You and AI: Exhibition Glossary

Artificial Intelligence – or AI for short – is part of computer science. There are different types of AI, but in general AI technology refers to computers or machines that are able to learn and perform tasks. AI is different to other types of computer programming, where we ‘tell’ the machine exactly how to complete a task through a series of specific instructions. Instead, in AI the machine figures out how best to do something based on learning from examples and experience. AI can often find surprising and unexpected ways to solve a challenge.

Nowadays, AI is an often invisible part of our everyday lives. From predictive text on our smartphones, to shopping recommendations online. It might even be screening your CV when you apply for a job. The fact AI is now so ubiquitous in our lives means it is important to understand how AI shapes our world, for better or worse, and who is shaping the AI.

In computer science, a traditional algorithm is a precise set of rules or instructions that tell a computer how to solve a problem. With a traditional algorithm, to complete the task the computer must follow the steps in the order they’re laid out by the programmer. By contrast, a machine learning algorithm creates the rules or ideal steps itself, by experimenting with the task and learning as it goes.

Algorithmic prejudice describes a computer system or AI that privileges one group and/or discriminates against another group(s). Artificial intelligence is shaped by humans after all, and often the data the machine learns from can reflect existing bias and prejudice in our society. Algorithmic prejudice such as racial profiling or gender discrimination can create unfair and often dangerous outcomes. Algorithmic prejudice is also often referred to as ‘algorithmic bias’ or ‘algorithmic discrimination’.

A dataset – or training data – is a collection of data used to train a machine / AI. Data exists in many forms. It could be a collection of photographs, a huge catalogue of text (books, recipes, blogs, for example), a library of sounds and audio recordings, and so on. AI uses the dataset to learn from.

Deep learning is a subset of Machine Learning, inspired by how the human brain works. Deep Learning relies on algorithms by which a machine/computer can teach itself how to do something based on ‘looking at’ and learning about a huge dataset of examples.

Facial recognition technology is able to recognise human faces and match them against a database. For example, you might use facial recognition to unlock your phone or pass through passport control. Facial recognition is also used in video surveillance.

Machine learning – or ML for short – is a subset of AI. Machine Learning uses algorithms to look at a dataset, find patterns or repeating features, and build a model. This model then allows the computer to make predictions or decisions. For example, looking at thousands of pictures of sheep, to make a model of what a sheep ‘looks like’, to then be able to look at other photos and work out if that photo contains a sheep.

A neural network is a type of AI that is modelled on how the human brain works. As the name suggests, it is a network of neurons or ‘nodes’ that can observe, process and analyse data to learn from it, make decisions and solve problems.

In a GAN – or Generative Adversarial Network – two neural networks compete (as adversaries!) to become better and better at making accurate predictions. A GAN is ‘generative’ as it creates (or generates) new data, which is then assessed for accuracy – does it fool the network? With each attempt the GAN learns how to generate with greater accuracy. GANs are often used to generate images, video and audio

Predictive analytics involves the use of statistics and search for patterns in current and historical data in order to make predictions about future events and identify risks and opportunities.

Taxonomy is a system for organising, classifying and naming things into groups that share similar qualities.

Biometrics are measurements and information about someone’s body, used to identify individuals.

Hashing is the process of converting a string of data with another value or code. Hashing is a common practice in computer science and it is used for data compression, indexing or cryptography.

Cryptography is the science of creating and using special codes to keep information in computer networks. A human readable message can be encoded with the use of an algorithm or mathematical operations into something hidden and unreadable.

Deepfake is a term used to refer to media produced and manipulated by automated means, in which a person in a video or image has been replaced with someone else’s likeness (especially a public figure) in a way that makes the video or image look authentic.

Hyperspectral imaging (HSI) is the capturing and processing of an image exploring the electromagnetic spectrum beyond the red, green and blue (RGB) bands that are detected by standard color cameras. This method can see a broader range of wavelengths extending beyond the visible. HSI technology applications can be found in different fields from healthcare, agriculture, physics to geosciences and surveillance.

A Cubesat is a miniature, low-cost satellite which was originally commonly used in low Earth orbit for applications such as remote sensing or communications. More recently cubesats are being deployed for interplanetary missions.