What's government's role in AI?
- By Matt Leonard
- Dec 12, 2017
At a Senate hearing on the future of machine learning and artificial intelligence, legislators wanted to know about the government’s role as both an end user and enabler of the technology.
“One of the things I think we should be thinking about is our role as an actually user,” Sen. Maria Cantwell (D-Wash.) said in the Dec. 12 hearing held by the Senate Committee on Commerce, Science and Transportation. She was specifically interested in what AI could do to improve cybersecurity.
Cyber defense systems will need to respond at machine speed and find vulnerabilities, said Edward Felten, a computer science and public affairs professor at Princeton University. “There is a huge opportunity there [for AI], and it is rapidly becoming a necessity as the bad guys are adopting AI and automation in their cyberattacks,” he said.
Besides helping secure systems, AI could also help government save money. The savings could be as much as “$41.1 billion annually by using AI to automate tasks,” Information Technology and Innovation Foundation's Vice President Daniel Castro said, citing a Deloitte publication.
But agencies haven't yet been able to get much traction for AI-based applications, Castro said. “We’re so far behind in actually doing this in government, and I think that’s an area that should be frustrating to policy-makers especially because if they want to understand it well, they should be using it on a daily basis,” he said. There are people asking questions, he said, but starting pilots and scaling them across agencies has “been a challenge.”
One of the reasons for the slow uptake could be the lack of resources for managers, Castro suggested. If someone is interested in automation, there isn’t a place to turn for best practices. “We’re starting to do this through [the General Services Administration], but we’re not there yet,” he said.
GSA launched the Artificial Intelligence for Citizen Services program last year for agencies interested in leveraging the capabilities of tools like Amazon Alexa, IBM Watson, Microsoft Cortana and Google Assistant. The Obama White House created a chatbot to make it easier for citizens to send the president questions through Facebook’s Messenger. There are also opportunities for automation when it comes to logistics (like fleet management) and process robotics, which helps move forms and data through workflows.
AI-enabled robotics also could benefit law enforcement agencies, according to Cindy Bethel, an associate professor in the computer science and engineering department at Mississippi State University.
Bethel is working with computer vision and sensing to help robots gather information and make decisions about a situation or environment before putting the lives of law enforcement officers on the line. Her research can help officers "determine what risks are in the area so they know what they’re entering into,” she told GCN.
However, it's much more difficult to put these kinds of capabilities in a robot than in a car, she said. The smaller size of robots can reduce the available computing power, which then limits capabilities for streaming video and other applications, she said.
While the government will no doubt benefit from AI, it will likely also play a role in improving the technology by supporting efforts to eliminate the potential for bias within the algorithms AI systems use. Castro said government data could help ensure AI-based models accurately reflect the people they’re analyzing by providing accurate and trusted data about those populations. The government could open up its data through open application programming interfaces or direct downloads, he suggested after the hearing.
Although Sen. Ted Cruz (R-Texas) raised the specter of Skynet -- the self-aware AI system trying to destroy the human race in the Terminator movies – just as he did at the Senate’s first hearing on AI just over a year ago, witnesses downplayed the worst-case scenarios. General AI, or artificial intelligence equivalent to that of humans, is far from being realized. Plus, researchers working in the field tend to be the least fearful of the technologies, Castro said.
“General AI is not going to kill us all,” he reassured the senators.
Matt Leonard is a former reporter for GCN.