Generative AI: An Introduction

There are numerous ethical issues and questions related to the development, implementation, and use of generative AI tools and applications. 

Generative AI is built on large language models (LLM) that are "trained" on enormous amounts of data.  These sources may reflect biases and stereotypes. When generative AI uses these resources to train the model, its output may reflect these same biases.  In addition, the Internet serves as a major source of data for LLM training and is overrepresented with English language resources,  which may lead generative AI models to underrepresent knowledge and ideas from non-English language communities and nations. 

While the content generative by AI can be useful, users should be aware of the possibility of bias in the output.

Hallucinations occur when generative AI tools and applications produce incorrect or misleading content. The fabricated content is presented in such a way as to appear authentic and can be difficult to identify the errors.  One common hallucination is when generative AI is prompted for research citations in a given area; the citations may or may not point to actual literature. 

Image and sound-based AI tools are also subject to hallucinations. Generative AI may add pixels in ways that do not accurately reflect the object.  This is why, for example, image-generation tools are notorious for adding fingers to hands! 

Another significant issue over the past two years has been the deliberate misrepresentation of images, audio, and text.