The overwhelming majority of human capability and function has to be learned. Breathing, for example, is something involuntary that ideally, we never have to think about doing. But whether it’s the development of an infant, or the continuing education of an adult, new information has to be presented in order to learn and grow.
Now often times this process is facilitated by another person taking on the role of a teacher. First your parents, then school teachers, friends, mentors, bosses, and so on. But when you apply this concept to a machine, their teacher is always represented by the introduction of new data and information.
A simple machine learning example is the predictive text algorithm in the keyboard on your smartphone. The more you use the keyboard, the more it starts to pick up on the words and phrases you use most. After a little while, it will get better at suggesting the next word in your sentence.
Your smartphone keyboard is an isolated example of Artificial Narrow Intelligence (which I cover in another post) but it also uses Machine Learning (ML) to get better at predicting your text. A more sophisticated example of ML is Amazon—the more you purchase from the online retailer the more it starts to understand your interests and will start making recommendations. Amazon even uses ML to optimize their inventory based on all its users shopping data.
An AI can perform a simple task indefinitely; an AI that is equipped with Machine Learning can find new and more efficient ways to perform the same task over time.
Machine Learning is a subset of Artificial Intelligence that gives systems the ability to learn and improve from datasets without being explicitly programmed to do so. In the same way a human learns, the more data that a machine learning AI has on a particular topic, the more adept it becomes in that area.
One early example of AI is commonly found in airplanes, otherwise called “autopilot”. The autopilot of an airplane knows not to fly into a mountain, which is great considering a human pilot may only spend seven minutes actively controlling the plane. But think if you applied Machine Learning to the AI in a plane, fed it as much flight data as you could, what would it learn? A faster route between D.C. and New York? More accurate arrival and departure times? A plane with no human pilot at all? It shouldn’t surprise you to learn that all of the above are possibilities currently being explored by the major airline carriers.
To show the tremendous growth potential of ML, Google once reported an error rate of 23% in their speech-to-text recognition software in 2013. In July, 2016 it was 8.5%. And last year in May, 2017, it was 4.9%. So, while an AI can save you time and effort by automating highly repetitive tasks, an AI that utilizes Machine Learning can exponentially improve over time.
When it comes to our own product, Shai, specific tasks and functions can be taught, using the exact same elements a human would require such as employee IDs, individual passwords, and desktop software. But Shai has the capability to learn and improve on the tasks assigned, creating the potential for evermore efficient solutions. Thanks to Machine Learning, Shai can sort through massive data sets to find new patterns and answers to complex issues, and improve overall operational efficiency.
And that’s just the beginning.