Model collapse is the loss of quality that can occur when generative models are repeatedly trained on data that was itself produced by AI. Rare content and edge cases are gradually lost, and outputs become more uniform and more error-prone. The issue is gaining importance because the share of AI-generated text and images on the internet is growing. Carefully curated, human-created data helps to counteract it.
Synthetic data is generated artificially rather than collected in the real world, but mimics the patterns and properties of real data. It helps when genuine data is scarce, expensive or sensitive from a data protection perspective – for example for testing software, training models or checking a chatbot with realistic sample queries. It is important that it reflects reality closely enough and does not allow conclusions about real people.
Training data are the examples from which an AI model learns patterns: large text collections for language models, labelled images for image models. Their quality, diversity and recency largely determine how reliable and fair a model is and how up to date its knowledge will be. For organisations it also matters whether a provider uses the customer data they enter for further training.
Generative AI refers to AI systems that create new content – text, images, audio, video or program code – rather than merely sorting or assessing existing data. They are based on models that learn patterns from large amounts of data and use them to produce plausible new outputs. In business it is used, for example, to draft emails, summaries, product descriptions or image ideas. Results can contain errors and should be checked before use.
‘Garbage in, garbage out’ describes the principle that a system can only produce results as good as the data it receives. This applies doubly to AI: incorrect, outdated or one-sided training or knowledge data leads to wrong or biased answers. A chatbot that draws on an outdated price list will quote outdated prices. Looking after data is therefore an ongoing task in every AI project.
