ODNI issues guiding principles for ethical artificial intelligence
The Office of the Director for National Intelligence issued guidance Thursday for the intelligence community‘s ethical development and use of artificial intelligence.
The two documents — new principles and a supporting framework — aim to set a foundation for how and when members of the IC should use, develop and procure AI applications.
Both strategic documents are meant to ensure “that our use of this technology always complies with all limits on the authorities of the American people have granted to our agencies to conduct our national security mission,” Ben Huebner, head of ODNI’s Office of Civil Liberties, Privacy, and Transparency, told reporters Thursday morning.
Huebner emphasized the extra sensitivity the IC has for making sure its AI is ethical, secure, explainable and consistent with the missions of the U.S.’s intelligence agencies.
“No one really has a better business case than the intelligence community for getting this right,” he said. “No one has a stronger case than the IC that our uses of AI have to be secure and resilient to adversarial attack. No one has a stronger case than the IC that the AI needs to produce results that our policymakers outside the IC, our customers for our intelligence, can interpret and understand and then use their human judgment to act upon.”
The principles are simply a set of basic ideas with brief explanations. None of them is unexpected or ventures into uncharted territory in the development of AI. They are: Respect the law and act with integrity; be transparent and accountable; be objective and equitable; rely on human-centered development and use; ensure the tech is secure and resilient; and be informed by science and technology.
ODNI developed its six principles to be consistent with those of the Department of Defense and others found in the private sector.
“We want it to be clear that the ethical principles that govern the use of artificial intelligence writ large — to act lawfully and in a manner that respects human dignity, to ensure an appropriate role for human judgment — those certainly govern us here in the intelligence community as well,” Huebner said.
Where the principles set the starting point for any conversation to be had around AI in the intelligence community, the supporting framework “really puts meat on the bones of how the intelligence community implements AI consistent with those principles,” Huebner explained.
“The framework that we developed is the result of data scientists, attorneys, privacy and civil liberties officer and other key stakeholders from throughout the intelligence community who all came together and worked collaboratively to find ways to fully incorporate the principles into the design and the use of artificial intelligence,” he said.
The framework boils down to a set of questions in 10 areas that span the entire lifecycle of AI. It’s not meant to be a checklist or a total solution in every scenario. Rather, it’s a tool to provide a “consistent approach to reasoned judgment,” Huebner said.
‘Strategic imperative’ for ODNI
Technologies like AI and machine learning hold the potential to revolutionize American intelligence gathering and are critical particularly as the world becomes enriched with new sources and greater volumes of data, said Dean Souleles, ODNI chief technology adviser.
“Our objective in developing AI and machine learning and related tools for intelligence is based on a strategic imperative. And that is that the mission of the intelligence community is to see over the horizon in enough time to impact the outcome and to understand foreign threats,” said Souleles, who also founded ODNI’s Augmenting Intelligence through Machines Innovation Hub. “And in order to do that we are awash with data that the world produces that is of interest to foreign intelligence, and we simply cannot hire enough analysts to deal with the vast amount of data that the world produces. And the hypothesis is that inside the data of the world are the answers to the questions that we must have.”
That said, AI is a different sort of beast than what the IC is used to dealing with — one that requires a guiding set of principles and framework, Souleles said. “We understand that these are different technologies than our traditional technologies. These are not fixed systems where a given input always produces a given output — they produce probabilistic results, they are based on very complex mathematical models called neural networks.”
He continued: “Our job It is to be able to produce intelligence for the president and the national leadership that we have confidence in, that we can state with some assurance that we understand why we made the conclusion that we made and we understand [the potential for error]. And AI doesn’t give you a yes or no.”
The documents are open to public feedback and will be updated over time, the ODNI officials said. The tough part of that is that much of what the IC will do with AI will involve classified data. But, ODNI will “talk in general terms about some of the things that we are doing, and use that to sort of inform the larger debate,” Huebner said. “We definitely want to apply this framework to some very highly classified work. But where we can we definitely want to be transparent.”