What is Machine Learning?

Computers aren’t taking over but they are getting smarter every day

Computer program learning traffic patterns for self-driving car technology
Machine Learning for Self-Driving Car.  Getty Images

In the simplest terms, machine learning (ML) is the programming of machines (computers) so that it can perform a requested task by using and analyzing data (information) to perform that task independently, without additional specific input from a human developer.

Machine Learning 101

The term “machine learning” was coined in the IBM labs in 1959 by Arthur Samuel, a pioneer in artificial intelligence (AI) and computer gaming.

Machine learning, as a result, is a branch of Artificial Intelligence. Samuel’s premise was to flip the computing model of the time upside down and stop giving computers things to learn.

Instead, he wanted computers to start figuring out things on their own, without humans having to input even the tiniest piece of information. Then, he thought, computers would not just carry out tasks but could ultimately decide which tasks to perform and when. Why? So that computers could reduce the amount of work humans needed to perform in any given area.

How Machine Learning Works

Machine learning works through the use of algorithms and data. An algorithm is a set of instructions or guidelines that tells a computer or program how to carry out a task. The algorithms used in ML gather data, recognize patterns, and use analysis of that data to adapt its own programs and functions to complete tasks.

ML algorithms use rule sets, decision trees, graphical models, natural language processing, and neural networks (to name a few) to automate processing data to make decisions and perform tasks.

While ML can be a complex topic, Google's Teachable Machine provides a simplified hands-on demonstration of how ML works.

The most powerful form of machine learning being used today, called “deep learning”, builds a complex mathematical structure called a neural network, based on vast quantities of data.

Neural networks are sets of algorithms in ML and AI modeled after the way nerve cells in the human brain and nervous system process information.

Artificial Intelligence vs. Machine Learning vs. Data Mining

To best understand the relationship between AI, ML, and data mining, it’s helpful to think of a set of different sized umbrellas. AI is the largest umbrella. The ML umbrella is a size smaller and fits underneath the AI umbrella. The data mining umbrella is the smallest and fits underneath the ML umbrella.

  • AI is a branch of computer science that aims to program computers to perform tasks in more “intelligent” and “human-like” ways, using reasoning and decision-making techniques modeled after human intelligence.
  • ML is a category of computing within AI focused on programming machines (computers) to learn (gather necessary data or examples) to make data-driven, intelligent decisions in a more automated way.
  • Data mining uses statistics, ML, AI, and immense databases of information to find patterns, provide insights, create classifications, identify problems, and deliver detailed data analytics.

What Machine Learning Can Do (and Already Does)

The capacity for computers to analyze vast amounts of information in fractions of a second makes ML useful in a number of industries where time and accuracy are essential.

  • Medicine: ML technology is being implemented in a range of solutions for the medical field, including helping emergency department physicians with faster diagnosis of patients with unusual symptoms. Physicians can input a list of the patient's symptoms into the program and using ML, the program can scour trillions of terabytes of information from medical literature and the internet to return a list of potential diagnoses and recommended testing or treatment in record time.
  • Education: ML is used to create educational tools that tailor themselves to the learning needs of the student, such as virtual learning assistants and electronic textbooks that are more interactive. These tools use ML to discover which concepts and skills the student understands using short quizzes and practice exercises. The tools then provide short videos, additional examples, and background material to help the student learn the needed skills or concepts.
  • Automotive: ML is also a key component in the emerging field of self-driving cars (also called driver-less cars or autonomous cars). The software that operates self-driving cars uses ML during both real-life road tests and simulations to detect road conditions (such as icy roads) or identify obstacles in the roadway and learn appropriate driving tasks to safely navigate such situations.

You’ve likely already encountered ML many times without realizing it. Some of the more common uses of ML technology include practical speech recognition (Samsung's Bixby, Apple's Siri, and many talk-to-text programs that are now standard on PCs), spam filtering for your email, building news feeds, detecting fraud, personalizing shopping recommendations, and providing more effective web search results.

ML is even involved in your Facebook feed. When you like or click on a friend’s posts frequently, the algorithms and ML behind the scenes “learn” from your actions over time to prioritize certain friends or pages in your Newsfeed.

What Machine Learning Can't Do

However, there are limits to what ML can do. For example, the use of ML technology in different industries requires a significant amount of development and programming by humans to specialize a program or system for the types of tasks required by that industry. For instance, in our medical example above, the ML program used in the emergency department was developed specifically for human medicine. It is not currently possible to take that exact program and directly implement it in a veterinary emergency center.

Such a transition requires extensive specialization and development by human programmers to create a version capable of doing this task for veterinary or animal medicine.

It also requires incredibly vast amounts of data and examples to "learn" the information it needs to make decisions and perform tasks. ML programs are also very literal in the interpretation of data and struggle with symbolism and also some types of relationships within data results, such as cause and effect.

Continued advancements, however, are making ML more of a core technology creating smarter computers every day.

Was this page helpful?