As a corporate executive, you’ve undoubtedly heard about Machine Learning (ML) and its potential to transform entire industries. However, most of literature about Machine Learning is either extremely technical or highly research specific. This is meant to be a practical guide on how you can think about implementing ML in a practical way within your organization.
What is Machine Learning?
Machine Learning is the study of teaching computers to program themselves. Historically, no matter how advanced an application may seem, a human programmer had to explicitly account for every branch of logic. This created a natural limit on the effectiveness of these applications. Over the past decade or so, a new way to program computers has emerged. Instead of telling a computer explicitly how to do something, ML algorithms program themselves by looking at training data to figure it out. For example, instead of a human writing an algorithm to determine if a picture contains a cat, an ML algorithm trains itself on millions of images of cats to figure it out itself.
This may seem like magic, and a common misconception is that there has been some revolutionary breakthroughs to enable this innovation. In reality, most of the ML algorithms have been around since the 1970s. The main enabling factors for the rise of machine learning have been exponentially cheaper access to computing power and an equally exponential rise in the amount of data being collected. The scale of this is absolutely staggering. It’s estimated that 90% of the world’s data has been collected within the last two years.
What can Machine Learning do for my organization?
There are two main ways that ML can help your organization.
Every organization has business processes that need to be optimized. Every business decision has the aim of maximizing gain (i.e. profit) given a set of constraints (time, budget, etc). ML is great for these type of problems. An example of this is Google using ML to optimize their data center power usage, which saved them hundreds of millions of dollars per year. Netflix uses ML to improve search results and save billions of dollars per year in user churn. Industrial firms like GE and Siemens are using ML to identify manufacturing anomalies and predict failures.
Another application of machine learning is to automate business practices that are currently being done manually. For example, most businesses have to process massive amounts of paperwork. This often amounts to humans reading through documents and inputing them into a computer system. Efforts are currently underway to replace this human effort with ML algorithms that can do the same thing in a fraction of the time, with much fewer errors.
Investment in Machine Learning is providing tangible ROI across nearly every industry. In addition to integrating ML into existing products and services, more and more companies are using ML to streamline internal operations.
What are the business risks of implementing ML
Since Machine Learning is a relatively new technology, there are a myriad of risks that a business leader must consider prior to implementing it within his or her organization. Because ML relies on large amounts of data, cybersecurity is a top concern. Researchers have found that some ML models have a hard time detecting adversarial input, essentially bad data designed to fool models. This allow hackers to potentially reverse engineer ML data sets and algorithms to steal intellectual property or get past security mechanisms.
Another top concern is trusting the recommendations made by ML models. Currently, ML is great at figuring things out based on certain criteria but terrible at explaining the reasoning behind it. This is an area of research at the moment but not quite ready for prime time.
How do I get started?
At Sensai Group, we advocate a multi-disciplinary approach to ML projects. Most ML projects that we take on is adding ML to an existing product line. This requires cooperation between several disciplines: UX Design, product, software development and data science. The reason for this is that most ML projects are made up of a series of what we call APIM (analyze-plan-implement-measure) loops. These loops require the cooperation of multiple disciplines in order to be successful. The analyze step consists of looking at existing data and third party integrations and cleaning up the data. The plan step is determining the right models to use to accomplish the objective. The implementation step is where the models are plugged into the existing application architecture, and potentially additional product changes are implemented to collect more data. The measure step is when we compare metric before and after implementing our ML models. For more details check out my colleague Tiger’s post here.
If you would like to learn more contact us at firstname.lastname@example.org we’d be happy to set up a free consultation.