COMPUTING NEWS

One of the most pressing needs in today’s data-flooded world is for accurate data interpretation. This is especially true in the intelligence community. The US government works diligently to build barriers against cyber-attacks that can take down our financial systems and wreak havoc on our nation’s infrastructure; terrorism both domestic and international; chemical, nuclear, and conventional weapon strikes, and even threats against our democracy and election processes.

Professors Gheorghe Tecuci, Mihai Boicu, and Dorin Marcu, researchers from George Mason University’s, Learning Agents Center are on to something big. A creative, technical solution that combines their research in artificial intelligence and learning agents with crowdsourcing. The goal is to develop a reliable, flexible technical assistant that can help intelligence analysts evaluate evidence, determine what is credible versus a red herring, and to share that information with other analysts, who working together, can quickly assess critical situations.

The system is called Co-Arg (Cogent Argumentation System with Crowd Elicitation). The Learning Agents Center team’s research has just been published in two books, Knowledge Engineering: Building Cognitive Assistants for Evidence-Based Reasoning and Intelligence Analysis as Discovery of Evidence, Hypotheses, and Arguments: Connecting the Dots. They’ve also taken their work into the classroom and out into the field.”

This past January, the LAC team, which is also supported by several researchers from Mason and five other universities, was awarded a $7.4-million-dollar contract from the Intelligence Advanced Research Project Activity (IARPA). The aptly named CREATE program: Crowdsourcing Evidence, Argumentation, Thinking and Evaluation is charged with finding technical solutions to assist intelligence analysts with data analysis, to dramatically improve analytic reasoning and the resulting analytic products.

Gheorghe Tecuci, the project PI speaks for his team when he says how eager and excited they all are to expand their work and ideas. They are confident they can develop a novel working solution and point to the fact that they are well on their way with the research.

The Co-Arg system has two main components, Cogent and Argupedia. Cogent is a cognitive assistant with advanced analytic and learning capabilities that directly supports a lead analyst in answering intelligence questions. Boicu explains that Argupedia is an innovative wiki for crowd problem solving allowing multiple crowd analysts to work through specific intelligence requests from the lead analyst, regardless of their physical location, and determining the most likely solution. Argupedia is also an encyclopedia of arguments. The stored information can be used over and over for multiple tasks and it will become more comprehensive and accurate as

intelligence problems are solved and incorporated into the wiki.

The Co-Arg system evaluates scenarios using an original Wigmorean probabilistic inference network. Tecuci explains that the idea emerged from an integration of previous LAC research on problem solving through analysis and synthesis, with the Wigmorean graphical method used in the legal profession to analyze legal evidence in trials. This part of the project was influenced by LAC’s David Schum, a founder of the science of evidence. One of the biggest hurdles is how to design a system that will be easily used by non-technical analysts. Tecuci says that the team fortunately has years of experience teaching the use of their AI systems at the U. S. Army War College and understands the population of people likely to use Co-Arg.

The Learning Agents Center is working on several projects related to the Co-Arg AI research including, sInvestigator, an education program to teach undergraduate students critical thinking skills in science, funding through NSF’s IUE Program; cognitive assistants that capture and automatically apply the expertise employed by cybersecurity analysts when they investigate Advanced Persistent Threat alerts, funded through the AFRL ADCO program; and a Mitre/NRO program, Towards Persistent Intelligence Processing.

Learn more online at: http://lac.gmu.edu