Employees at the San Francisco-based machine learning and artificial intelligence company Figure Eight was unwittingly involved in Google’s highly controversial Project Maven AI program, based on a series of new emails received and reported by The Intercept.
According to information contained in the emails, after Project Maven kicked off in April of 2017, Google began looking to outsource the work of analyzing and labeling satellite images for the associated AI to learn from. In October of the same year, it began working with Figure Eight — known as CrowdFlower at the time.
The search giant was effectively sending the company raw images from a satellite technology known as wide-area motion imagery with the goal of training an AI for use by Air Force drones. Figure Eight’s role was to label objects, people, buildings, and other subjects in the image according to Google’s instructions, essentially drawing boxes around the items.
Left in the dark
That Figure Eight and its workers were kept in the dark about who or what the AI was being used for. Former employees have reportedly noted that isn’t uncommon in scenarios where a business set up to provide exactly those types of identification services for machine learning. More often than not, employees of those companies aren’t made aware of who a given client is or what the project will be used for.
Google’s run-in with its own employees and more general backlash associated with the project have been widely reported since news first broke. Summarily, in the case of Project Maven, the overarching goal was to enable rapid identification of enemy combatants and other key objects using military drones. In particular, the technology was expected to see use on active battlefields and similar areas.
The search giant has since abandoned the project and made sweeping changes to its AI policy but news about the specifics of image analysis and how Project Maven progressed may be indicative of a deeper problem.
A source of future controversy?
The apparent fact that Figure Eight and its employees were left unaware of the military-centric goals of their work and that it is a common industry practice is likely to be concerning to many. Similar work may have been conducted or could be in the future, to begin with, but Figure Eight isn’t the only so-called “human-in-the-loop” business in operation either. The concepts behind the business are beneficial in creating realistic AI models that are contextually aware and even Amazon has a division dedicated to the task called Mechanical Turk.
The newly reported emails, although exact details have not been reported, will open an array of questions for those opposed to the idea of AI being used on a battlefield or in similar scenarios. Although Google has suspended work on Project Maven and vowed to not use or sell its AI for any use without consideration for the consequences moving forward, it has stopped short of promising not to get involved in similar projects.
Taken in combination with the fact that other organizations and companies are actively pursuing futures in AI, that could leave plenty of room for future controversy and backlash.