What’s machine learning? That’s a complicated question. So where do we begin?
The machine learning story is not a simple one, especially given the fact that what goes by that name in 2019 differs in many ways from the machine learning technology of the past. What’s more complicated is the fact that the internet is bursting with descriptions of machine learning at every imaginable level of expertise and abstraction. Try punching the term into a search engine and expect to find yourself digging through a mixed bag of technical discussions, academic publications, user forums, blogs, tech reporting, and so on.
To cut through the noise and clear away confusion, it’s best to establish a clear foundation. What we can do is start with a simple, working definition of machine learning and use it, and the basic concepts connected with it, to better understand some fundamental issues around this rapidly developing technology and its growing practical applications.
This article provides a brisk but deep dive into the world of machine learning by addressing the following key issues:
- What is it?
- How it differs from past technologies?
- How it is being used to guide decision-making and improve performance within many different kinds of organizations?
- What are some practical examples of applied machine learning?
Machine Learning: What Is It?
Perhaps the simplest, most general, but still helpful definition of machine learning (ML) is that it is the science of teaching computers to learn in ways that can help human users make better, more well-informed practical decisions. To be more specific, ML embraces a large and growing set of methods and technologies for using ever-more sophisticated algorithms to gather and analyze ever-greater varieties of ever-bigger data.
In this way, ML systems help users make sense of data and arrive at determinations or predictions about future outcomes based on data analysis. That is the practical value of ML, and in this sense, ML is nothing new. Businesses and other organizations have always needed to understand the terrain on which they operate, make reliable predictions about how operating conditions may change in the future, and use these insights to allocate their finite resources in order to optimize their chances of success.
Machine Learning: What’s New?
Where ML differs is in the way it helps users pursue the age-old goal of leveraging good insight to inform decisions and optimize performance. What so new about machine learning as we know it today?
A large variety of sources converge on an answer: machine learning is a type of artificial intelligence that enables modern computers to perform complex tasks, and refine and improve their performance over time, without relying on programming.
That’s the difference-maker. Whereas computer systems of the past relied on explicit instructions, ML systems use AI to identify patterns in data sets and automatically learn from what they observe in order to make inferences and act accordingly.
This capability can seem a little unsettling. It’s one thing for computers to mimic human intelligence by, say, excelling at chess and even outperforming the greatest human grandmasters. Most of us have gotten used to that idea.
But it can be another thing entirely to discover that computers can now learn like humans, to the point of enhancing their own capabilities through autonomous data gathering and analysis. Indeed, it turns out that modern chess-playing computers continue to outpace their human opponents by using machine learning AI to teach themselves how to play like human beings—only better than the best humans can. The meteoric evolution of chess-playing programs suggests another way to frame the difference between leading computer systems of the past and modern AI-powered machine learning.
Older technologies could overpower humans in terms of brute force. They could summon more raw computational power, for instance, or (to stick with our chess example) analyze many times more chess moves per second than their human counterparts. But due to their ability to recognize patterns and act and adapt accordingly, ML systems are capable of something much more like creative or intuitive learning. They are often less muscular, performing fewer computations per second, than their AI counterparts of yesteryear. But their new and unique, human-like knack for pattern recognition gives them the ability to develop novel ways of handling big data sets and performing complex tasks.
Machine Learning: Its Uses & Applications
When someone first discovers that modern AI systems are capable of these impressive feats of automatic and adaptive learning, it’s natural for them to immediately begin wondering what ML applications might impact the field, industry, or area in which they operate.
For the remainder of this article, we’ll begin to address this question by reviewing two key current areas of application for ML.
One flourishing area of application for ML is called predictive analytics, or the extraction of patterns from data sets in order to support predictions about future trends and developments in a given area. Statistical algorithms are the tools of choice for predictive analytics, which must contend with the bigger, faster, and more diverse flow of information that defines our era of “big data.”
Crucially, it’s not necessary for organizations to employ a statistician or data science expert in order for them to leverage predictive analytics tools to gain a competitive edge. Many such tools now exist as user-friendly software platforms that can be used by non-specialists.
Predictive analytics tools have a wide variety of practical applications. They can help organizations solve old problems, avoid future ones, identify new opportunities, and better understand the terrain on which they operate.
For instance, ML’s pattern recognition power can be used to uncover abnormal behavior in financial networks, resulting in more effective fraud detection.
Dataiku and Teradata are two companies that illustrate how new ML techniques are being developed to aid financial organizations prevent and detect fraud. Managers or fraud detection teams at banks can use the ML services provided by Dataiku in order to gather and analyze the large amounts of raw data produced through their routine operations and interactions with clients.
Non-technical staff can benefit from predictive insights about data patterns by using dashboards and other software tools that flag anomalies in transactions and suspicious client activity with notifications, prompts, and alerts.
One way this works is for users to upload data into their analytics platform, which then organizes and formats the information into a spreadsheet. The results can be visualized in a range of ways using charts, graphs, and other formats. For instance, traits (such as age or gender) can be associated with data to shed light on past behavior and help predict future behavior.
Predictive models can boost a company’s efficiency by more accurately predicting sales activity and inventory problems, facilitating resource allocation, and driving success in other areas of business operations.
For instance, predictive models are being used to turbo-boost marketing campaigns. Data-driven insights help marketers attract and retain the best audience for their portfolio.
ML platforms can be used to gather and interpret raw data about customers to provide actionable insights about the success of different marketing campaigns on social media. They can also provide data about trends in buying behavior in different online venues or among different populations. Knowing what does and doesn’t work in different areas helps marketing teams design outreach strategies with better chances of playing well to the audiences being addressed.
Finally, risk reduction is another benefit offered by predictive analytics, which can be applied to insurance claims or credit scores to determine the likelihood of default and other factors pertinent to creditworthiness.
Modern computer systems continue to evolve in revolutionary ways. Artificial intelligence outstripped human speed and raw computational power decades ago. The latest AI milestone is machine learning, where programs use statistical algorithms to learn and derive insights from huge data sets without relying on explicit programming.
One thing to bear in mind is that for all their power to boost efficiency and potential to transform organizational decision-making, algorithms, and therefore the AI and ML systems that totally rely on them, are not perfect or immune to error. Their reliability will continue to depend substantially on the quality of due diligence practiced by their human users. A topic for another time!
Was this helpful?