MIT News Guidelines on Generative AI
Serving the MIT community as well as journalists and the general public, the Institute Office of Communications (IOC) manages platforms including MIT News, the MIT Daily/Weekly and other newsletters, the MIT homepage and associated web pages, the MIT YouTube channel, and Institute-level social media accounts.
Maintaining the trust of our audiences is paramount, which means our content on these public-facing platforms must be accurate, above all, and meet high editorial standards.
With these observations in mind, we have adopted the following principles, effective Dec. 6, 2023, to guide the use of generative AI in content appearing on IOC channels. These guidelines reflect input from communicators both inside and outside the IOC, as well as MIT faculty members:
- Our articles, newsletters, and video scripts will continue to be written by humans. The possibility of inaccuracy in the output of AI tools is well-established. While much of our content is vetted by experts before publication, we strive to minimize the risk of publishing incorrect or biased information on IOC channels. Further, the importance of communicating science is core to MIT’s ethos and a distinguished area of our educational pedagogy.
We acknowledge that there will be circumstances when non-IOC communicators at MIT may use AI to generate text for use on their own platforms around the Institute. However, we feel that human judgment and expertise that goes into writing long-form text provides a better service to IOC audiences and helps protect the trust that the public has placed in MIT as a source of accurate, credible information.
- We will allow the use of generative AI tools in limited circumstances. There are multiple ways in which MIT communicators use generative AI to assist in the early stages of writing or production, including for generating ideas, researching background information, creating outlines, or analyzing data. AI chatbots may also help with copyediting, cutting text, or drafting headlines, image captions, alt-text, meta descriptions, and social media posts.
As these uses generally fall into the categories of preparation and summarization, we will allow the publication of content on IOC channels that was prepared with help from generative AI in these ways. Communicators across MIT are already responsible for fact-checking and vetting their work to ensure its accuracy before submitting it to IOC platforms, and this will continue.
- We will no longer publish images produced by generative AI. In the past, we have accepted such images submitted to accompany MIT News articles. But in order to align with our position on human authorship of written materials, as of Dec. 6, 2023, we will no longer publish AI-generated images. We have compiled a list of resources for free or low-cost stock images that we can provide to colleagues upon request.
We may make exceptions for images generated with AI tools for research purposes, in cases where these images directly illustrate the research being described in an article.
Similar to our position on editorial content, it’s acceptable to use genAI in a preparatory way, such as for brainstorming, planning color palettes, moodboarding, or developing an image concept. AI-generated thumbnail images pointing to content on non-IOC websites are acceptable — for example, in a newsletter or in the automatic preview that appears for a link on a social media post.
While other MIT departments, labs, and centers are welcome to adopt these guidelines for their own communications purposes, they are also free to pursue their own approaches.
These guidelines will change as AI and its use throughout society continue to evolve. In the event of significant changes to this guidance, we will communicate those changes to colleagues around the Institute. For now, we believe the principles outlined above will help us maintain the trust, accuracy, and editorial standards that our audiences have come to expect of us, while allowing for the responsible adoption of new technologies that have the potential to enhance our communications.