What is generative AI?
Generative AI is a type of artificial intelligence that creates new content, such as text, images, audio, video or software code, in response to a prompt. Generative AI learns patterns from very large amounts of existing material and uses them to produce output that resembles what a person might write, draw or record.
Why it matters for your business
A large share of office work is producing and reworking content: emails, proposals, reports, summaries, meeting notes, product descriptions, code. Generative AI can produce a usable first version of most of these quickly. The gain is time on the first draft and on routine reading, such as summarising a long contract or a lengthy email thread. It also needs no technical skill to use, because the instruction, called a prompt, is written in ordinary language. That is why staff often start using it before the business has agreed any rules for it.
How it works
A model is trained on a very large body of text, images or other material and learns the patterns in it. General-purpose models of this kind are called foundation models; those built for language are large language models (LLMs). The model can then be adapted to a particular task or given a company's own documents to work from. When it receives a prompt, it generates a response piece by piece, each time predicting what is most likely to come next.
The model does not look facts up the way a database does. It produces what is plausible given its training and the material in front of it.
Example
Picture an 80-person engineering consultancy that answers several tenders a month. A bid manager gives a generative AI tool the tender document and the firm's past submissions, and asks for a first draft of the methodology section. The bid manager then corrects the technical details, checks every claim against the project records and rewrites the parts that sound generic. The time saved is on the blank page, not on the review.
Costs, limits and when not to use it
It can be confidently wrong. Generative AI can produce statements that read well and are false, known as hallucinations. Anything that matters needs checking by someone who knows the subject. It can repeat bias present in the material it learned from. Confidentiality and ownership. What staff paste into a public tool may leave the company, and output can resemble copyrighted material. It is the wrong tool for exact work. For calculations, forecasts from company data, or decisions that must be explained and repeated identically, conventional software or predictive models fit better.
Generative AI vs predictive AI
Predictive AI, the longer-established form, analyses data to classify or forecast: which customers are likely to leave, which transaction looks fraudulent. Generative AI creates new content. The two solve different problems, and a forecasting question is not answered by a content tool.