The story of a tech noob often starts like this: if I’m not familiar with how XYZ is built, I shouldn’t be using it.
This is often the case as far as machine learning is concerned. However, if we regress a little and apply the same pattern to the case of using a microwave oven, it doesn’t seem to make much sense —
If I’m not familiar with how a microwave oven is built, I shouldn’t be using it.
Many of us are not engineers by trade, yet most of us have used a microwave oven before, and this is exactly what was proposed by Cassie Kozyrkov, Google’s Chief Decision Scientist — a machine learning model is like a microwave oven. It doesn’t take an engineer to know how to use it.
Build that Winning Recipe, Chef. Not the Microwave.
In her short presentation at the TNW Conference, Kozyrkov explained: “machine learning is a general-purpose tool that helps other people solve their business problems”, much like how a microwave oven is built to make food edible. What business executives should do is not to train staff on how to build a “microwave”, but to focus on forming a winning business strategy and consider what tools would help achieve that strategic vision. In other words, in order to make Michelin gourmet, it’s more important for the chef to create that jaw-dropping recipe than to understand how an industry-grade oven is built. We can leave that to a professional engineer.
The Machine Learning Attitude – Test, test, test.
Thanks to the engineers, you don’t need to know how to build the tools from scratch. However, there’re some important aspects to think about when applying machine learning to your businesses.
In another quick machine learning talk hosted by Google, Kozyrkov demonstrated that not all machine learning models are created equal. Some models may perform better than the others for certain purposes. Let’s take an example from the world of cooking again – it’s like rice cooker versus saucepan for cooking rice. A saucepan can do the job but it’s not quite the same. So, how do you know which machine model is good for your business?
Repeat after me — test, test, TEST.
“Don’t be afraid to fail and make mistakes. Before attaining the state of perfection, a machine learning model would have to fail many times in order to know what to improve. Failure is a necessary step. Test as many times as you can.” Kozyrkov advised that this “machine-learning attitude” is something important to have when integrating the tool as part of your business strategy. If you’re too scared to make mistakes, it’s almost certain that you’ll never succeed.
Feature Engineering – Convert Domain Knowledge into Good Solutions
Finally, it’s all about data. Data is like that quality ingredient you’re about to put into a Michelin dish. The skill of a chef is crucial, but no chef can make magic without good ingredients.
In the terminology of machine learning, the process of picking the right data for your model is called Feature Engineering. Does it take an engineer to build the right features? No, but it does take a domain expert to decide what sort of data should be gathered and how to weigh their respective importance when looking for the right solution for a business problem. Again, that’s similar to building a recipe. This applies to every industry. From our very own experience as an educator, it takes not only data scientists to find out the right algorithm for providing the best learning experience for our programming students, but also the domain knowledge from education experts, researchers, and scholars.
In short, it’s not difficult to apply machine learning to your business as long as it fits into your business strategy. The key is to try until you find the right approach.
Alright cheffies, whenever you’re ready – it’s time to cook up a storm!