What Is Artificial Intelligence (AI)? | Google Cloud (2024)

Get started for freeStay informed

While the specifics vary across different AI techniques, the core principle revolves around data. AI systems learn and improve through exposure to vast amounts of data, identifying patterns and relationships that humans may miss.

This learning process often involves algorithms, which are sets of rules or instructions that guide the AI's analysis and decision-making. In machine learning, a popular subset of AI, algorithms are trained on labeled or unlabeled data to make predictions or categorize information.

Deep learning, a further specialization, utilizes artificial neural networks with multiple layers to process information, mimicking the structure and function of the human brain. Through continuous learning and adaptation, AI systems become increasingly adept at performing specific tasks, from recognizing images to translating languages and beyond.

Want to learn how to get started with AI? Take the free beginner's introduction to generative AI.

Artificial intelligence can be organized in several ways, depending on stages of development or actions being performed.

For instance, four stages of AI development are commonly recognized.

1. Reactive machines: Limited AI that only reacts to different kinds of stimuli based on preprogrammed rules. Does not use memory and thus cannot learn with new data. IBM’s Deep Blue that beat chess champion Garry Kasparov in 1997 was an example of a reactive machine.
2. Limited memory: Most modern AI is considered to be limited memory. It can use memory to improve over time by being trained with new data, typically through an artificial neural network or other training model. Deep learning, a subset of machine learning, is considered limited memory artificial intelligence.
3. Theory of mind: Theory of mind AI does not currently exist, but research is ongoing into its possibilities. It describes AI that can emulate the human mind and has decision-making capabilities equal to that of a human, including recognizing and remembering emotions and reacting in social situations as a human would.
4. Self aware: A step above theory of mind AI, self-aware AI describes a mythical machine that is aware of its own existence and has the intellectual and emotional capabilities of a human. Like theory of mind AI, self-aware AI does not currently exist.

A more useful way of broadly categorizing types of artificial intelligence is by what the machine can do. All of what we currently call artificial intelligence is considered artificial “narrow” intelligence, in that it can perform only narrow sets of actions based on its programming and training. For instance, an AI algorithm that is used for object classification won’t be able to perform natural language processing. Google Search is a form of narrow AI, as is predictive analytics, or virtual assistants.

Artificial general intelligence (AGI) would be the ability for a machine to “sense, think, and act” just like a human. AGI does not currently exist. The next level would be artificial superintelligence (ASI), in which the machine would be able to function in all ways superior to a human.

When businesses talk about AI, they often talk about “training data.” But what does that mean? Remember that limited-memory artificial intelligence is AI that improves over time by being trained with new data. Machine learning is a subset of artificial intelligence that uses algorithms to train data to obtain results.

In broad strokes, three kinds of learnings models are often used in machine learning:

Supervised learning is a machine learning model that maps a specific input to an output using labeled training data (structured data). In simple terms, to train the algorithm to recognize pictures of cats, feed it pictures labeled as cats.

Unsupervised learning is a machine learning model that learns patterns based on unlabeled data (unstructured data). Unlike supervised learning, the end result is not known ahead of time. Rather, the algorithm learns from the data, categorizing it into groups based on attributes. For instance, unsupervised learning is good at pattern matching and descriptive modeling.

In addition to supervised and unsupervised learning, a mixed approach called semi-supervised learning is often employed, where only some of the data is labeled. In semi-supervised learning, an end result is known, but the algorithm must figure out how to organize and structure the data to achieve the desired results.

Reinforcement learning is a machine learning model that can be broadly described as “learn by doing.” An “agent” learns to perform a defined task by trial and error (a feedback loop) until its performance is within a desirable range. The agent receives positive reinforcement when it performs the task well and negative reinforcement when it performs poorly. An example of reinforcement learning would be teaching a robotic hand to pick up a ball.

A common type of training model in AI is an artificial neural network, a model loosely based on the human brain.

A neural network is a system of artificial neurons—sometimes called perceptrons—that are computational nodes used to classify and analyze data. The data is fed into the first layer of a neural network, with each perceptron making a decision, then passing that information onto multiple nodes in the next layer. Training models with more than three layers are referred to as “deep neural networks” or “deep learning.” Some modern neural networks have hundreds or thousands of layers. The output of the final perceptrons accomplish the task set to the neural network, such as classify an object or find patterns in data.

Some of the most common types of artificial neural networks you may encounter include:

Feedforward neural networks (FF) are one of the oldest forms of neural networks, with data flowing one way through layers of artificial neurons until the output is achieved. In modern days, most feedforward neural networks are considered “deep feedforward” with several layers (and more than one “hidden” layer). Feedforward neural networks are typically paired with an error-correction algorithm called “backpropagation” that, in simple terms, starts with the result of the neural network and works back through to the beginning, finding errors to improve the accuracy of the neural network. Many simple but powerful neural networks are deep feedforward.

Recurrent neural networks (RNN) differ from feedforward neural networks in that they typically use time series data or data that involves sequences. Unlike feedforward neural networks, which use weights in each node of the network, recurrent neural networks have “memory” of what happened in the previous layer as contingent to the output of the current layer. For instance, when performing natural language processing, RNNs can “keep in mind” other words used in a sentence. RNNs are often used for speech recognition, translation, and to caption images.

Long/short term memory (LSTM) is an advanced form of RNN that can use memory to “remember” what happened in previous layers. The difference between RNNs and LSTM is that LSTM can remember what happened several layers ago, through the use of “memory cells.” LSTM is often used in speech recognition and making predictions.

Convolutional neural networks (CNN) include some of the most common neural networks in modern artificial intelligence. Most often used in image recognition, CNNs use several distinct layers (a convolutional layer, then a pooling layer) that filter different parts of an image before putting it back together (in the fully connected layer). The earlier convolutional layers may look for simple features of an image, such as colors and edges, before looking for more complex features in additional layers.

Generative adversarial networks (GAN) involve two neural networks competing against each other in a game that ultimately improves the accuracy of the output. One network (the generator) creates examples that the other network (the discriminator) attempts to prove true or false. GANs have been used to create realistic images and even make art.