RTX’s BBN division has secured a contract award from DARPA’s “In The Moment” program, aiming to develop trusted algorithms that can make independent decisions in complex and dynamic scenarios. The program specifically focuses on situations like mass casualty triage and disaster relief, where there is often no clear right answer and no consensus among humans.
According to Alice Leung, the principal investigator at Raytheon BBN, the goal of the program is not simply to train AI to provide correct answers in controlled scenarios. It involves creating AI systems that humans would trust to make decisions independently in uncontrolled environments. To achieve this, the team will conduct research on both decision-making and trust.
The Raytheon BBN-led team, which includes Kairos Research, MacroCognition, and Valkyries Austere Medical Solutions, will employ a cognitive interviewing technique to understand how medical professionals and first responders evaluate information and make difficult trade-offs during critical decision points. This qualitative information will help design scenario-based experiments to study individual decision-making attributes and their impact on choices. It will also explore how the alignment of attributes between different individuals affects the willingness to delegate decisions to others. Ultimately, this research will enable AI to be customized to match the attributes of expert populations or individual experts.
Leung highlights that due to the inherent variability in human decision-making, it is unlikely that a one-size-fits-all trusted AI model exists. Instead, the goal is to create AI systems that can adapt to different users and domains. By tuning an AI’s attributes such as risk tolerance, process focus, or willingness to change plans, it can better match the preferences of users or user groups.
DARPA has brought together multiple teams to collaborate on the “In The Moment” program. These teams will focus on developing prototype AI decision-makers that can be tuned to match specific target attributes after being initialized with baseline knowledge. The research products from these teams will be integrated and evaluated to determine how well the algorithmic agents align with target human attributes when faced with challenging scenarios. The program will also assess whether human experts trust these aligned agents more than baseline agents or even other humans. In evaluations of trust, human experts will review decision records without knowing whether the decision-maker was an AI or a human.
The work carried out under this contract, sponsored by DARPA and the Air Force Research Lab, will take place in Cambridge, Massachusetts; Dayton, Ohio; and Anniston, Alabama.
Raytheon BBN, founded in 1948, focuses on advanced technology research and development with a national security priority. The company has a history of transitioning advanced research into innovative solutions for its customers, from the creation of the ARPANET to pioneering email systems. Raytheon BBN specializes in analytics and machine intelligence, networks and sensors, intelligent software and systems, and physical sciences.
RTX, the world’s largest aerospace and defense company, comprises leading businesses like Collins Aerospace, Pratt & Whitney, and Raytheon. With over 180,000 employees worldwide, RTX continuously pushes the boundaries of technology and science to redefine how we connect and protect our world. The company is headquartered in Arlington, Virginia and recorded $67 billion in sales in 2022. Their focus areas include advancing aviation, engineering integrated defense systems, developing next-generation technologies, and manufacturing solutions to address critical challenges globally.
In summary, RTX’s BBN division has received a contract award from DARPA’s “In The Moment” program to develop algorithms that can make independent decisions in complex scenarios. The research will explore decision-making attributes and trust factors to enable the customization of AI systems that match the preferences of users or expert populations. The program brings together multiple teams to collaborate on the development and evaluation of AI decision-makers aligned with human attributes. The work will be conducted across different locations, and the contract is sponsored by DARPA and the Air Force Research Lab.