US Army Creating New Approach to Real-Time Conversational AI
With the new classification technique, the classifier would be able to properly respond to new commands and determine when to ask for help appropriately, given incomplete information.
US Army researchers have created a pioneering technique for combat vehicles and autonomous systems to interpret the intent of a soldier through spoken dialogue and respond appropriately, the army reported on this week.
With this approach — developed by the US Army Combat Capabilities Development Command, Army Research Laboratory, and the University of Southern California’s Institute for Creative Technologies — a robot may receive a command to turn 45 degrees and send a picture, and it will interpret the instruction and carry out the task, the report stated.
According to the army, this technology is currently the primary dialogue processing component for its Joint Understanding and Dialogue Interface system, a prototype that allows for back-and-forth dialogue between soldiers and autonomous systems.
New Classification Technique
With the new classification technique, the classifier would be able to properly respond to new commands and determine when to ask for help appropriately, given incomplete information.
“We employed a statistical classification technique for enabling conversational AI using state-of-the-art natural language understanding and dialogue management technologies,” Army researcher Dr. Felix Gervits said.
“The statistical language classifier enables autonomous systems to interpret the intent of a soldier by recognizing the purpose of the communication and performing actions to realize the underlying intent,” he added.
The technology also excels at handling noisy speech, including pauses, fillers, and disfluencies — features that are expected in a normal conversation with humans. It can also operate in real-time with no processing delay in the conversation.
New Versus Commercial Approach
The new approach is said to require fewer training examples compared to commercial deep-learning approaches that require large, expensive data sets. It is also able to reduce deployment time and cold start capability for new environments.
Compared to commercial dialogue systems, this pioneering approach can also focus on a search-and-rescue task that could occur in a future tactical environment involving both soldier and robot.
Additionally, the researchers claim that the new approach allows for better transparency. It could also better explain why the system produces a certain behavior, which is critical for military applications where ethical concerns are involved.
