For a long time, AI was primarily geared toward classifying data: Is this email spam, does this image show a defect, how will sales develop next month? Generative AI goes one step further and creates content itself. A short instruction in natural language is enough to obtain a text draft, a graphic, or a piece of code.
This changes knowledge work in particular, that is, activities around writing, researching, designing, and analyzing.
How does generative AI work?
Generative models are trained on very large amounts of sample data and learn their statistical patterns in the process. When used, they create new content that fits these patterns and the respective input. Different methods are used depending on the type of content:
- Text and code: Large language models generate text piece by piece, each time as a likely continuation of what came before.
- Images: Diffusion models gradually shape random noise into an image that matches the description.
- Audio and speech: Specialized models generate spoken language or music.
Many current systems combine several of these capabilities and therefore count as multimodal AI.
The results are newly generated but can, in individual cases, closely resemble training content. In addition, they are based on probabilities and are therefore not automatically correct.
Typical examples
- First drafts of texts, quotes, product descriptions, or presentations
- Summaries of meetings, reports, or email threads
- Support with programming and with documenting code
- Image ideas and illustrations for internal documents
- Suggested replies in customer service that employees check and adjust
What companies should keep in mind
- Check content: Generated texts can contain errors or fabricated information, see hallucination.
- Protect confidential data: Define which tools are approved and which information may be entered into them.
- Clarify rights: For published content, observe copyright and trademark rights as well as the services' terms of use.
- Transparency: The EU AI Act provides for labeling and transparency obligations for certain AI-generated content.
- Build competence: Employees should know how the tools work and where their limits lie.
How it differs
Analytical AI classifies existing data, for example through classification or forecasting. Generative AI creates new content. In practice, the two are combined, for example when a customer inquiry is first automatically assigned to a category and then a draft reply is created.
View AI consulting