A Startup Is Suing Facebook, Princeton For Stealing Its AI Data
The legal wars over AI training data are just beginning.
A Lithuanian company called Planner 5D is suing both Facebook and Princeton University for stealing its artificial intelligence training data — an early skirmish in the strange new legal frontiers of AI.
Princeton computer scientists scraped more than 45,000 files from Planner 5D’s software, according to a lawsuit filed on Wednesday, and used them to train their artificial intelligence algorithms. They then made that information, which they named the SUNCG dataset, available to other computer vision and artificial intelligence researchers, the legal document claims.
Through partnerships with Princeton, the data made its way into the hands of Facebook scientists, who used it to develop the company’s AI — and the lawsuit also suggests that the data could have benefitted Facebook’s virtual reality company, Oculus.
Facebook made the dataset, full of original and lightly-modified Planner 5D files, available as a resource to contestants in the 2019 Scene Understanding and Modeling challenge for computer vision researchers, who are competing for a cash prize and a speaking slot at a conference later this month.In short, Planner 5D claims that Princeton and Facebook have benefited off data that was wrongfully taken from it — and now the company wants them to pony up.
Let’s Back Up
Planner 5D offers software that lets users design homes, buildings, and interiors by arranging 3D models of various objects. It’s sort of like building a home in “The Sims,” except it’s meant for people who will actually build their creations. People can also navigate their designs in virtual and augmented realitysimulations.
Per the lawsuit, Planner 5D says it created 3D models of over 4,500 objects as well as over a million pre-designed scenes for use through its software. All of that data, the company says, is invaluable for scientists training AI algorithms to navigate physical spaces and recognize objects.
Alexey Sheremetyev, CEO and Founder of Planner 5D, declined to answer specific questions about the ongoing lawsuit. But he gave Futurism a general statement on the matter.
“A copy of our data has been used by researchers at Princeton, Facebook, and elsewhere,” he said. “We are very troubled by this, because this is a core asset of our company. We spent a lot of time and money creating this asset. We think this data is a key ingredient for AI scene-recognition research. We brought this lawsuit to address what we regard as a serious threat to our company.”
A Princeton representative declined to comment for this story. As of this article’s publication, Facebook representatives have not responded to multiple requests for comment.
The URL for the SUNCG dataset provided in the court filings now redirects to Princeton’s homepage, suggesting that it has been removed. But the page still shows up in the website directory, and references to the dataset exist on other pages of Princeton’s website as of June 7. A previously-archived versionof the website shows that the dataset was available to researchers as recently as October 2018.
Princeton University scientists, collaborating with Adobe Research, first published AI research that used the SUNCG dataset in late 2016. The next year, Princeton computer scientist Thomas Funkhouser presented Facebook-funded studies that used the SUNCG dataset to teach computers to navigate indoor environments. In that presentation, he identified “data” as the main roadblock for the field of 3D scene understanding research.
The best place to find that data, according to the lawsuit, was Planner 5D. That 2016 study acknowledges that the 3D models used to train the AI system were downloaded from Planner 5D’s website. And thanks to partnerships with Princeton, Facebook got its hands on the data shortly thereafter, using it to further company’s own computer vision and virtual reality research.
Both Facebook and Princeton scientists shared the data, which originated from Planner 5D, with scientists in their field, according to the suit. While this open-source mindset — common among computer scientists and AI developers — can accelerate scientific progress, it only really works when everyone agrees to share, not when one group helps itself to another’s work.In this case, Planner 5D wants Princeton and Facebook to pay up, arguing that its continued success depends on its sole ownership of the models and scenes that it created.