NBC Built a Tool That Shows If IBM Used Your Face to Train AI

Interactive feature is part of a report on the company's data use practices

NBC recently beefed up its data journalism team. - Credit by Getty Images
Headshot of Patrick Kulp

NBC News has created a widget that will tell you whether IBM used photos from your Flickr account to train facial recognition artificial intelligence.

The interactive feature is part of a deep dive the network did on the widespread use of publicly available images to train algorithms without their subjects’ consent, a common practice in the artificial intelligence community with questionable privacy implications.

NBC News’ Olivia Solon reported that IBM entered this fraught territory in January when it released a collection of nearly 1 million photos taken from Flickr and tagged with AI-generated descriptors. The tech giant promoted the diverse makeup of the data set to researchers as a step toward solving the bias problems that plague AI systems.

IBM declined to share the data set with the news outlet, arguing it was only available for corporate or academic research purposes, but a source leaked it anyway, and NBC News data editor Joe Murphy saw an opportunity. Together with developer Jeremia Kimelman, he built out a searchable input for the data set to embed in the article.

Users simply enter their Flickr handle, and the feature returns the number of your photos included in the data set and an example of one.

“It’s a mix of technical chops and journalism chops,” Murphy said. “We’re able to find the right way to visualize data together with the editorial team and help drive story decisions in general, rather than simply act as a service desk.”

Like other news outlets navigating the digital media landscape, NBC News has been beefing up its capacity to integrate big data and other technical skills into its reporting. The company hired Murphy last year along with a four-person data-focused team to support him.

Well-groomed data sets that easily lend themselves to interactive features don’t come along often, but when they have, it’s generated traffic, Murphy said. He cited a tool the team built to inform people about the California wildfires last November as another example.

“We found that the more interactive our visual or our piece is, the more people are interested in it,” Murphy said. “That’s something that doesn’t happen every month, and that we wish happened more.”

@patrickkulp patrick.kulp@adweek.com Patrick Kulp is an emerging tech reporter at Adweek.
Publish date: March 14, 2019 https://dev.adweek.com/digital/nbc-built-a-tool-that-shows-if-ibm-used-your-face-to-train-ai/ © 2020 Adweek, LLC. - All Rights Reserved and NOT FOR REPRINT