White House AI Bill of Rights Looks to Rein in ‘Unaccountable’ Algorithms

A lengthy statement of AI policy principles announced Tuesday(Opens in a new window) by the White House doesn’t actually say, “An AI may not injure a human being or, through inaction, allow a human being to come to harm,” but it might as well lead off with that sentence.

This “Blueprint for an AI Bill of Rights(Opens in a new window),” put forth by the executive branch’s Office of Science and Technology Policy after months of consultation, sticks to the harm-reduction priorities that science-fiction author Isaac Asimov famously laid out in his Three Laws of Robotics(Opens in a new window).

But where Asimov was trying to predict the future, this report comes after such AI-driven systems as facial recognition have become part of the basic workflow for such tasks as applying for a job, signing up for a car-share service, or flying to another country, often with little consideration of possible failure modes.

As the White House blog post about the blueprint says in a paragraph that could land in a future speech by President Biden: “This blueprint is for the older Americans(Opens in a new window) denied critical health benefits because of an algorithm change. The student(Opens in a new window) erroneously accused of cheating by AI-enabled video surveillance. The fathers(Opens in a new window) wrongfully arrested because of facial recognition technology. The Black Americans(Opens in a new window) blocked from a kidney transplant after an AI assumed they were at lesser risk for kidney disease. It is for everyone who interacts daily with these technologies—and every person whose life has been altered by an unaccountable algorithm.”

This lengthy document–the printable version(Opens in a new window) (PDF) runs 73 pages and exceeds 30,000 words, counting appendices and endnotes–sets out five policy planks for public and private actors, citing dozens of real-world examples of what can go wrong. What this report doesn’t say: Without action by Congress to pass meaningful privacy legislation, many of these suggestions will remain just that. 

Some civil-liberties groups quickly applauded the OSTP’s work. 

The Center for Democracy & Technology released a statement from its president and CEO Alexandra Reeve Givens saying “we commend the White House for considering the diverse ways in which discrimination can occur, for challenging inappropriate and irrelevant data uses, and for lifting up examples of practical steps that companies and agencies can take to reduce harm.”

The American Civil Liberties Union, meanwhile, provided one from ReNika Moore, director of its Racial Justice Program, that called for “a ‘bill of rights’ to protect us against the use of faulty and discriminatory artificial intelligence that infringes upon our core rights and freedoms.” 

The U.S. Chamber of Commerce, meanwhile, sent a comment from Jordan Crenshaw, vice president of the business group’s Chamber’s Technology Engagement Center, that commended the overall effort but warned against impeding US competitiveness: “For starters, the Blueprint’s definition of ‘automated systems’ is so broad that basic computing and technology could be subject to the same oversight or regulations as highly complex decision systems.” 

The blueprint is the latest in a series of policy directives from OSTP as the Biden administration–with some concurrence of Congress in the form of such legislation as the CHIPS and Science Act–has moved to take a more active role in setting science, technology and industrial priorities for private industry.