Machine learning may seem like a high-level concept, one reserved for the brainy experts at M.I.T. However, its capabilities are widespread, and can be harnessed to provide real power to everyday companies. Integrating machine learning into a product can easily be done without breaking the bank. To get started, it’s vital to understand the basics of machine learning as well as some practical examples of how machine learning works and where it can be used.
It’s a common query to ask the difference between machine learning versus artificial intelligence (AI). In truth, machine learning is actually a form of AI. However, machine learning doesn’t actually think for itself. It uses programmed directives to analyze and automate tasks that companies used to rely on humans brain to do. Machine learning products do this by recognizing and identifying patterns with some introductory help from developers. They set the program up by inserting machine learning algorithms that teach the system how to do its work. Typically after that, it’s off to the races for the system as it analyzes and organizes given data.
As powerful as machine learning tools are, many myths exist today. Many product managers think the barriers to entry are just too high, or that it’s unusable from their business’ end. Usually, this fear is rooted in one of two concerns:
1. Machine learning technology doesn’t really exist.
Machine learning is used every second of every day in a wide range of processes, including everything from building smarter websites to choosing playlists on streaming services.
2. If this technology did exist, the cost of implementation has got to be prohibitive.
Nope. There are lots of plug and play machine learning applications that can be integrated into your current software, thus eliminating the need to develop anything by hand. In reality, many machine learning products can actually save money. Think about the cost of hiring people to individually tag an unending mountain of photos that users share on social media platforms. Machines can do that task 24/7— in a fraction of the time.
In general, machine learning falls into one of three categories. The type of machine learning algorithm you use depends on what kind of goals you are trying to accomplish.
This is what most practical applications use. Essentially, the software is given a defined input variable as well as options for what it analyzes and can then output. Overtime, as it gets more information, it will begin to get better and better at recognizing those patterns so that eventually it can be left to its own devices.
Let’s say the goal is to use machine learning to identify different types of animals. It would be given the known number of classes and features. If the algorithm is given the following features such as land-born, has fur, performs live birth, and has warm blood, any animal that matches these criteria would fall into the class of “mammal” and be labeled as such. Just as an animal with the features of water-born, lays eggs, and breathes with gills would be classed as a fish.
This is where one starts with an algorithm and then feeds the computer a lot of data and it spits out different ways to categorize and organize that data. The goal here is simply to learn more about the data that it’s given.
Again let’s take the example of animal species and see what unsupervised learning would do. In this scenario we would input all the information we have and see what the algorithm comes up with. Here we learn that in 2018, there were 220 new animal species and 9 plant species discovered. This result comes from the fact that out of all the data given, 220 animal and plant species didn’t match the existing data points and could not be classified as anything that was known to exist.
This is where one starts with an algorithm that is then given negative or positive reinforcement as it analyzes data. Think about how children learn to identify animals. At first everything with four legs might be considered a dog. They are then told that “no, that’s a cat,” or “yes, that’s a poodle.” Eventually, with enough guidance, they will soon be able to tell the difference between a basset hound, golden retriever, and more.
There are a few different ways that that machine learning can make existing digital products more effective, enable features that hadn’t been thought of before, and inspire all new products that many might not have thought possible.
Websites that provide stock photos and social media websites have vastly improved their search capabilities thanks to machine learning. If, for example, one of these websites needs to search for a certain picture in a database of many different images, machine learning can be used to make the process more accurate and faster. The machine learning algorithm will go through and tag photos using a couple of different categories so that when users search, it will be more likely that they find the picture they are seeking.
How can machine learning inspire new ideas? Spotify was at one point solely an app that let people listen to their favorite music on demand. As a result of being a great idea, many people bought into it and began to stream on a daily basis. With that said, the music streaming market has grown to be a market of many competitors. In order to stay ahead of the curve, Spotify used machine learning to analyze the data of what people were listening to. By using this approach, they realized they could offer a brand new feature that their customers would love: suggested playlists based on listening history.
Machine learning also helps drive new technology, literally. Without it, the driverless car would still be a pipe-dream. A driverless car needs an extensive amount of data to function. It needs to be able to predict other drivers, driving conditions, and how to get from point A to point B efficiently in the real world. Thanks to machine learning, all of that is possible. Driverless cars can recognize and account for fast or reckless drivers. They can also compensate for icy conditions and recognize road signs. This is made possible through extensive GPS data and thousands of hours worth of research.
While there is a lot that machine learning can do, it’s important to understand notable limitations to today’s machine learning technology.
The first big limitation is having the data to analyze in the first place. Product managers shouldn’t assume that users are going to give up their data easily. If users are expected to fill out a bunch of queries in order for machine learning to have something to look at, there won’t be much to go on. Instead, what product managers need to do is incorporate data science from the start. This will give them some idea of how exactly the data that the machine learning algorithm needs will be harvested.
Additionally, machine learning is still plagued by some inherent biases that might affect its ability to produce an accurate outcome. As a result it is important to understand and review for any bias that may occur. There are a few different kinds of biases to look out for. The most common forms of machine learning bias have the potential to delay or negatively impact a project due to either skewed results or just plain giving out inaccurate information.
This misses major key points like the “why” of the project in the first place. This type of bias occurs when there is a flaw in the algorithm resulting in data that might look correct, but isn’t.
This is when data does not represent the realistic group at hand. It’s the same idea as comparing apples and oranges. Imagine trying to calculate a route to run somewhere; it wouldn’t make sense to use a map of the highway.
This type of machine learning bias can skew data based around gender, race, or socio-economic background. If there are sets of pictures where more men are skiing and more women are swimming, then the machine learning will skew the results. This is because it will assume that more men ski and more women prefer swimming.
This bias may be due to poorly gathered data or by letting the machine learning algorithm control the measurement of data. Faults in measurement lead to skewed data. For example, providing a survey with leading questions will influence responses and make them less genuine. As a result, the skewed data should not be used for machine learning.
Machine learning provides many advantages that will help product managers develop better products. While there are still plenty of kinks to work out, there are many ways that product managers can take advantage of these resources today. We’ve seen lots of different ways that machine learning can help clients, while not taking too big of a bite out of their budget. If this seems like something that you would be interested in, we’ve helped lots of clients dip their toes and then dive head first into the wide world of machine learning. Contact Crafted today to learn how we can help you do the same.