The Government Accountability Office (GAO) released an AI accountability framework back in July, but a GAO official said at NextGov’s Emerging Tech conference August 18 that the organization’s work on AI oversight remains a work in process.
GAO’s Chief Scientist Dr. Tim Persons explained how the office envisions the use of that framework, as well as took a look towards the future of GAO’s oversight of AI technologies. Persons said it’s important to get AI accountability right due because of the bad consequences that would flow form biased AI results.
“There’s no future in mitigating cybersecurity without some machine learning adaptable system being able to say, ‘This looks like something, bad actors are trying to hack into this event,’” Persons said at the virtual conference. “The future is now; this is already happening. The key thing is in how do you connect [the need for these systems with accountability and oversight.]”
“The risk is unjustly pulling out let’s say a US citizen and maybe identifying them and referring them to law enforcement unjustly,” Persons added. “So, the risk in that is PII (personal identifiable information), civil liberties kind of risk.”
Persons also warned of potential financial risks and regulation of autonomous vehicles by use of AI. He said the area of concern GAO hears from Congress the most about are when it comes to bias in AI, and making sure the programs are abiding by the Civil Rights Act of 1964 and protecting civil liberties.
Persons said in order to avoid issues like this, as well as avoid merely automating processes that are already not working as fairly as possible – such as the justice system, as Persons pointed out – developers have to start with the risks they’re trying to avoid.
As far as what comes next with GAO AI oversight, Persons said the GAO will learn through practice and look for patterns and ways to update its oversight framework.
“We are going to learn by doing, and, as it’s been the case in the past, … they will iterate, or update, or become sharper through lessons learned,” Persons said. “We can learn from the good, and the bad, the ugly from our position in the Federal space.”
“We believe that we will see over time patterns develop, tweak them or update the framework in some way, whatever the cadence might be,” Persons added. “But ultimately, hopefully, we will compile stories, tell the stories in an honest way, and then share best practices with the Federal government.”