Artificial intelligence

Artificial intelligence (AI) is any system that uses a computer or software to replicate the idea of human intelligence. AI involves the development of computer programs and algorithms that can learn from and make decisions based on data, without the need for human intervention. Some of the tasks that AI performs include speech recognition, language translation, chatbots, text generation, and image and video analysis.

Types of artificial intelligence

There are two major types of AI: expert systems and machine learning.

Expert systems

An expert system uses AI to mimic the thinking of a human who has expertise in a particular area. At its core, an expert system is a very complicated flowchart made by a programmer. The computer follows the flowchart using if-then rules to make decisions. Those who operate expert systems can make them more proficient by patching (updating) the system.

Almost all video game AI uses the expert system method. The diagram below shows the kind of if-then flowchart a Minecraft creeper might use when approaching a player.

A real-life example of an expert system is banking software in which a group of banking experts refine and distil their knowledge of fraud detection. The software has a high level of expertise and can detect fraud issues and respond to them more quickly than a human.

Machine learning

Machine learning describes any process where the computer figures out how to perform a task by itself. Simply explained, a machine (computer) has access to information and uses algorithms to find patterns, generate understanding and make decisions. This process is similar to learning to play Minecraft. A new player discovers that creepers explode when they get too close. After a few explosions, the player gains information and understanding about the creeper and makes the decision to avoid it while playing the game.

The way the machine is taught to perform a task is dependent on the problem the machine learning software is asked to solve. The most common examples are classification and regression. Classification is where the user inputs a particular type of information and receives a discrete output – for example, you’ve taken a photo of a plant and you want to know what it’s named or what species it is. On the other hand, regression produces a numerical output – for example, content recommendation algorithms that assign a value to how likely you are going to enjoy a video based on what the algorithm knows about you.

Artificial neural networks and deep learning

An artificial neural network is a mathematical system that is based on how we once thought the brain worked. Neural networks take in data, train themselves to recognise patterns within the data and then predict the outcome for similar data. Given it has enough training data, it is mathematically provable that a neural network can solve any classification or regression problem.

Deep learning involves the use of several neural networks. Deep learning algorithms have many layers of neurons and nodes – dozens or even hundreds of them. The many layers are referred to as the depth, which is how deep learning gets its name. The image below is a simplified model of an artificial neural network. The green dots represent layers of nodes.

To put this all together in simple terms, deep learning, which uses neural networks, is a subset of machine learning, which is a subset of artificial intelligence.

Generative machine learning

Another type of machine learning is generative – where the aim is to generate a piece of data, like an image, from scratch. The primary method for creating an image is generative adversarial networks.

The process works by having two separate neural networks. One is trained to generate images while the other is trained to discriminate or tell the difference between generated images and training images. During training, the generative AI tries to fool the discriminator AI. The discriminator AI tries to tell the difference between a generated image and a training image. Both AIs train off each other until a realistic image is achieved. This type of training takes a lot of computer time given that millions of images are generated and tested. The end user has a different experience as they are only interested in generating one image, and this results in a much faster experience.

Generative learning is also used to create human-like text. Generative text models are trained using vast amounts of example text. Text generation software usually works by taking a piece of text and asking the question, “What is the next word that would be most appropriate?” This is used for predictive text with search engines and modern chatbots. Another example of generative learning is the generative pre-trained transformer (GPT) used by ChatGPT.

AI in our lives

Artificial intelligence has been decades in the making. It is positively transforming many aspects of our lives from healthcare to using maps to navigate foreign streets and translating languages to recommending movies we might like to watch. AI is also disruptive – educators are grappling with how to use platforms that generate realistic text. There are ethical concerns about bias within the training data and the fear that some jobs might disappear.

One thing we do know. AI is here to stay. In the words of Colin Angle, CEO of iRobot, “It’s going to be interesting to see how society deals with artificial intelligence, but it will definitely be cool.”