DENVER — Early efforts to regulate artificial intelligence programs that play a hidden role in the hiring, housing and medical decisions of millions of Americans are facing pressure from all sides and floundering in statehouses across the country.
Only one of seven bills aimed at preventing AI's tendency to discriminate in making subsequent decisions — including who gets hired, money for a home or medical care — has passed. Colorado Governor Jared Polis hesitantly signed the bill into law on Friday.
The Colorado bill and those that failed in Washington, Connecticut and elsewhere faced battles on many fronts, including between civil rights groups and the tech industry, and lawmakers were wary of getting into a technology that few understand, and governors worried about being the odd ones out. and fearsome AI startups.
Polis signed the Colorado bill “with reservations,” saying in a statement that he was wary of regulations that would hinder AI innovation. The bill has a term of two years and can be amended before it becomes law.
“I encourage (lawmakers) to significantly improve this before it goes into effect,” Polis wrote.
Colorado's proposal, along with six sister bills, is complex, but will broadly require companies to assess the risk of discrimination from their AI and inform customers when AI was used to make a resulting decision for them.
The bills are separate from the more than 400 AI-related bills that have been debated this year. Most are focused on bits of AI, such as using deepfakes in elections or to create pornography.
The seven bills are more ambitious, apply to major industries and target discrimination, one of technology's most perverse and complex problems.
“We actually have no insight into the algorithms that are being used, whether they work or not, or whether we are being discriminated against,” said Rumman Chowdhury, AI envoy for the US Department of State, who previously headed the AI Twitter ethics team. .
Although anti-discrimination laws are already on the books, those who study AI discrimination say it's a different beast, one that the US is already behind in regulating.
“The computers are making biased decisions on a massive scale,” said Christine Webber, a civil rights attorney who has worked on class-action discrimination lawsuits, including against Boeing and Tyson Foods. Now Webber is nearing final approval of one of the nation's first settlements in an AI discrimination class action.
“I shouldn't say that the old systems were completely free of bias,” Webber said. But “everyone could only view a limited number of resumes per day. So you can only make so many biased decisions in one day, and the computer can do that quickly for large numbers of people.”
When you apply for a job, an apartment or a home loan, there is a good chance that AI will review your application: forward it, assign it a score or filter it out. According to the Equal Employment Opportunity Commission, it is estimated that as many as 83% of employers use algorithms to assist with hiring.
AI itself does not know what to look for when applying for a job, so it teaches based on previous CVs. The historical data used to train algorithms can smuggle in biases.
For example, Amazon was working on a hiring algorithm that was trained on old resumes: largely male applicants. When reviewing new applicants, it downgraded resumes containing the word “women” or mentioning women's colleges because they were not represented in the historical data – the resumes – it had learned from. The project was scuttled.
Webber's class action lawsuit alleges that an AI system that scores rental applications disproportionately assigned lower scores to black or Hispanic applicants. A study found that an AI system built to assess medical needs was passing over Black patients for specialty care.
Studies and lawsuits have allowed a look under the hood of AI systems, but most algorithms remain obscured. Americans are largely unaware that these tools are being used, according to Pew Research polls. Companies generally do not have to explicitly disclose that an AI has been used.
“Just pulling back the curtain so we can see who is actually doing the assessment and what tool is being used is a huge first step,” Webber said. “Existing laws don't work if we can't get at least some basic information.”
That's what Colorado's bill, along with another remaining bill in California, seeks to change. The bills, including a flagship proposal in Connecticut that was defeated amid opposition from the governor, are largely similar.
Colorado's bill will require companies that use AI to help make consistent decisions for Americans to annually review their AI for potential bias; implement a monitoring program within the company; tell the attorney general if discrimination has been found; and informing customers when an AI was used to make a decision for them, including the ability to appeal.
Unions and academics fear that the reliance on companies to police themselves means it will be difficult to proactively tackle discrimination in an AI system before it causes damage. Companies fear that forced transparency could reveal trade secrets, including in potential lawsuits, in this hyper-competitive new field.
AI companies also pushed for, and generally received, a provision that would allow only the attorney general, not citizens, to file lawsuits under the new law. The details of enforcement are left to the attorney general.
While larger AI companies are more or less supportive of these proposals, a group of smaller Colorado AI companies said the requirements may be manageable by giant AI companies, but not by early-stage startups.
“We are in a brand new era of primordial soup,” says Logan Cerkovnik, founder of Thumper.ai, referring to the field of AI. “Having overly restrictive legislation that forces us to define and limits our use of technology as it is forming will only harm innovation.”
Everyone, like many AI companies, agreed that what is formally called “algorithmic discrimination” is critical to address. But they said the bill as written does not meet that goal. Instead, they proposed strengthening existing anti-discrimination laws.
Chowdhury worries that lawsuits are too expensive and time-consuming to be an effective enforcement tool, and that laws should instead go beyond what even Colorado is proposing. Instead, Chowdhury and academics have proposed an accredited, independent organization that can explicitly test for potential biases in an AI algorithm.
“You can understand and deal with a single person who is discriminatory or prejudiced,” Chowdhury said. “What do we do if it is embedded in the entire institution?”
___
Bedayn is a staff member for the Associated Press/Report for America Statehouse News Initiative. Report for America is a nonprofit national service program that places journalists in local newsrooms to report on undercovered issues.
