What is Generative AI (genAI)?
Citing IBM Research’s blog, generative AI refers to deep-learning models that can generate high-quality text, images, and other content based on the data they were trained on.
MIT News has it as a machine-learning model that is trained to create new data, rather than making a prediction about a specific dataset. A generative AI system is one that learns to generate more objects that look like the data it was trained on.
Finally, Cambridge dictionary describes it as a particular artificial intelligence … that is able to produce text, images, etc.
These definition, and the very name “generative AI,” all point to the quality of generating or producing something. Most often, the stuff produced is content of some kind, more-or-less directly presented to consumers (text, images, video, audio), but can also be program code, mathematical formulas, or science simulations. In some areas there is a fine line between content generation and transformation, for example in translation services. As good translations involve translating idioms and word play, there is always some new content generation involved.
For the purpose of Certified Human, the term “Generative AI” applies to any use of machine learning / artificial intelligence algorithms to generate data, including translations, code generation, scientific research, etc. It specifically does not apply to other applications of machine learning that analyze or classify data - as long as the output of the algorithm is not new data used as baseline for further applications or processes, or for presentation to consumers.
What about edge cases?
Generative AI is a new enough topic that rules and definitions need to be developed on the fly, and there will always be a degree of arbitraryness around them. Here are a couple of examples and Certified Human’s stance about them:
Cases that fall under the “AI generated” definition
- Images “photoshopped” with an AI Fill function - YES, falls under “AI generated images.”
- Images modified by camera software to supstitute or generate details on the photos - YES, it’s genAI-generated, even if it’s done by the camera software.
Cases that do NOT fall under the “AI generated” definition
- Images “photoshopped” with a function that adjusts features like white balance, hue, colorisation / pallette change - as long as the original image is drawn or photographed by a person.
- Advanced multi-lingual translation - as long as the role of the genAI is constrained to translation, not generation of text / phrases / ideas not present in the original text.