Distinguishing Between AI, Machine Learning and Deep Learning

You’re going to be hearing a lot about artificial intelligence (AI), machine learning and deep learning in the near future, judging by this year’s Consumer Electronics Show. Major manufacturers such as Samsung and LG showcased AI-powered smart refrigerator screens that can be controlled by voice-activated digital assistants such as Bixby and Alexa. These tech giants also highlighted other smart home appliances, such as washing machines, that use machine learning to study user laundry habits and optimize settings to factor in variables such as air quality and weather. Meanwhile, Nvidia rolled out a self-driving car that uses deep learning facial recognition to determine whether a driver might be looking in the wrong direction to see oncoming vehicles or whether they look too drowsy to drive.

All these technologies share a generic relationship to AI, but there are also important distinctions between artificial intelligence, machine learning, and deep learning. Here’s a deeper look at how these technologies are related to each other, as well as what makes them different.

Artificial Intelligence: Handling Everyday Tasks Automatically

You can think of artificial intelligence as a general category that includes machine learning and deep learning as sub-categories. AI is a specialty within computer science that uses computers to replicate and automate cognitive tasks performed by humans, such as sensory perception, language processing, mathematical operations and logical deduction.

The Holy Grail of artificial intelligence research is general AI, which would be an artificial brain that could duplicate human intelligence, like Data from Star Trek. But experts aren’t sure this achievement is possible, and most AI research pursues more limited goals. One of the first AI products to reach the consumer mainstream was Apple’s Siri, which uses artificial intelligence to provide voice control as a simpler way for users to interface with their devices.

Next-generation mobile devices include built-in AI capability. For instance, Qualcomm’s new Artificial Intelligence Platform uses a lightning-fast processor chip to run on-device AI applications without needing to depend on remote cloud resources. Other examples of consumer AI applications are self-driving cars, chatbots, and computer-controlled video game characters.

Machine Learning: Making AI Adaptable

Machine learning is an artificial intelligence application that allows computer programs to adapt more fluidly than standard AI. In standard AI, as in traditional computer programming, programs run set operations based on pre-assigned rules that generate predictable output. Taking a more flexible approach, machine learning applies probability in order to generate and test a series of different mathematical models that match the same data, with each model yielding a closer approximation to the data until the best match is reached. By using this approach to “learn” from data, machine learning can generate a less restrictive, more adaptive range of outcomes than standard AI.

Machine learning can be used for applications such as identifying trends, predicting possible outcomes or recommending decisions based on probable outcomes. For instance, Amazon uses machine learning to build recommendation engines that analyze customer browsing and buying trends in order to make recommendations customized to fit individual preferences. Other applications of machine learning include customizing computers to fit user habits, teaching robots and self-driving cars how to navigate and detecting credit card fraud and cyberattack patterns.

Deep Learning: Using Data to Discover Patterns

Deep learning is an AI specialization within machine learning that copies human neural networks in order to efficiently identify patterns in big data. Based on the way the human brain receives and processes sensory input, artificial neural networks receive and process digital input by passing it through a series of layers before producing output. Each layer analyzes the data signal by assigning it a logical or mathematical value that represents how closely the input represents the desired outcome. For instance, a neural network’s goal might be to find an image that matches a target image. To achieve the desired outcome, each network layer performs a transformation on the previous layer’s signal, then sends it on for further evaluation until a final output is reached. The series of layers in this sequence is what inspires the metaphor of “deep” learning.

One of the most important applications of deep learning is pattern matching. Facebook’s DeepFace software, for example, uses a deep analysis of Facebook’s user photo database to tell whether two different pictures represent the same user with over 97 percent accuracy. Deep learning is also used for speech recognition, natural language processing, and computer vision.

Artificial intelligence, machine learning, and deep learning have a close generic relationship, but they also differ in some important ways. AI is the most general category and is characterized by artificially automating human cognitive tasks. Machine learning uses probability analysis to identify data trends, make predictions and recommend decisions. Deep learning copies human neural networks in order to train computers to spot patterns. All three technologies are growing in importance and will increasingly become a part of our everyday lives.

Some of the link on this post may have affiliate links attached. Read the FTC Disclaimer.