One of the main themes I’ve seen in 2019 thus far is the advancement of Artificial Intelligence (AI) and its growing importance in our social, economic and political discussions. At one point, these concepts were fictional, only to be read about in science fiction novels and seen on television. Today, AI is much more than a subject reserved for nerds to argue over. It’s a powerful tool, and if used correctly, can help generate real business value by freeing-up resources. Having a better understanding of AI opens doors to gaining unprecedented insight into business markets, and the world around us by using data.
Let’s start by building an understanding of the three most common terms in AI: Artificial Intelligence, Machine Learning, and Deep Learning. These terms will increase in value with time, and therefore become more important to understand why they are an increasingly important part of business conversations.
The simplest way to differentiate deep learning, machine learning and artificial intelligence is to look at them as a hierarchy. As shown in Figure 1, Machine Learning is a way to achieve AI, while Deep Learning is a system of methods that are applied to Machine Learning. To break this down, let’s start with the broadest of the three: AI.
Figure 1. Machine Learning is a way to achieve AI and Deep Learning is a system of methods that falls under Machine Learning.
Artificial Intelligence (AI) is human intelligence exhibited by machines. “General AI” is the kind we see represented in the characters of some of our favorite sci-fi movies (think C-3PO from “Star Wars” or The Terminator from the movie of the same name). In these movies, the machines think like humans and seem to move through their experiences using the same sensory faculties that we do (sight, smell, etc.) They also experience emotion and can autonomously make decisions based on reason and information they’ve gathered using their senses. This kind of AI is currently nonexistent outside of the imagination.
We do however have “Narrow AI” which is technology that performs very specific tasks as well as, and sometimes even better than, humans. Google’s Translate and Amazon’s Alexa are everyday examples of narrow AI, both of which can perform their designated tasks with skills rivaling or exceeding the average human’s. This kind of intelligence is made possible by machine learning.
Machine learning is the scientific study of algorithms used by computer systems to perform tasks, without explicit instruction. This is the key differentiator from other forms of AI. Machine learning algorithms allow us to iteratively improve on specific tasks without the need for manually specified instructions or intervention. In practice, these algorithms yield continuous, progressive improvements to the assigned tasks efficiently and independently.
There are three major categories of machine learning:
Supervised machine learning is the task of the machine learning a function, without human guidance, that maps an input to an output based on input-output pairs. Doing so requires training data, which includes a set of training examples. The result is an algorithm which analyzes the training data to yield an inferred function that can be used as a prototype for new examples.
Unsupervised machine learning is a self-organized style of learning where the machine finds patterns in your data, without previously existing labels. This style allows a given algorithm to act on that data without manual human input. Using this kind of learning uncovers underlying structures and patterns in sets of data. Unsupervised learning is more complex and less advanced than supervised machine learning but can still be a valuable tool when there is little to no data on desired outcomes.
Reinforcement learning (RL) is used to create software agents, which take actions in an environment based on something, such as an input or response from the environment. This method is useful when needing to look at data to answer an internal question like how to maximize performance based on the notion of cumulative reward. Cumulative reward is defined by a person and is a reward which may constitute positive reward for good actions or negative for bad actions. RL does not require labeled input and output pairs. Instead, it focuses on finding a balance between the exploration of uncharted territory and the exploitation of existing knowledge. For example, RL was used by Alphabet’s Deepmind to generate programs that could play the game of GO, an abstract, strategy board game, and beat some of the best players in the world.
Deep learning is a class of machine learning algorithms in which learning can have supervised learning, unsupervised learning, or some combination of the two. By using multiple layers, deep learning is able to progressively extract higher-level features from your raw input. It is also a part of the broader family of machine learning methods, which are based on artificial neural networks (ANN). ANNs, as the name suggests, was inspired by information processing in biological systems. However, unlike biological brains where any neuron can interact with any other neuron in the nearby vicinity, ANNs have discrete connections, layers, and directions of data propagation. In fact, image recognition algorithms, based on deep learning, can outperform humans in things like identifying tumors in MRI scans or classifying cat images.
Artificial Intelligence has great potential to change the socioeconomic and political landscape of how humanity understands and operates in the world. With the ongoing acceleration and advancement in research, new types of machine learning, including deep learning, are advancing progress at a rapid speed. As we continue to build a society where machines learn, create efficiencies and facilitate discovery in every existing industry, we must continue to stay ahead of the curve by familiarizing ourselves with the tools available to us, and to take advantage of them. Or else prepare yourself to be left in the dust.