009 For the Love of Algorithms – For the Love of Data

Worst Pun Ever: Today, we are talking about Al… Al Gore… Al Gore Rhythms… Algorithms!

Definition: a step-by-step procedure for solving a problem or accomplishing some end especially by a computer¹

Inputs:

Many algorithms use census data or FICO score as one of their prime inputs
Plus any custom information you give a website
Plus any information they glean about you from other sites (when you visit a site with a Facebook Share button, Facebook can track that you’re there¹⁷)
Websites are constantly looking at ways to break our anonymity (fingerprinting) so they can track us and serve us more relevant or lucrative ads⁵.

Fun Stuff

Chess – algorithms are so good that humans haven’t been able to beat a 4- CPU PC since about 2005⁶
Rubik’s Cube – the machine record is 0.887 seconds vs. just over 5 seconds for a human⁷

Poker – scientists solved all moves for Heads Up Limit Hold ‘Em – 3.16 x 10^17 moves. You may be able to win individual games, but it is HIGHLY unlikely that you can win over time⁸

Machine Learning

Can include intentional or unintentional bias.
@JonathonMorgan did a post on Medium and a podcast on Partially Derivative about using a machine learning model to find alt-right white supremacists on Twitter and track their degree of radicalization over time. He did this by training a model with their tweets and analyzing their usage of words like “Jewish” vs. more mainstream usage^3,4.

Pricing²:

Amazon lists its results over competitors, even when higher including shipping for non-prime customers; however, it claims it’s algorithm is customer-centric
Princeton Review charges between $6,600 and $8,400 for its online course in some zip codes. It charged higher in zip codes with higher incomes and some with higher Asian populations.

News/Search:

Link Analysis, how two entities relate to each other is used by Google’s PageRank, Facebook’s News Feed, and LinkedIn’s job/connection recommendations. It was developed in 1976 and first used by two other search indexes before Google began using it in 1998.¹³
However, algorithms cause sites to cater to information similar to your preexisting views, or for what they think you will find interesting, rather than presenting balanced, holistic content.¹⁴
- Medium recommends articles based on how long it thinks you will read.
- Some sites tailor related content, content types (video, etc.), and sharing buttons based on where you enter their site from.
- These choices and filters can lead you into a content bubble that leads you down a path of more and more specific, and sometimes extreme, viewpoints.
Facebook uses hundreds of features, or input variables, when assigning a relevancy score to posts you see in your news feed.¹⁵
When you have a Facebook account and you visit a page that has a like or share button, Facebook can log your visit and use that to tailor content or ads when you visit their site.¹⁶ See here¹⁷ for a relatively up-to-date list of features used in Facebook’s newsfeed algorithm (time spent viewing, friend’s posts receive priority, likes/reactions, etc. are all key inputs).

Serious Consequences

Some algorithms for car insurance weight FICO credit scores higher than drunk driving convictions.^9,10,11
Cathy O’Neil calls them “Weapons of Math Destruction” (WMDs) if they are: widespread, secretive, and have the potential to do great harm^9,10
Kronos, a small big data HR company hired by large firms to screen applicants employs a personality test as part of their screening of candidates. Some argue that this unfairly excludes them from jobs, with no explanation of the reason, in a manner that violates the American’s with Disabilities Act (ADA).^9,10,12

Theme Music: Algorithm of Desire by Measles Mumbs Rubella, courtesy of FreeMusicArchive.

Sources:

Podcast: Play in new window | Download