker black patients (Washington Post)

Assignment

Part of this assignment will be different for fully online versus hybrid students and part of it will be the same.

Fully online students

• Part 1: Go to the canvas link (just so the link isn’t open to the world) and help to write a group policy proposal for policy that could be implemented to keep this problem from occurring again.  Note, your policy link is different than the fully online students because their part 2 is to debate it while your part 2 is to try to persuade each other to vote on a final document.
• Part 2: In slack on #ethics, try to persuade people to vote on the finalized policy document.  Bonus for everyone who votes if you “pass” a good final document by the deadline!  Declare that you participated!

Hybrid students

• Part 1: Go to the canvas link (just so the link isn’t open to the world) and help to write a group policy proposal for policy that could be implemented to keep this problem from occurring again.  Note, your policy link is different than the fully online students because of part 2!
• Part 2: Come to class prepared to debate the policy, as if it was something we wanted to pass in congress.  I will assign you randomly to different sides of the issue when you arrive at class.  This will be graded as participation only, not based on your viewpoints nor on your debate skills.  This is not a debate class, but it is a class where you need to learn to think critically about issues and that is what we will be doing!  Declare that you participated!