Deep learning is a concept that essentially represents that giant leap forward in the ability of machines to think for themselves. Through deep learning, a computer can produce output without an explicitly programmed algorithm. This still falls within the realm of artificial intelligence which, broadly speaking, encompasses the range of abilities that allows machines to do what humans do. Although it may have begun with something harmless like programming machines to beat humans at chess, deep learning allows machines to act in the absence of programming. For many people, that idea may be a little scary.
What Are AI & Deep Learning?
Artificial intelligence is a buzzword that we are bombarded with across numerous media. We rarely hear the term itself explored or explained, however. Here, we’d like to cover what exactly the terms artificial intelligence, deep learning, and a related term called machine learning mean. These terms are interrelated because they all signify abilities that demonstrate progress in the capability of machines to act independently or “think,” like us humans.
Let’s start with our examination of artificial intelligence, or AI, and proceed from there. This is a term that arose in the 1950s, a time when computers had been used for about a decade to break codes and work complex algorithms. By then, computers were programmed to complete tasks, but the idea also arose that perhaps computers could be programmed to think in at least some of the ways that humans think. This was how the idea of AI came about.
So what is AI, really? Because AI encompasses a range of abilities, just as human beings and the human mind encompass a range of abilities, definitions are always general. AI can be defined as an intelligence or ability that results in an optimal solution to a problem. To put it another way: Given a problem, the machine (computer) is able to come up with a solution in just the same way that a human being might. This can involve anything from playing a board game and successfully beating an opponent to solving a mathematical problem or navigating a maze.
As AI has gone from something speculative to something in practical, everyday use, the fact that machines are able to solve some problems more quickly and reliably than human beings has become apparent. Computers can now routinely defeat masters at their various games and perform calculations far faster than the smartest, most experienced scientists. In order to understand how they are able to do this, we need to understand the concepts of machine learning and deep learning, which lie within the larger circle of artificial intelligence.
Machine Learning vs. Deep Learning
Although it can be confusing to sort out how these terms are different, understanding them a bit better can help to shed light on just how frightening (or exciting, depending on your sense of adventure) machine capabilities may be in your view. How will we define what deep learning and machine learning are? We will start out by keeping the definitions pretty simple:
- Machine learning is the ability of a machine to complete a task without being explicitly programmed.
- Deep learning is a machine’s ability to process data outside of task-specific algorithms, sometimes without human monitoring or intervention.
The distinctions between machine learning and deep learning can be confusing. Going a bit further, deep learning can be thought of as lying within the broader spectrum of machine learning. It helps to understand that early work in these areas was inspired by biological neural networks. The nervous system is able to collect sensory data of various kinds, process it, and produce an output. Consequently, if a machine is to be capable of behaving in ways similar to humans, they also need to be capable of perceiving and processing information independently.
The key to understanding the leaps forward in artificial intelligence technology is to first grasp the basic concepts of computer programming. A machine can be programmed to complete a specific task using an algorithm. It can also be programmed to collect data and make a decision or choose a solution based on that data. This latter case reflects the direction of current AI advancements, and it represents the deep learning side of AI. Today, some machines are able to collect information without being programmed and reach an optimal solution based on the data they have collected.
What Are Their Differences?
The issue of the meaningful differences between AI and deep learning is an interesting one because AI is an umbrella term that encompasses other related terms. We can say that, over time, the true capabilities of machines have become better understood and that artificial intelligence has expanded to encompass capabilities that we understand better now than we did when computing was developed more than 60 years ago.
The term AI came out of the Dartmouth Conferences of 1956. Minds were emblazoned with the idea that the leap forward expected from computers would be that these machines could eventually behave much in the way that the human mind behaves. Eventually, the terms General AI and Narrow AI came to the fore, and a brief discussion of these areas may help your understanding of deep learning become even clearer.
Let’s consider what General Ai and Narrow AI are. General AI applies to machines that are capable of exercising a broad range of the abilities that we associate with human beings. In contrast, Narrow AI describes machines that are able to intelligently complete specific tasks like facial recognition.
Narrow AI is already in widespread use today, as many computer programs are able to recognize similarities between things, as when they suggesting similar pictures on a search engine, for example. General AI still is the stuff of legend, or at least the stuff of movies, which may feature villains that have been manufactured by human beings and motivated by artificial intelligence serving as the antagonists in the story. Much of the wonder and worry about AI has revolved around these images of the great capabilities of AI. Frankly, we can say today that those projections are not too far off the mark.
So where does deep learning fit into all this? We can think of deep learning as the link that makes General AI possible, or even as the link between Narrow AI and General AI. Of course, we see that a machine or computer program can complete a task like facial recognition, but how do we go from that to creating the Terminator? The key is patterning machine abilities on things that exist in nature, like neural networks. Developers study the ways that other intricate systems detect and process information, and then they apply that knowledge to computing.
Once machines have accomplished the leap from always needing a programmed algorithm to being able to collect different types of information and process them using different networks, ultimately reaching an optimal solution, they possess incredible new capabilities. Such a machine, developed to its full potential, would possess deep learning and approach the full potential of artificial intelligence.
Pros, Cons, & Potential Risks
Now we can weigh some pros and cons of artificial intelligence, with deep learning positioned as the natural endpoint of artificial intelligence development. This concept that was addressed, but not well understood, back at the Dartmouth Conferences of 1956.
The theoretical possibility that artificially intelligent machines may, individually or collectively, reach the conclusion that humans are deleterious to nature and should be removed remains far-fetched. However, this is the concern that has been the main plot device of many futuristic movies, and it is not impossible, given that AI, by its designed-in nature, involves machines reaching “optimal” conclusions on their own.
The question of whether artificial intelligence is something that is truly beneficial to humans is an interesting one. Indeed, it begs the philosophical question of whether the interests of humans may run contrary to nature and whether artificially intelligent machines may side with nature over us. Whatever your take on the debate is, the reality is that the deep learning side of AI is something that is no longer in the distant future. It is already here, and work being done in deep learning has very important implications for what reality may look like in the future.