By Dr. Randal S. Olson
Whether it’s reducing the time you wait at a traffic light, allowing your phone to respond to your voice, or preventing fraudulent charges to your credit card, it’s becoming apparent that our systems are increasingly being managed by the machines we built to operate them. And surprisingly, they’re often proving to be better at it than we are.
As someone who has spent the last ten years of my life devoted to developing and using artificial intelligence, machine learning, and data science tools, I’m comfortable with that trend. It’s making us better at a lot of important tasks, including the work I’m doing at Life Epigenetics, where we’re using machine learning to analyze epigenetic biomarkers from saliva, a process that will lower the cost of life insurance.
They’re also great tools to have some fun with. For instance, I used these tools to create a strategy to find Waldo as quickly as possible in the “Where’s Waldo” books, to optimize the driving route on road trips that will take you to all of the lower 48 U.S. states, and to determine the optimal attack strategy in the board game Battleship. All of these projects were featured on my blog and received some attention from others who weren’t necessarily data nerds. Whether you think those projects have any value is an issue for you to decide, but I couldn’t have completed them without some form of artificial intelligence, machine learning, and data science.
One thing to keep in mind is how these terms can be used and misused. Allow me take you through my world, the terms and the science that will help you understand what they mean, and how they can apply to your work and life.
Artificial Intelligence (AI): At a high level, artificial intelligence is an umbrella term for the dozens of approaches to creating intelligent machines that can perform a task and become better and more efficient at it without our direct intervention. AI doesn’t refer to any particular approach to creating intelligent machines, so saying that a problem was “solved with AI” makes about as much sense as saying that a product was “sold using business practices.” If someone tells you they are using artificial intelligence to solve a problem, make sure to have them explain how they’re using it and what kind of artificial intelligence they’re using. It’s important to be specific and say what kind of AI was used.
The best example of AI in everyday life is speech recognition, the technology that is heavily involved in Siri on iPhones. Whenever you use Siri, it has to record your voice and then translate it for a computer to understand and process. We take that process for granted, but it wasn’t long ago that speech recognition couldn’t be done quickly nor accurately enough to be used in everyday life. Now, speech recognition technology is transforming interactions such as automated phone systems that provide customer service at a fraction of the cost of human operators.
Machine Learning: Machine learning is a subfield of artificial intelligence that seeks to give machines the ability to learn to perform a task without manual – or human – guidance. Most often, machine learning algorithms learn to perform a task by analyzing data and building an algorithm that accomplishes that task. One amusing example of machine learning is an actual iPhone app based on an episode of the HBO series “Silicon Valley.” It uses a machine learning algorithm to look at an image and tell us whether or not there’s a hot dog in it.
Hot Dog or Not Hot Dog may seem easy, or at the very least dumb, but it’s a good example of how most machine learning algorithms work. To start, the system was given pictures of hot dogs – I know this sounds ridiculous, but stay with me – and asked to learn what to expect when a picture has a hot dog in it. Next, the system was given pictures that did not contain hot dogs – like someone’s arm with ketchup on it or a sausage next to a bun – and asked to learn what to expect when it viewed pictures that don’t contain hot dogs. The system had to view thousands of images of Hot Dogs and Not Hot Dogs, which enabled it to build an algorithm that takes any input image on an iPhone and decide whether it contains a hot dog or not.
In a sense, most modern machine learning should be thought of as a tool that learns to produce predictions from data just in the same way as this Hot Dog or Not Hot Dog algorithm. There are other machine learning approaches that don’t fit into this description, but they are currently much less commonly used in commercial applications.
Of course, there are more serious applications of machine learning than looking at an image and deciding whether there’s a hot dog in it or not. At Life Epigenetics, for example, we’re integrating existing underwriting information—such as driving records, financial records, and health records—with epigenetic biomarkers from saliva. Our machine learning algorithms can use this information to make individual-level predictions about significant personal health issues, such as disease risk, smoking habits, and drug and alcohol abuse habits. We then use those predictions to provide low-cost life insurance rates for our customers, which makes life insurance more accessible to the public.
Data Science: Data science is a practice that’s focused on drawing insights from data using a variety of analytical techniques. Oftentimes a data science project requires only basic data analysis, such as visualizing counts or percentages in a dataset. One example of basic data analysis is when I discovered the deadliest movie actors of all time by visualizing the MovieBodyCounts.com database, which counts the number of on-screen kills that every actor has made in movies. It’s important to know that although a data science project may sometimes benefit from the use of machine learning for prediction tasks, the two terms shouldn’t be used interchangeably. Machine learning is just one tool in a data science practitioner’s tool belt.
To sum it all up, you can think of these terms like this:
- Artificial intelligence is an umbrella term that encompasses machine learning and other approaches to creating intelligent machines.
- Machine learning is a tool that produces predictions from data.
- Data science is the practice of drawing insights from data.
With how quickly the data science and artificial intelligence fields are growing, it can be difficult to keep track of the terminology. Hopefully this article will serve as a helpful reference when thinking about these terms and how to use them.
Dr. Randal S. Olson is the Lead Data Scientist at Life Epigenetics, the insurtech subsidiary of GWG Holdings in Minneapolis, Minnesota. He invented and developed the award-winning Automated Machine Learning (AutoML) tool TPOT, which is one of the most widely used open-source AutoML tools in the world.