I’ve been on a bit of a SciFi kick lately, and Amazon knows it. When I log on to my Amazon account, I’m greeted with these recommendations. No surprise: it’s more science fiction novels. What is surprising is that I’ve never actually read or owned any of them, making these likely candidates for books I might buy next.
Amazon knows I own Douglas Adams’ The Hitchhiker’s Guide to the Galaxy (since I bought it from them in April). Because I already own another book in the series, So Long and Thanks for All The Fish is a safe bet. I also recently listened to Andy Weir’s The Martian on Audible (which is owned by Amazon). This means another book about Mars, Red Mars by Kim Stanley Robinson, makes sense as well. Lastly, Do Androids Dream of Electric Sheep? is one of Philip K. Dick’s most popular books, and Philip K. Dick is a very popular SciFi writer. Anyone who is a fan of SciFi is likely to be a fan of this book as well.
How was Amazon able to know all this and make such intelligent recommendations for me? The short answer is machine learning. It seems similar authors, similar topics, and similar genres are among the criteria used in selecting recommendations to advertise. Moreover, users who bought one of the books I already own also bought the books above. If you completed ESS 101 or DEV 362, you know this type of machine learning algorithm is called collaborative filtering, but is more commonly referred to as a recommendation engine.
Recommendation engines allow companies like Amazon, Spotify, and Netflix to guess which books I want to buy, which songs I want to listen to, and which shows I want to binge. In 2006, Netflix announced a $1,000,000 prize for any team that could improve their recommendation algorithm based on user ratings of films, and in 2009 a team won the Netflix Prize by improving the accuracy of the algorithm by over 10%.
Machine learning has continued to improve since the Netflix Prize. In 2012, Target’s recommendation engine predicted a pregnancy before the rest of the mom-to-be's family knew. But recommendation engines are not the not the only kind of machine learning:
- In 2011, IBM’s supercomputer Watson beat Jeopardy champions. Watson uses a combination of natural language processing and regression models to search through vast amounts of data and choose the most likely answer.
- In 2014, Facebook released facial-recognition software that identified users with over 97% accuracy. Facial recognition uses deep learning to find faces and match them with tagged pictures of Facebook users.
- So far in 2016, Google’s artificial intelligence AlphaGo beat the world-champion human player at Go, a game which is notoriously difficult for computers to understand. AlphaGo also uses deep learning.
What are your favorite machine learning applications, real or fictional? Share them in the comments below!