In this fortnightly online column, ADM journalist Corey Lee Bell covers defence news across Japan and the Republic of Korea.
March saw developments in two artificial intelligence (AI) research projects commissioned by Japan’s Acquisitions, Technology & Logistics Agency (ATLA), as Japan’s Ministry of Defense (MoD) ramps up efforts to implement directives on AI induction set out in the 2022 National Defense Strategy and Defense Buildup Program.
Sakana AI wins contract for data analysis/decision making study
On 13 March, Nvidia-backed unicorn Sakana AI announced that it won a tender from ATLA’s Defense Innovation Science and Technology Institute (DISTI) to conduct research on multi-model data analysis and decision making AI.
According to a company press release, a primary aim of the project is to produce an AI system that can integrate visual, acoustic and other forms of data generated across land, air and sea domains, and accelerate decision making tasks such as resource allocation to enhance the effectiveness of Japan’s Command and Control system (C2 system).
Another objective is to create a small vision language model (SVML) that can make drones and portable control terminals “capable of rapid and efficient situation reporting and information management.”
The research project falls under the umbrella of DISTI’s Breakthrough Empirical Research, a US Defense Advanced Research Projects Agency (DARPA)-inspired initiative that seeks to induct civilian-sector technology and know-how which could potentially transform the nation’s defence capabilities.
During last December’s ATLA Defense Industry Promotion Exhibition (DIPEX) Sakana AI showcased technology relevant to the project, including its Temporally Adaptive Interpolated Distillation (TAID) AI technology.
The company claims it can create high performing Small Language Models (SLM) that can function on small edge devices including mobile phones and small drones.
Fujitsu sets up Accelerator Program for Defence Tech
Sakana AI’s announcement coincided with Japanese information and communications technology (ICT) giant announcing the establishment of the Fujitsu Accelerator Program for Defence Technology.
The new initiative is described as an open innovation program for supporting groundbreaking research on defence-related multi-AI agent systems, according to the program’s official webpage.
Multi-AI agents refers to a framework where a number of specialised AI with different roles, data sets and capabilities collaborate to solve problems that are too complex for a single AI model.
The development followed Fujitsu securing a two-year contract from ATLA to develop an experimental multi-AI agent system which can support military commanders to accelerate decision making and reduce staff/manpower requirements.
Concrete aims for the program outlined by Fujitsu also include developing a large language model (LLMs) that can enable the use of simple natural language prompts to direct drones to perform complex tasks semi-autonomously.
The Accelerator program will provide funding and other resources to support non-defence sector startups with expertise in multi-AI agent systems to collaborate with Fujitsu on the ATLA commissioned project.
An information session for program applicants was held on 12 March, with Fujitsu set to announce the successful applicant in June.
Open innovation to tap into “software-centric” civilian sectors
The rise of open innovation contracts comes as Japan seeks to reach out beyond its traditional hardware-centric defence industry stalwarts to bolster its software/AI-capabilities.
This follows calls in last year’s “Fundamental Strengthening of Defence Capability” report to further invest in AI and broaden collaboration between the MoD/ATLA, startups and research institutes.
Recent initiatives to bolster the participation of tech startups in the defence sector include ATLA announcing a “Fast Pass Procurement” program and industry support sub-agency for startups late February, and the concurrent formation of a Small Business Innovation Research (SBIR) System in the MoD.
A rapidly growing number of smaller firms specialising in AI have also been invited to collaborate with ATLA’s research institutes on projects involving automation.
High profile examples include ATLA last month awarding eight small tech firms/independent developers with cash prizes and opportunities for ongoing collaboration on defence projects, after thousands of developers submitted algorithms for publicly open simulated wargaming competitions aimed at training AI-driven air-to-air combat-support UAVs.
The last two months have also seen the release of additional ATLA tenders for AI research projects, including a study on AI tools for comprehensive analysis (information gathering, information management and analysis) and scoping research on shipborne generative AI platforms.