Dive Brief:

• The White House on Tuesday unveiled a draft memorandum detailing 10 principles to guide federal regulation of artificial intelligence (AI) technologies and applications.
• The principles prioritize public trust in AI; public participation; scientific integrity and information quality; risk assessment and management; benefits and costs; flexibility; fairness and non-discrimination; disclosure and transparency; safety and security; and interagency coordination.
• During a keynote discussion at CES on Wednesday, U.S. Chief Technology Officer Michael Kratsios said the memorandum is "a very, very big step for both the U.S. and the world." While the private sector has made strides in detailing its own AI principles, these guidelines will ensure the federal government is "looking out for and protecting Americans as these technologies are developed and deployed around the country," he said. 

Dive Insight:

The Trump administration introduced AI guidance in 2019 with Executive Order 13859, or the American AI Initiative, a national strategy resting on five main pillars: promoting a sustained AI R&D investment; increasing federal AI resources; removing barriers to AI innovation; fostering workplace opportunity for AI jobs, with AI-focused education and training opportunities; and promoting international engagement.

The new memorandum seeks to expand the federal government's oversight of AI-powered technologies and to ensure that all federal government agencies — from the Food and Drug Administration (FDA) to the U.S. Department of Transportation (USDOT) — have the same information and guidance in how to regulate the private sector.

It recognizes some huge barriers to AI development, however, including cost-benefits and algorithmic biases. In regard to costs, Kratsios said flexibility will be key in finalizing guiding principles, to ensure regulatory frameworks can adapt with the ongoing tech revolution.

“If we start putting in regulations today that aren’t applicable years from now, we’re going to be stifling innovation in the end," he said.

Biases many not be as simple to combat. A recent study from the National Institute of Standards and Technology (NIST) found top facial recognition systems misidentify people of color at far higher rates than white people. A clear solution to this problem requires more diverse data training and engineering teams to develop AI-driven algorithms. However, there are limited requirements to ensure such diversity is imposed in AI engineering.

The draft memorandum will be open for public comment soon, which Kratsios said is "critical" to establishing an AI regulatory framework that is favorable to all stakeholders.

“Public interaction, public engagement is key," he said. "We don’t have all of the answers in Washington, never have, and we can’t begin to just start blindly putting out new rules on AI without engagement."