Generative Artificial Intelligence (AI): Home

Generative artificial intelligence (AI) refers to computational systems able to create (generate) new text, images, or other media. By drawing on typically vast datasets used to train them, generative AI models can produce novel outputs in response to human input.

Generative AI is by no means new but has recently gained extraordinary popularity, beginning with the 2022 release of OpenAI's ChatGPT, a conversational bot able to generate text (and more recently images) in response to prompts in plain language (i.e., without programming code input).

The potential uses of generative AI include writing, editing, and summarizing text (including programming code), creating graphics in various styles, and searching for information, all through the medium of conversation—i.e., without requiring specialized skills on the part of the user.

However, the risks of using generative AI are significant, particularly in highly sensitive contexts, such as medicine. It is imperative that before you use these tools in your work, you understand the risks and how to avoid them.

Never enter sensitive information into an AI model

• Even models that claim to protect input data may not adequately do so.
• Entering protected information may constitute a HIPAA violation, whether or not the data are stored on the developers' servers.

Always double-check the outputs yourself

Care should be taken when using the outputs of GPT-4, particularly in contexts where reliability is important.

• Generative AI models are known to "hallucinate" non-existent references or confidently declare falsehoods.
• AI is not a substitute for systematic literature search, which is a necessary component of evidence-based practice.

Treat AI as a tool

• Use it to support and complement your own efforts, not as a replacement for them.
• Declare that you have used AI, including which tools and how they were used, in the Methods or Acknowledgements of your manuscript.

See the Limitations tab for more information.