Times Insider explains who we are and what we do and delivers behind-the-scenes insights into how our journalism comes together.

Last fall, the Climate desk at The New York Times asked readers to submit their questions on a complex topic: climate change.

Thousands poured in. Many of them were questions about the science of global warming and solutions. So editors and reporters narrowed the list to about 50 questions and began reporting out the answers.

Those responses appear in an interactive project that was published online last week. Readers can use a search function to browse questions or ask one of their own. To help guide readers to pertinent information, the team turned to an unlikely tool: machine learning, a type of artificial intelligence that draws connections and predicts outcomes based on data. Type in a question that has yet to be addressed (“What should I recycle?”) and the technology will direct you to a relevant answer, reported and written by a journalist. (In this case, the tool recommends reading the answer to “Should I bother recycling?”) The system also notes the new questions, so that journalists can consider addressing them.

“With this technology, we aim to get a response to each question in about 250 milliseconds,” said Jack Cook, a Times engineer who developed the technology for the project. He and other members of The Times’s Research and Development team have been working on machine learning since 2019; similar technology had also been used in 2020 and 2021, in F.A.Q.s about the coronavirus and its vaccine.

In an interview, Amelia Pisapia, a manager of emerging projects at The Times, and Jesse Pesta, the Climate desk’s deputy editor, explained how the tool came together and why human oversight is key to its success. This conversation has been edited and condensed.

What inspired this project?

JESSE PESTA Climate change is, fundamentally, a simple concept. We’re putting too much carbon dioxide into the air and it’s warming the world. But once you get past that, it can start to feel really complicated and confusing. There’s all kinds of jargon and acronyms. The science can start to feel a little bit tricky to understand. And the totality of all this can feel like, How do I even know what I’m supposed to know? This F.A.Q. is an exercise in meeting readers exactly where they are. We’re not guessing what readers want to understand the most — they’re telling us.

Coming into the project, did you have concerns that readers would look at this and decide: I can’t do much?

PESTA That is actually addressed in several places in the F.A.Q. itself, partly because these are the questions readers are asking, but it’s also reflected in places like the answers to: Are we doomed? Or how do I recognize and rebut bad information that I see people spreading in my life? There’s the feeling that there’s seemingly so much bad news on the climate front, but there are actually success stories. There are ways to grapple with these serious problems. And there are reasons for optimism. There’s one section in the F.A.Q. that addresses that head-on: “Where is the good news?” It’s quite a long section.

How does the search box work exactly?

AMELIA PISAPIA The search box’s technology is powered to understand both the questions that readers are asking and their intentions, and it attempts to match them with answers that have been written and edited by the Climate desk.

It’s a type of machine learning that is often called “human-in-the-loop.” We like to call it “editor-in-the-loop” here because it’s really important that we have very high oversight over this tool, monitoring it to make sure that it’s responding appropriately. Part of my job is monitoring questions and make sure they’re going to the right places. If it’s slightly off or there’s a slightly better answer for it, we can make these changes in the tool. The next person to ask that question is then directed to the most appropriate answer.

So I would imagine it requires near-constant human intervention to guide it.

PISAPIA I wouldn’t say near-constant, but I would say oversight, definitely. You never want to let these things loose in the world without having tight oversight over them.

PESTA A.I. is having its moment right now, as we all know — ChatGPT and so forth. This isn’t that kind of project. This is a human editor- and New York Times-journalist-driven project, first and foremost. It uses some sophisticated software to help pair peoples’ questions with the best answer that The New York Times can provide.

Did you encounter anything unexpected when receiving queries?

PISAPIA I would say no. My job is to be pretty skeptical of this kind of technology. So I try to operate from a place of deep skepticism and try to anticipate all the ways that this could falter.

It’s really important if you work with emerging technologies to think critically about them, about how they can work really well, what their weaknesses are and how can we account for some of those weaknesses. When I say that we’ve been thinking about this since 2019, we’ve been thinking about how to make this tool deliberate and in line with the work that we do across the report.