Both artificial intelligence and machine learning are still emerging technologies. Even highly technical people may not always understand the difference between AI, machine learning and other related tech. Sometimes, you may see and hear machine learning and AI used as synonyms, but they aren’t the same. Before discussing real-world uses of machine learning applications, it might help to briefly review a few basics.
Machine Learning vs. AI
Technically, machine learning is something that intelligent systems may do, so it's more accurately considered a subset of AI.
- AI machine learning refers to software that can alter its reactions after gathering more experiences and data.
- AI simply refers to computer systems that can recognize patterns or make decisions in ways normally associated with human intelligence.
In other words, AI describes smart machines, but machine learning algorithms can make the machines smarter — without needing a programmer to step in to modify any code. As a simple example, you might expose a machine learning algorithm to a labeled set of images or descriptions of cats and dogs. Later, the algorithm should learn enough to sort images or text into cats and dogs on its own.
As the machine finds that it mistakenly categorized a large cat as a dog, it will learn from that to avoid the error in the future. Feeding the algorithm data and then correcting mistakes is referred to as training it. A less-trivial example might include systems that scan thousands of images of moles to learn to detect which are probably benign and which ones might indicate skin cancer.
Real-World Applications of Machine Learning Algorithms
Almost everybody remembers getting frustrated with automated phone systems that didn't seem either intelligent or capable of learning. You may have to be a certain age to realize what a breakthrough even rudimentary voice recognition was a few decades ago. Because of improvements in all sorts of technical advances, you probably interact with machine learning algorithms without even knowing it more often than you think today.
Certainly, phones provide plenty of examples of speech recognition. Besides automated phone systems, you may use voice-to-speech to compose text messages or even ask Siri for cooking temperatures. Some voice-to-text programs can learn from the errors that you correct in documents to improve in the future, so you "train" them to understand your accent and vocabulary better.
Today, image recognition has advanced to the point where you may not realize a smart and trained machine powers an automated teller machine or even a banking app on your phone. According to Apple, its new Face ID security system relies on an interface between a camera with depth perception and machine learning algorithms.
A few years ago, Internet users would mostly see advertisements based on the content of the site they visited. Retargeting systems made it possible to also base ads on the user's recent online visits. Maybe the first thing that typical consumers noticed were advertisements from companies that seemed to follow them around the web.
While some folks found these creepy, over twice as many consumers say they'd prefer to see targeted ads than general ones. As marketers began to enjoy more success with these first semi-intelligent ads, they also invested in machine-learning algorithms that could covert better with different groups of consumers, based on consumer behavior. Today, these systems don't limit themselves to simply retargeting online ads, but may schedule emails, text messages and even phone calls.
Today's chatbots rely on advancements in natural language processing (NLP). This could be one area of machine learning that holds the most promise for businesses. After all, one good chatbot on a commodity server could replace hundreds of live phone operators for first-level phone support and customer service.
By 2015, Google trained a machine algorithm that could handle tech support fairly convincingly. The software could also answer other factual questions and even attempt to participate in ethical and philosophical discussions.
Today, Facebook's public chat algorithms can rarely fool people into believing they are messaging a human, but that could be intentional. As the social networking company worked to improve the conversational and even negotiating skills of its algorithms last year, it accidentally created a problem. When it dialed up the self-interest aspect of chatbots it wanted to train to negotiate, conversational skills improved dramatically, but during the process, the algorithm learned to lie. Now chatbot engineers have to decide just how human they want to make their messaging software.
IBM's supercomputer Watson has already added artificial intelligence and the ability to learn the healthcare field for half a decade. Some examples include genetic research, scheduling social programs, drug discovery and, of course, medical diagnosis, treatment suggestions and predicting outcomes.
As an example, oncologists might rely on Watson to spot important information in a patient's history, find treatment options from reams of medical publications, and predict the outcomes of various treatments. Some medical providers also use learning algorithms to help engage and reach out to patients via mobile and web apps.
How Machine Learning Will Drive Progress
Other examples of current and upcoming uses for AI machine learning include self-driving cars, "also like" lists, fraud detection, and software that judges prevailing attitudes and predicts trends by sifting through millions of social posts. Still, machine intelligence has limits because algorithms don't understand context, humor or sarcasm well at all. Specialists still speak of training and not teaching their machine learning algorithms.
For instance, medical AI machine learning might help a doctor but cannot replace one. Nevertheless, judging by the common uses of this still-emerging technology, machine learning can help a lot. In this era of big data, smart machines capable of learning can consume and analyze information much faster than people. They also have photographic memories and never get bored or distracted. By using machine learning for what it does best, all kinds of organizations can help people focus on what they can still do better.