Introduction to Generative AI

Generative AI is AI that creates new content, such as text, images, music, code or video, that looks as if a person made it. It learns patterns from huge amounts of existing content, then uses those patterns to make something new. It has grown fast in the last few years.

Conventional AI analyses, processes and classifies existing data: is this email spam, will it rain, is this a cat? Generative AI creates new content. Both learn from data, but they are built for different goals.

Generative AI can produce text, images, audio, video and code. Under the hood, different kinds of models do the work: GANs, VAEs, RNNs and autoencoders, and today mostly transformers.

A text generator learns from a huge amount of text which words tend to follow which. It then writes one word at a time: it looks at what it has written so far, works out the probability of every possible next word, and picks one. It learns by itself from text that nobody labelled, finding patterns without a teacher.

A GAN has two neural networks. The generator is like a forger: it makes fake data. The discriminator is like a detective: it compares fakes with real data and decides which is which. The detective's feedback tells the forger how to improve. After many rounds the fakes look real.

Generative AI can boost creativity, save time, personalise content and make tools accessible to more people. But it can repeat bias from its data, produce unpredictable or wrong results (sometimes called hallucinations) and needs a lot of computing power.

Generative AI raises new ethical questions: Who owns AI-made art? What if fake photos or deepfake videos spread misinformation? Could it reveal private information or repeat bias? As AI content becomes harder to tell from human work, we must think for ourselves and use it honestly.