AI-enabled voice assistance for responders
- By Stephanie Kanowitz
- Sep 25, 2019
First responders are testing a NASA-developed artificial intelligence tool that may seem like just another Alexa or Siri. AUDREY, however, can turn voice commands and images into data that facilitates emergency teams’ lifesaving work.
Most recently, the Next Generation First Responder (NGFR) Apex program at the Department of Homeland Security’s Science and Technology Directorate tested the tool, called the Assistant for Understanding Data through Reasoning, Extraction and Synthesis, or AUDREY for short. The experiment, which took place in April in Ontario, examined how AI could make paramedics more effective and efficient. It was conducted in partnership with Canada’s Department of National Defence Science and Technology Organization, Defence Research and Development Canada Centre for Security Science (DRDC CSS).
Through a contract S&T had with NASA’s Jet Propulsion Laboratory, which developed the system, the directorate, NASA JPL, DRDC CSS and the Hastings-Quinte Paramedic Service in Ontario set up a scenario in which a dummy patient was experiencing chest pains. AUDREY was asked to perform several functions such as responding to voice inputs, transcribing those voice commands and tracking and validating medication dosage and dispensing frequency. AUDREY was also tasked with recognizing medications from images of bottles. At the end, the tool automated the completion of a paramedic treatment record.
Similar to the voice activation feature on a smartphone or voice-controlled digital assistant, AUDREY is personalized to responders and can recognize their jargon. AUDREY also uses human-like neural symbolic processing for cognitive reasoning, meaning it collects data from sensors on responders’ personal protective equipment and a suite of plugin tools to improve machine intelligence. It turns those into insights for responders that they might not have had during an emergency, according to S&T.
“AUDREY can receive data without being required to respond one-on-one, and at the same time, it can feed thousands or hundreds of thousands of data points -- such as reporting water levels from sensors -- and then [get] that back to the first responder,” said Denis Gusty, a program manager at S&T.
During the experiment, paramedics held a cell phone-like device that could accept voice commands and “read” photographs, transcribe them and synthesize all of that as data. In a real-world situation, the data in the cloud would be accessible to a 911 call center, a receiving hospital or both, depending on the customer, Gusty said.
A major takeaway from the experiment was that the form factor is not viable yet, he said. “Paramedics need both hands to operate and, in this case, they couldn’t do that using AUDREY,” Gusty said. “They actually had to hold in one hand a handheld device they had to talk into to get the commands back to AUDREY. That just didn’t work very well. That was the biggest limiting factor in our experiment.”
AI has great value for paramedics; the practicality of AUDREY isn’t there yet, he added.
“Unless you ride in the back of an ambulance, you don’t know how crazy it is to be a paramedic. They’re just constantly moving and moving and moving. Then, at the end of the day, they’re forced to sit down and write up their notes," Gusty said. "So one instance where artificial intelligence comes into to play is it can actually help the first responder with that treatment record, constructing that while it’s happening, rather than waiting until after the fact.”
Another way AUDREY can help is by confirming a drug and dosage is correct before paramedics give it to the patient. “We took a picture of the medication the paramedic was about to administer and sent that to AUDREY, which confirmed it was the right drug,” Hastings-Quinte Paramedic Service Chief Doug Socha said in a DHS announcement. “Paramedics are in the back of the ambulance by themselves making critical decisions, and they don’t have the luxury of the emergency room with other professionals who can be the second set of eyes to ensure the right medication is about to be administered.”
Because the contract with JPL expired, additional testing of AUDREY is on hold for now, Gusty said.
S&T has conducted pilot tests of AUDREY with other first responders, too. For instance, in fall 2017, it ran its first test at the Multi Agency Communications Center in Grant County, Wash. There, the focus was on data collection during 911 calls and how to improve text-to-911 services.
In a staged video about NGFR, DHS shows how a police officer responding to a building explosion could contact other officials by saying, “AUDREY, patch this call to Corp. Williams on his radio,” or “AUDREY, give me incident command.”
“I think the potential for artificial intelligence coming into play with first responders on a day-to-day basis is just astronomical,” Gusty said. “For someone or something, in this case, to process millions of data points in split seconds, I think that’s the beauty of artificial intelligence -- the computing power that someone has at their fingertips.”
Stephanie Kanowitz is a freelance writer based in northern Virginia.