When your enterprise sets out to build an artificial intelligence and machine learning team, are you targeting the right people to hire? Or is it possible that you are seeking to reinvent the wheel, or the microwave oven?
Many organizations today may indeed be hiring the wrong skills when they look to build out that machine learning team, according to Google Cloud Chief Decision Scientist Cassie Kozyrkov, who provided an overview of applied machine learning at Google and beyond during a keynote address at the Strata Data Conference in New York last week.
First, Kozyrkov provided a simple and elegant definition of machine learning as a second way for humans to communicate with computers. The first way is through instructions by coding. The second way is by providing the computer with multiple examples. That is what machine learning does.
Kozyrkov noted that humans, too, teach each other with these two methods -- instructions and examples. Previously, humans only used instructions to teach computers. Now humans are adding this second method of examples.
Much work has already been done to teach computers with examples, creating machine learning algorithms. Yet rather than creating engineers who can build on the work already done, so many universities are teaching students to build those initial algorithms again.
Kozyrkov used the analogy of microwave ovens. You don't need to take apart a microwave oven and see how it is made to be able to cook with it, or to make sure that it will cook your recipe correctly. Yet when universities train engineers, they are not teaching them to cook with the microwave, they are training them to build more of the appliances.
"We end up training people to make more microwaves," Kozyrkov said. "Then when you hire them into your kitchen, they end up wanting to build you a microwave. But there's already warehouses upon warehouses of microwave appliances already there."
What you need is someone to innovate with new recipes. What you need is someone to apply the technology that has already been built. What you need is someone who knows how to use machine learning to achieve business outcomes.
Google is trying to change this with a new approach, according to Kozyrkov.
"We have started training our personnel in applied data science and applied machine learning, and we are calling that decision intelligence engineering," she said.
"This is about taking all those applied machine learning principles and augmenting them with insights on how to make this useful for this business. It focuses on using data to solve business problems," Kozyrkov said.
A very public example of how Google has applied machine learning is in cooling its own data centers. When the company applied machine learning to automatically control the cooling of data centers, it achieved a 40% energy efficiency improvement
But no machine learning practice is complete without testing, Kozyrkov said. These algorithms should be tested the same way you would test students -- with brand new questions they haven't seen before. Testing is the best basis for establishing trust.
For instance, if you want to be confident the airplane you are flying in won't crash, you could read all about aerodynamics and examine how an aircraft has been designed and manufactured. Or you could look at that aircraft's record -- has it successfully flown many times?
"I'm all about testing," Kozyrkov said. She advises organizations to test everything. "Blind trust is a terrible thing. Force the algorithm to earn your trust by testing."