Skip to Main Content

Artificial Intelligence

A guide to resources for learning basics of artificial intelligence

Artificial Intelligence

Artificial Intelligence, or AI, is a subject that's increasingly in the headlines, and increasingly present in our day-to-day lives. But many people don't understand what AI is or how it works. This guide provides basic education about what AI is, how some commonly-used AI tools work, and the applications and ethics of AI. The goal of this guide is to provide readers with enough information about AI to understand the implications of news and advancements in the field, and to help readers interested in creating their own AI tools to find resources to learn more.

Artificial Intelligence v.s. Machine Learning v.s. Deep Learning

You may have heard the terms "artificial intelligence," "machine learning," and "deep learning" discussed interchangeably, or without any explanation of what the differences are between them. Although these terms are related, they're not interchangeable. Deep learning is a subfield of machine learning, which is a subfield of artificial intelligence.

Three nested circles. The outermost circle is labeled Artificial Intelligence. The middle circle is labeled Machine Learning. The innermost circle is labeled Deep Learning.

Artificial intelligence is any computer program or software that mimics some aspect of human intelligence. Because human intelligence is comprised of many types of tasks, many different kinds of software are considered AI. Some common artificial intelligence tasks include: object recognition; speech and language; game-playing; decision-making; and movement through physical space.

AI can be achieved either through predictive algorithms, which use large datasets to learn patterns, or through hard-coded rules. AI software that uses hard-coded rules is sometimes referred to as "good old-fashioned AI," or GOFAI.

To demonstrate the difference between predictive AI and GOFAI, think about the game of tic-tac-toe. The rules of tic-tac-toe are simple, and so is the best strategy. We could easily create an AI program that plays tic-tac-toe according to a few hard-coded rules:

List of tic-tac-toe rules. 1. If there is a row, column, or diagonal with two of my pieces and a blank space, then play the blank space and win the game. 2. If there is a row, column, or diagonal with two of my opponent's pieces and a blank space, then play the blank space and block my opponent's win. 3. If there are two intersecting rows, columns, or diagonals with one of my pieces and two blanks in each, and the intersecting space is blank, then play the intersecting space and create two paths to a win. 4. If there are two intersecting rows, columns, or diagonals with one of my opponent's pieces and two blanks in each, and the intersecting space is blank, then play a space that gives me two pieces in a row if one exists and force my opponent to block me, or play the intersecting space and stop my opponent from creating two paths to a win. 5. If the center is blank, then play the center. 6. If my opponent is in a corner and the opposite corner is blank, then play the opposite corner. 7. If there is a blank corner, then play the blank corner. 8. If there is a blank side square, then play in the blank side square.

(Crowley and Siegler)

But another way that we could create a tic-tac-toe AI is to write a program that looks at thousands of tic-tac-toe games, and finds the patterns of moves that are the likeliest to win in that data set. At each move, the program would consult the patterns it had learned, and make the move that, in the data, leads to the highest probability of winning. That would be an example of predictive AI.

A tree chart of possible moves for a tic-tac-toe game. The root node has two Os and two Xs filled in. The three leaf nodes each have a possible placement for the next O. Each of the leaf nodes is labeled with a percentage: 33%, 75%, and 33%.

 

Machine learning is the subset of artificial intelligence that is concerned with predictive AI. Its name comes from the fact that in machine learning, computers learn what to do, rather than being explicitly told. Machine learning can be used for many tasks that would be difficult to program using GOFAIā€”for instance, object recognition. Most current advancements in AI, and most AI that you read about in the news, is machine learning.

You may have also heard some forms of AI called deep learning. Deep learning is a subset of machine learning that uses an algorithm called a neural net. The learning is called "deep" learning because a neural net algorithm involves many layers of processes. Deep learning algorithms are usually very complex.