Pew Research Center outlines internal AI guidelines: surveys stay human-answered, disclosure rules for production use, and careful experimentation while keeping research people-centered.
At Pew Research Center, we’re approaching artificial intelligence from many perspectives.
As public-facing social scientists committed to innovation, we’re exploring what AI might add to our toolkit. As public opinion researchers, we’re studying and explaining how people are reacting to the advance of AI technologies. And as information providers whose fundamental values include accuracy and methodological rigor, we’re moving with great deliberation to ensure the quality of our work remains high.
In all of these spaces, our primary commitment is that our work remains people-centered:
Real people, not machines, answer our surveys. We don’t use AI to create or model synthetic public opinion. Our survey results are based on the views that real people report to us. For U.S. polls, for example, we rely on a multimode, probability-based survey panel made up of roughly 10,000 adults who are selected at random from across the entire country.
Humans decide what we study. Choosing research topics and deciding what to ask the public are among the most consequential things we do. Our researchers make those decisions.
Humans, not AI, write and review our research reports. Humans determine what’s analytically important and interesting.
Accuracy and rigor remain paramount. Humans oversee every aspect of our work, and humans hold the ultimate responsibility for its quality and accuracy.
We protect respondents and their data. We never expose our survey respondents’ personal identifying information to any open, public AI tool.
The photographs that accompany our work are of, and by, real people. They’re selected by human editors. We don’t use AI to generate or modify photos. The same goes for our illustrations.
How we currently use AI
In the production of our website. Our engineers use AI code assistants to write the code underlying pewresearch.org.
In parts of our research process. Researchers may use AI assistants to write the code needed to prepare, organize or run analysis on survey datasets. AI tools also assist with tasks involving textual data, such as coding open-ended survey responses into categories or scraping websites for key data. But even when we use AI, researchers design the analysis plans and interpret the data for our audiences.
In the final stages of our editorial process. Many widely available tools already use AI to help clean up grammar and punctuation. We’re experimenting with its use in initial phases of copy editing, though humans will continue to review and approve our final copy.
In creating derivative products. We’re also experimenting with using AI to assist in creating derivative products based on human-authored research, such as social media posts. AI helps adapt existing findings for different formats and audiences, but it doesnot generate the Center’s original analysis or introduce new interpretations. Humans review all AI-assisted content prior to publication.
We’ll be transparent
If we make meaningful use of AI in research production – for example, in analyzing textual data or coding open-ended responses – we’ll describe its use in the report’s methodology section.
If AI use expands in ways that materially affect how our external products are produced, we’ll revisit how we disclose that use publicly and update this article accordingly.
We’d love to hear your thoughts, hopes and concerns on this topic.
Note: This is an update of a post originally published on Aug. 27, 2024.