Issue link: https://read.uberflip.com/i/1173432
This and similar non-technical hurdles are harming the US military's ability to leverage the best commercial AI capabilities; hurdles most international competitors are not burdened with. The good news is that these challenges can be mitigated with a modern business approach to defense electronics investment, design and manufacturing. This will be explored in more detail in another white paper in this series as it pertains to the acquisition of commercial AI capabilities for military missions. Incidentally, Eric Schmidt (Alphabet chairman) is now calling Google an AI company, not a data or advertising company". Technology challenges - Embedding High Performance Computing (HPC) for AI Industry knowledge regarding AI and deep learning system architecture is from the perspective of high performance computing (HPC) or supercomputing and is synonymous with the data center environment. Traditional look-up tables that powered previous generations of EW and SIGINT missions are being upgraded with smarter, cognitive payloads that can analyze new threats and signals on the fly. HPC data centers favor more hardware over optimizing their software's performance. More servers, reduced footprints or higher quantity units are secondary considerations within data centers as abundant resources mitigate overall processing limitations. Within data centers, multiple- way redundancy is common, as is the installation of over-capacity. As such, data center servers are managed like cattle rather than pets as hardware is expendable compared to software execution. HPEC or high performance embedded computing is HPC embedded within a system or platform as is the often the case with military applications. Military HPEC systems are required to fit into physically restrictive, power-constrained environments and are deployed in isolated, hostile and contested locations. These applications emphasize hardware efficiency and their software is equally refined for efficiency. 5 • Monitor international AI progress, especially in regard to achieved technology milestones • Work closely with industry and academia • Prioritize basic and long-term AI research vectors • Include the disruptive influence of AI on cybersecurity and cybersecurity on AI The policy's intent was clear - Regulation and funding must support commercial and defense agencies in AI and cognitive computing efforts, to bridge the gap between the theoretical and the practical. Open standards for hardware, data and algorithms should be adopted to de- silo information, and aid interoperability and rapid development. Nowhere is the need to adopt AI more critical than in the military domain. New military missions will require an AI advantage to take on international competitors who are also adding AI to their offensive systems. Most current artificial intelligence IP is being developed within the commercial domain. Thus, future global stature within an AI-enabled era requires leveraging this commercial IP and making it ready for military and defense applications. Non-technology related challenges complicate an already impressive technical challenge associated with embedding AI capability into military missions. Non-technical challenges include accessing the best commercial IP, limited commercial partnerships and co-investments. Further, whereas the commercial domain is very agile, it lacks product sustainment and business practices that are compatible with the defense industry as a whole. Leveraging the best commercial artificial intelligence IP requires a different approach to acquisition than is currently widely practiced within the defense industry. Consider the case of the much publicized Maven Project, a venture between government and a leading commercial AI provider. Commercial challenges - The Maven project The government-funded project nicknamed Maven, leveraged Google technology to process information more efficiently and accurately through machine learning. The project used Google's open-source TensorFlow artificial intelligence API to analyze UAV sensor data, detect objects of interest and flag them for human analysis later. Currently, sensor processing algorithms can detect basic objects including cars and people, but have a harder time detecting the greater array of objects that are of military interest. The Maven Project uses Google's TensorFlow API to fill this capability gap. As Google worked on the project, it encountered resistance from its employees who viewed enabling military missions contrary to their values. Suppling their technology for use on a military platform caused many employees to refer to their company motto: "Don't be Evil." Facing a revolt, Google announced that they would not be pursuing future contracts in the military realm after the current one expires in 2019. Google onboard caused an employee revolt within the tech giant

