Budgets Confirm Tech Inquiry's Reporting on Scope of Project Maven
Budgets Confirm Tech Inquiry's Reporting on Scope of Project Maven
Beyond using artificial intelligence for processing aerial imagery, Maven also experimented with satellite imagery, "publicly available information", and captured enemy materials.
[2023-04-15, 6:22pm] Added a note on earlier budgets similarly detailing Project Maven’s scope.
The Office of the Secretary of Defense’s 2024 budget details much of the scope of the Pentagon’s pathfinder project for operationalizing artificial intelligence, Project Maven. While first reported by Tech Inquiry in 2021, the OSD’s budgets in fact detail Maven’s incorporation of “Publicly Available Information” as far back as in February 2018. With rare exception, reporting has generally been fixated on Maven’s application to drone targeting.
In two separate lengthy paragraphs, the 2024 budget request details Maven’s usage of the various intelligence collection ingredients which typically make up “all-source intelligence” analysis:
“The AWCFT (Project Maven) is the pathfinder artificial intelligence (AI) initiative for the DoD that accelerates the integration of AI into DoD systems to improve warfighting speed and lethality for the Joint Force…Maven’s AI architecture initially automated and augmented Processing, Exploitation and Dissemination (PED) of Full Motion Video (FMV) from Tactical Unmanned Aerial Vehicles (TUAVs). Maven additionally developed algorithms to Medium Altitude, High Altitude, and Wide Area Motion Imagery (WAMI) Intelligence and multiple other Surveillance, and Reconnaissance (ISR) platforms to support the National Defense Strategy (NDS). Maven includes AI tools used on Captured Enemy Material (CEM), Maritime, and Public Available Information (PAI) exploitation [emphasis added]…Maven developed a path forward to eliminate substantial costs and coordination among myriad legacy projects to instead use a single screen with multiple AI-enabled layers and tools…”
Rather than explicitly naming Project Maven’s 2023 transfer to the National Geospatial-Intelligence Agency (NGA), said intelligence agency is instead referred to as part of “various classified mission partners”. And a further paragraph notes that Maven is in support of the National Defense Strategy’s “peer/near peer competitor strategy”.1
Tech Inquiry analyzed proactively disclosed subcontracting records in order to report in September 2021 on what could be deduced about the scope of Project Maven. Major subcontractors underneath ECS Federal include:
Microsoft receiving $31.6 million for “[Electro-Optical] and [Infrared] Full Motion Video Machine Learning” and to “automate and augment the analysis of Wide Area Motion Imagery”, and
Clarifai being paid more than $25 million for the “Development of Facial Recognition”, to build a “[Captured Enemy Materials] & [Publicly Available Information] NSFW filter”, and to develop “Machine Learning...for Multispectral...still imagery from...satellite sensors..[and] [High Altitude] and [Broad Area Search] algorithms”.
The relationship between Project Maven and social media monitoring was also highlighted through a separate contract to Booz Allen which disclosed paying Yonder $600,000 for “social media data collection, normalization, aggregation, storage, and advanced analytics services through the New Knowledge Platform”. (Yonder, which was previously known as New Knowledge, was caught disseminating fake Russian disinformation in a U.S. Senate race. The company was acquired by Primer Technologies in June of 2022.)
Project Maven’s incorporation of social media monitoring is newly relevant in light of recent calls for U.S. intelligence agencies to more closely monitor social media and fora.
(Anyone interested in other classified projects mentioned in the 2024 defense budget should see Bill Arkin’s post from earlier today.)
The Office of the Secretary of Defense’s 2020 budget also notes the incorporation of the “High Accuracy Video Object Classification” (HAVOC) effort into Project Maven: “High Accuracy Video Object Classification (HAVOC): A prototype system that provided an automated, machine learning enabled, target recognition capability for expeditionary forces using a desktop computer. HAVOC transitioned to an intelligence community partner in 2018, and HAVOC technologies have been incorporated into Project Maven.”
A nearby item describes the “Social Network Aided Geolocation (SNAG)” project, which “prototyped a suite of automated machine learning algorithms that accurately estimate geolocations for social media messages from location-indicative terms and metadata features. This capability provides location estimates for the large volume of social media content that is not explicitly geotagged. Geospatial analysis of social media has proven effective for identifying and tracking actors of interest as well as understanding local concerns and sentiments within an area of interest. SNAG transitioned to the Defense Intelligence Agency’s GOSSIP architecture, which is widely used across the intelligence community.”