Phase II year
2018
(last award dollars: 2023)
Phase II Amount
$2,499,922
The Navy must train and deploy effective teams in complex domains, from intelligence analysis to field medicine. Training individual personnel in all the skills and knowledge required by these tasks would be cost-prohibitive and inefficient. Instead, teams should be constructed to optimize their collective expertise. To construct these teams, the Navy needs a team development system that manages task and learning assignments to dynamically build the expertise required to achieve evolving task objectives. To meet this need, and based on our Phase I success, we propose to design and develop a full-scope system for Group Learning and Optimization of Collaborative Workflows (GLOW). GLOWs Collaborative Workspace will enable teams to decompose complex tasks into work chunks that can be performed by contributing team members, execute (or learn to execute) those tasks while efficiently applying the collective expertise of the team, and assess performance on those tasks through an integrated peer review process. We expect this system to immediately and tangibly benefit the Navy and DoD by enabling collaboration across a number of complex tasks, including collaborative intelligence analysis with DCGS-N
Benefit: N/aWe expect the full-scope GLOW system to immediately and tangibly benefit the Navy and DoD by enabling collaboration across a number of complex tasks, including collaborative intelligence analysis with DCGS-N. We also anticipate immediate benefits for personnel management; detailers will be able to construct more effective teams with the collective expertise to succeed. Commercially, the results of this effort are positioned to significantly impact the do-it-yourself learning community, a $200 million market in the US and UK alone.
Keywords: ADAPTIVE LEARNING, Market-based optimization, Argumentation, peer review, probabilistic programming, Crowdsourcing, collaborative analysis, collective expertise ----------- The DoD can maximize their strategic advantage by increasing the scale of collective problem solving to exceed current individual and team limits. Though promising results have been demonstrated for foundational technologies such as Artificial Social Intelligence (ASI), additional technology advances are required to develop more generalized frameworks and cross domain collective problem-solving support. We propose ASI Group Learning and Optimization of Collaborative Workflows (AGLOW). AGLOW will (1) integrate the Charles River ASI agent, from our DARPA ASIST work, into the GLOW framework; (2) extend GLOW with new domain-general processing components and reasoning algorithms; (3) adapt our ASIST ASI agent for a new domain; and (4) deploy the ASI agent via AGLOW to a real-world puzzle-hunt environment, a task that shares similarity to teamwork performed by Intelligence Fusion Cells. We expect AGLOW results to materially benefit the ASIST program, the human-machine-teaming AI research community, commercial products, and military technology transition roadmaps. AGLOW provides reusable data logging and intervention listeners used in real-world deployments, as well as an ontology for domain packages and abstractions for cross-domain and domain-specific activities, which can serve as reference implementations to help other ASI technologies generalize. AGLOW adds social-intelligence team support to commercial products.