Generative AI

In short

Generative AI is AI that creates new content, such as text, images, audio, video or code, in response to a request. It learns patterns from huge amounts of existing material and produces fresh output that resembles it.

Also called: GenAI, Gen AI, GAI
Topic: The basics
Level: Beginner
First described: An early modern method came from Ian Goodfellow and colleagues, who proposed generative adversarial networks in June 2014⁴.

A definition used by the US standards body NIST describes generative AI as "the class of AI models that emulate the structure and characteristics of input data in order to generate derived synthetic content", including images, videos, audio and text¹.

Much older AI sorts or predicts: is this email spam, will this loan be repaid. Generative models produce something new. ChatGPT, launched by OpenAI on 30 November 2022, generates text in conversation². Image generators such as Stability AI's Stable Diffusion produce pictures from typed prompts³.

One early modern method was the generative adversarial network, proposed by Ian Goodfellow and colleagues in 2014. It trains a generator alongside a second model that tries to tell its output from real data⁴. By 2020, OpenAI's GPT-3 could write news articles that human evaluators had difficulty telling apart from human-written ones⁵.

Generative AI produces plausible output, not checked facts. NIST lists confabulation, "the production of confidently stated but erroneous or false content", among the main risks¹. It also warns of harmful bias, such as image tools under-representing women in pictures of CEOs, doctors and judges¹.

Use is spreading fast. Ofcom figures quoted by the Government Digital Service suggest 54% of UK adults say they use AI tools, up from 31% the year before⁶.

Example

Asking a chatbot to draft a polite complaint to your energy supplier, or asking an image tool for "a fox reading a newspaper on a London bus", are both uses of generative AI.

Why it matters

It is the kind of AI most people now use directly, and it raises new questions about accuracy, copyright and fake content.

Questions people ask

Is ChatGPT generative AI?

Yes. ChatGPT generates text in conversation. At launch in November 2022, OpenAI said it could answer follow-up questions, admit mistakes and challenge incorrect premises².

What is the difference between AI and generative AI?

Generative AI is the part of AI that creates new content. NIST's definition covers models that generate synthetic images, video, audio and text¹. Other AI, such as a fraud filter, classifies or predicts instead.

Can generative AI be wrong?

Yes. NIST lists confabulation, confidently stated but false content, as a key risk of generative AI¹. Check anything important against a reliable source.

Related terms

Tools that use it

Sources

Back to the glossary