Updated: Jul 11, 2020
My Amazon Echo helps take care of many of my tedious tasks, whether it’s turning off lights, playing music, or letting me know what the weather is like on a particular day. Though it’s frustrating when Alexa misunderstands me, I’m often astonished at how much Alexa can process. As a coder, I’m used to converting my thoughts into code – the computer’s “language.” But natural language has always been very different, and how we communicate as humans is vastly different from code. There are idioms and intonations, emotions and accents. While most of the commands Alexa processes seem basic, when I think of the number of accentuation and dialects it must understand, I can’t help but wonder: how was that coded and programmed? Researching and reading online made me realize that Alexa actually started out “dumb” – like a more advanced version of MIT’s older ELIZA program; coders “trained” Alexa to adapt to our language by recording examples.
Just like any other machine learning algorithm, the program is trained by being fed bits of our lives, whether that’s speech, text, our handwritten numbers, or even pictures of our pets. From there, the program slowly learns more and more about our life, eventually developing its own speech/text (forming words and sentences), being able to recognize our handwriting, or having the ability to generate a picture of a dog or a cat (that isn’t actually real, simply a generic photo of what a cat could look like).
Additionally, this connects to our social media apps today. How does Instagram know me so well? How can it generate the perfect explore page for me, knowing that I’ll click on the post, and be interested in it? Even though it is borderline frightening, the programmer in me is mostly impressed. It truly is insane how much data is being collected all around us, and how effectively it’s done. The data Instagram collects can be so conclusively adaptive, as once I click on a particular post, it automatically hunts for other posts to suggest followers/accounts, or tracks my reading time and click rates.
Data analysis is truly fascinating to me, and although a tad bit stalker-y, the code to build it astonishes me to this day.