There has been a lot of talk about Artificial Intelligence lately; bots, self-driving cars, machine learning, siri, cortana, alexa, etc. The term was coined in the 1950s and quickly became popularly thought of as machines simulating a human’s ability to think and learn.
Thanks to science fiction writers and movie makers the mind generally jumps to a myriad of books & movies. Steven Spielberg’s AI, Bicentennial man portrayed by Robin Williams, Stanley Kubrick’s “Hal” and my personal favorite, “Deep Thought” in The Hitchhiker’s Guide to the Galaxy by Douglas Adams.
In all of these the robot, advanced computer, etc can think for itself, it can receive and process data, learn and gain knowledge and apply that learning to achieve an outcome.
But while we focus on the end product of AI and predictive analytics, in the real world of today, a lot of menial, gritty, tedious and hard work goes into just sourcing the data to achieve these outcomes. The not so glamorous world of AI. I’m reminded by this as I work with a group of post-grad students from MIT’s A-Lab. As part of a class taught by Erik Brynjolfsson (Co-Author of “The Second Machine Age“) students are given the opportunity to work on real life examples to put the theory into practice. Through a generous client I’ve been able to give them access to a rich data-set. And this is where the not so glamorous side of predictive analytics and AI makes an appearance.
The data-set in question lives in some legacy system where the data can’t be accessed directly – not that anyone would let you access data in a production system for analysis …. so we go back to the timeless technique of exporting data to a delimited text file and going through a file transfer mechanism we all know too well. The machines can do some of the work, but us humans have to connect the dots for this one-off process. Siri has not yet been taught to understand “Siri, grab 5 years worth of data from that legacy system and get it over to my students at MIT for analysis”.
So now, we eventually have the file, it’s time to look at the data. Of course, any rich data set is going to include text, and inevitably that text is going to include a character that the machine is going to confuse for a delimiter. It’s obvious to the human that the “rogue” character is just a form of punctuation, but the machine didn’t go to school and learn about grammar.
And that’s the key to where we are today with AI. We should not be concerned about the machines taking over the world and rendering us humans useless. The machines are our students. They need to learn, and we are the ones to teach them. All those things we do instinctively, that knowledge needs to be codified before machines can think for themselves. And that is a lot of nitty gritty tedious work.
Ultimately my students and I are hoping to develop the algorithms to better understand and predict the consumption patterns of thousands of supplies for individual medical facilities. The goal is to utilize advanced data & analytics to optimize the supply chain – so that when the child with the cut on her knee sees her Doctor, he can apply the right Dora the Explorer band-aid to make everything alright.