AI Bulletin – April 24, 2020
Want the AI Bulletin mailed to you?
- Spotlight on copyright and trade secret suit involving AI-assisted computer vision
- U.S. initiative on combatting COVID-19 with AI
- FTC guidance on the use of AI
- Guilty plea for trade secret theft by former Uber employee
- CFAA found inapplicable to terms-of-service violations
- Impact of COVID-19 on EU digital legislation regarding AI training data
- New SCOTUS petition on access to public user data under CFAA
SPOTLIGHT – PLANNER5D LITIGATION
Background In June 2019, Planner5D filed a complaint alleging copyright infringement and trade secret misappropriation against Facebook and Princeton University. The case is before Judge William Orrick in in the Northern District Court of California.
Planner5D is a mobile app development company based in Lithuania. Its 2D and 3D home design tools allow users to select from thousands of virtual objects, such as furniture and electrical appliances, to create simulated interior design scenes.
As described in its complaint, Planner5D created a library with thousands of digital objects as well as compilations of design scenes comprised of these objects—a process spanning over seven years. In 2016, Facebook allegedly encouraged Princeton researchers to make unauthorized use of these objects and scenes as AI training data for enhanced computer vision. According to Planner5D, the company’s lifelike digital scenes were only possible because human designers crafted each three-dimensional object, with Princeton recognizing such realism “was vital for accurate machine learning.”
Complaint and Causes of Action The complaint asserts claims of copyright infringement for unauthorized duplication and distribution of Planner5D’s objects and scenes, and of the compilations of the same, as well as state and federal trade secret misappropriation claims over the underlying data files of those objects and scenes. Planner5D specifically asserts that the Princeton researchers crawled the company’s servers and downloaded thousands of objects and scenes, then used that data to publish articles regarding scene-recognition and released the dataset to the public. Planner5D also accuses Facebook of launching a scene-recognition competition (with the help of Princeton researchers), that directed contestants to use the Planner5D dataset in their submissions.
Motions to Dismiss. Facebook and Princeton moved to dismiss the complaint for failure to state a claim. The court granted both motions.
Copyright. The court held that Planner5D had not adequately alleged that it registered its work with the Copyright Office, nor had it shown it was exempt from doing so on the basis that the objects and scenes were non-United States works.
Facebook additionally argued that Planner5D failed to plausibly allege that its works were copyrightable because they were not original and creative. The court agreed, holding that Planner5D had to sufficiently allege “more than time and effort” in creating the individual objects and scenes as well as facts regarding the “creative selection process” and “creative choices” behind its scene compilation.
Trade Secret. The district court determined that Planner5D failed to plead its trade secret claims because the complaint did not allege how the company maintained the secrecy of its underlying data files, or what improper means the defendants had taken to obtain those files.
In its order, the Court also granted Planner5D leave to amend to address these deficiencies.
Amended Complaint. In December 2019, Planner5D filed an amended complaint asserting only its trade secret misappropriation claims, as well as a separate notice expressing its intent to reallege its copyright claims once it had fulfilled the registration requirement with the Copyright Office. In March 2020, the parties sought clarification with respect to how the litigation should proceed with respect to the trade secret claims, given that Planner5D could not yet refile its copyright claims. The Court determined that upon satisfying the registration requirement, Planner5D could reassert its copyright claims in a new lawsuit, which would then be related to and consolidated with the pending trade secret claims.
What’s Next. On March 31, 2020, Planner5D filed the new lawsuit with its copyright claims and filed an administration motion to relate the two cases. On the same day, the parties also filed a joint stipulation agreeing that Facebook and Princeton would each file a consolidated response to the amended complaint and the new complaint within 21 days after the Court issued an order consolidating the two actions. The Court has yet to rule on either request.
U.S. GOVERNMENT & REGULATION
The U.S. Office of Science and Technology Policy has announced the COVID-19 Open Research Dataset, an initiative aimed to better understand how AI can help analyze scientific literature regarding coronavirus and avoid information overload. The initiative builds on the Allen Institute’s Semantic Scholar project and was created in conjunction with various technology leaders including the Chan Zuckerberg Initiative, Microsoft Research, and Kaggle (Google’s machine learning and data science community).
Anthony Levandowski, the former Uber Technologies executive, pled guilty to one count of Theft and Attempted Theft of Trade Secrets. Mr. Levandowski was originally referred for criminal investigation on May 11, 2017 by Judge William Alsup, in connection with the Waymo v. Uber llitigation over which he was presiding (and in which Quinn Emanuel represented Waymo). The misappropriated technology at issue involves Google’s self-driving car program (now known as Waymo), where Mr. Levandowski worked until January 2016. In the plea, Mr. Levandowski admitted that he “intended to use the [misappropriated Google trade secret document] to benefit myself and Uber.” Mr. Levandowski agreed to pay restitution to Google in the amount of $756,499.22. He faces up to 10 years in prison, although prosecutors are only recommending a sentence between 24 to 30 months. A sentencing hearing has been set by Judge Alsup for August 4, 2020.
COMPUTER FRAUD & ABUSE ACT
A federal district court has ruled in Sandvig v. Barr that violating a website’s terms of service does not violate the “Access Provision” of the Computer Fraud and Abuse Act. The CFAA criminalizes "access[ing] a computer without authorization or exceed[ing] authorized access.” A group of academic researchers wanted to test whether job websites’ proprietary algorithms were discriminatory by providing false information, thereby violating the sites’ terms of service. The researchers brought a pre-enforcement challenge arguing that the CFAA violated their First Amendment rights to free speech. On March 27, 2020, the District Court for the District of Columbia determined that websites’ terms of service do not constitute “permission requirements” under the CFAA because their long, dense, and changing terms provide inadequate notice to trigger criminal liability (among other reasons). Moreover, the court wrote, allowing private websites to define the scope of criminal liability “raises considerable nondelegation issues.” The court did not reach the First Amendment issue raised by plaintiffs in dismissing the case.
The coronavirus pandemic has prompted the European Union to rethink the rules and safeguards regarding AI it unveiled in a white paper earlier this year. Although the strategy called for European AI systems to be trained using European data, the EU noted the outbreak has exposed the limitations of restricting training data and delaying access. According to an official, the EU is “thinking more actively about the unintended consequences” of its proposal. The consultation period for public comment on the white paper was set to end in mid-May, but is likely to be pushed back.
The EU may also revise its Product Liability Directive, which was enacted in 1985, in light of the technological advancements such as self-driving cars and smart self-learning appliances. If the directives remain unchanged, a “manufacturer's liability could be at stake each time it is believed that the technology attached to a product could not really be controlled by the user.”
Quinn Emanuel represents technology startup hiQ Labs against LinkedIn before the United States Supreme Court in a Computer Fraud and Abuse Act (CFAA) appeal. On March 9, LinkedIn petitioned the nation’s highest court to determine whether hiQ violated the CFAA by accessing public LinkedIn user data to support its “people analytics” tools. HiQ’s business clients use these tools, for example, to identify employees who are at risk for leaving to another company. After receiving a cease and desist letter from LinkedIn, hiQ brought suit in 2017 and requested preliminary injunction from the Northern District of California. The district court found that hiQ was likely to succeed on its claims and granted hiQ’s motion. The Ninth Circuit upheld the decision, recognizing that that “[i]t is likely that when a computer network generally permits public access to its data, a user’s accessing that publicly available data will not constitute access without authorization under the CFAA.”