Maintaining Your Competitive Advantage – Machine Learning
This is part two of my three part blog series titled “Maintaining Your Competitive Advantage”. For part one, please click here.
Machine Learning (AI)
The phrase Artificial Intelligence often elicits imagery from 1980’s science fiction thrillers. In recent years, it is often misapplied to any task that a machine can complete autonomously.
But perhaps one of the biggest misnomers is that AIs think. This is rooted once again in the recurring science fiction story, where an “evil” artificial intelligence becomes sentiment and declares war on the human race. However, machines do not think, at least not in any way representative of human thought.
A good analogy to explain this is a submarine. Submarines move rapidly through water, but they do not swim. Just as submarines do not swim, AI’s do not think. Rather, like the submarine that moves through water through propulsion technology, machines achieve their goals through self-improving pattern recognition algorithms and not through biological and chemical processes that power human thought.
In this context, let’s focus specifically upon machine learning. Just as the phrase implies, machine learning is any algorithm in which a machine is able to learn by trial and error, improving its accuracy in pursuit of eventually solving the problem. Machine learning requires pattern recognition algorithms to identify paths to problem resolution, and a feedback mechanism (often managed by humans) to indicate mistakes.
That’s Cain. He’s a dog. Specifically, he’s my dog and he’s a Rottweiler. Cain is a smart dog and he learns quickly. Cain’s goal is to obtain a delicious treat and in order to receive the treat, he needs to learn how to play dead. There are several steps to playing dead.
Cain needs to identify the prompt, then to lay down, roll over, fold his paws, close his eyes and stick out his tongue.
Each time that Cain makes a successful step toward meeting his goal, he is rewarded, and when he fails, he is not. Eventually, Cain begins to eliminate actions that do not further is goal, and string together actions that bring him closer to his goal.
Once Cain is able to play dead, he receives the delicious treat and remembers the full series of steps to achieve the goal in the future.
A machine learning AI, in fact, is much like Cain. Using neural networks, an AI starts to develop patterns in pursuit of a specific goal. At first these patterns are quite random, but with corrections and enough computations, the AI begins to get smarter. Eventually, it solves the goal and learns the new behavior.
Machine learning algorithms can be used to solve some of our industries’ most challenging and expensive problems. We use machine learning in the healthcare industry, for example, to help our customers detect fraud and eliminate waste, problems that cost the industry billions of dollars every year.
Like Cain, the AI starts by making what seem to be guesses. Each time that the machine incorrectly identifies fraud, a human corrects it. With each correction, the AI becomes smarter, improving accuracy. While accuracy can never be expected to reach 100%, the machine eventually becomes far more efficient than any human in identifying healthcare fraud.
In scholarly publishing, the problems are again different. Publishers want to identify, maintain and attract the best authors, researchers and papers. One way to do that is to identify which article submissions will have the biggest impact. With machine learning, an AI can analyze tens of thousands of papers and identify patterns that made them impactful. It can predict trends and suggest more efficient workflows. With time and sufficient course corrections, it can become quite proficient.
In my next blog post I address the third technology that is creating a competitive advantage for our customers. Blockchain.