Generative AI (or generative artificial intelligence) refers to a class of artificial intelligence models that create brand new content – text, images, audio, video, code or synthetic data – rather than simply analyzing or classifying existing data. A generative model learns the underlying patterns and structure of its training data, then uses that learned representation to produce original outputs that resemble, but do not copy, the material it learned from.
Generative AI differs from traditional (discriminative) AI in its objective. Discriminative models answer questions such as “is this email spam or not?” by drawing boundaries between categories. Generative models instead learn the distribution of the data itself, which lets them produce a plausible new email, image or paragraph on demand.
The main families of generative models include:
- Large language models (LLMs) – Transformer-based models that generate text and code. Examples include the GPT series, Claude and Gemini.
- Diffusion models – Models that generate images, audio and video by learning to reverse a gradual noising process. Examples include Stable Diffusion, Midjourney and DALL-E.
- Generative adversarial networks (GANs) – Two competing neural networks that produce realistic synthetic data, widely used before diffusion models became dominant.
- Variational autoencoders (VAEs) – Models that compress data into a latent space and generate new samples from it.
Common applications of generative AI include drafting and summarizing documents, writing and debugging software, customer service chatbots, image and video creation, drug discovery, voice synthesis and generating synthetic training data for other machine learning models.
Alongside its benefits, generative AI raises well-documented concerns: hallucination (confidently stated false output), copyright and training-data disputes, deepfakes and misinformation, and the environmental cost of training very large models. These concerns have made responsible deployment and AI governance central topics for organizations adopting the technology.
