Person
1 episodes · 22 quoted claims
Jacqueline Acker said she had worked at the Central Intelligence Agency for over eleven years, having joined after working at a law firm in Oklahoma.
“I've been at CIA for over eleven years now, and I think that's, you know, I when I started there, I didn't know how long I'd be there.”
Starting around 2019, the intelligence community wrote principles for AI ethics adapted from existing principles for ethical intelligence.
“starting in about 2019, we wrote the principles for intelligence transparency, and those principles were based on, you know, general, uh, principles for ethical intelligence that had been around for a while.”
Acker said the intelligence community's AI ethics framework was developed with data scientists, lawyers and people from what she recalled as 17 intelligence community agencies.
“we worked really hard with data scientists, lawyers, and people from, eh, at the time, I think it was 17 different intelligence community agencies.”
The intelligence community's AI ethics framework was signed in June 2020.
“It got signed in 2020, um, in June, which was a kind of a crazy time.”
The intelligence community AI ethics framework was signed by the then Director of National Intelligence, who acts as integrator across the intelligence community.
“For the ethics framework, that was signed by, um, the then director of National Intelligence.”
Acker used an analogy that cars with brakes can go faster to argue that governance guardrails allow faster AI adoption.
“cars originally, you didn't have brakes, and so they couldn't go very fast, but once they had brakes, they could go faster.”
The Privacy Act of 1974 has required federal agencies to protect information about U.S. citizens and lawful permanent residents for over fifty years.
“for over fifty years, we have been doing privacy, um, thanks to the Privacy Act of 1974.”
Under the Privacy Act, systems that retrieve information by a unique identifier require public notice describing the system, its purpose and its protections, published on the agency's website.
“if you're able to retrieve information by a unique identifier, you have to have transparency to the public about, you know, what that system is, why does it exist, what protections are on that.”
The Privacy Act gives the public a right to access records about themselves and to amend them if wrong.
“they basically have this right to access their records and then amend them if they're wrong.”
Acker said many organizations are adapting federal privacy impact assessments into AI impact assessments.
“a lot of organizations are looking at the federal government's privacy impact assessments and saying, Oh, we can shift this into an AI impact assessment.”
Acker said the intelligence community cannot always be fully transparent because it must protect sources and methods.
“in the intelligence community, we can't always be as transparent as we would like because we have to keep sources and methods safe.”
Intelligence oversight includes cleared members of Congress, the Privacy and Civil Liberties Oversight Board, and the CIA's own privacy office.
“there's cleared members of Congress, there is the Privacy and Civil Liberties Oversight Board, there's even my office, so we have our own internal oversight requirements.”
The National Security Commission on AI issued a report of over 700 pages, released a few years before the generative AI boom, which frames AI as a values competition.
“We saw this even with the National Security Commission on AI report. It was, like, 700 and something pages long. It came out a few years ago before, again, the generative AI boom.”
Acker cited the OECD and the United Nations as sources of international frameworks of shared AI values.
“We've seen a lot of these, you know, OECD, um, definitely has several. Uh, we're seeing things out of the UN and otherwise.”
The intelligence community's AI ethics framework includes a full section of questions on mitigating undesired bias.
“our IC AI ethics framework, um, has an entire section on mitigating undesired bias, uh, and has a set of questions there.”
Acker said AI systems should not be reviewed only once a year, and that review frequency should depend on purpose, rate of change and consequences.
“you can't just, like, set it and forget it and come back to a system, like, once every year or something.”
Acker helped develop the IAPP's AI governance professional certification and teaches the class.
“I actually helped them develop the AI governance professional certification over the last couple of years, and I help teach that class”
Acker suggested an AI incident response program probably differs from privacy or cyber incident response because AI failures can come from wrong data and involve different stakeholders.
“what should an AI incident response program look like? Is it the same as a privacy incident response or a cyber incidents response program?”
Acker argued that balanced regulation and governance can speed up AI innovation.
“I really do believe that when there's the right balance of regulation and governance, it can help, um, actually speed up some of that innovation.”
Acker pointed to the EU AI Act and U.S. state-level rules as examples of legal limits on AI.
“Of course, you know, with the EU, there's the EU AI Act, so you have it at, you know, for AI specifically at those kind of levels.”
Acker said she had worked on AI governance for over six years.
“I've been really lucky to be able to work on this for over six years at this point.”
Acker had completed her sixth semester teaching in the master of legal studies program at American University's Washington College of Law.
“I just wrapped up my 6th semester teaching at American University for their Washington College of Law, um, their master of legal studies program.”