The Senate voted unanimously Thursday to approve legislation designed to scrutinize the use of algorithms and artificial intelligence by Connecticut’s state government to ensure automated systems are not permitted to make discriminatory decisions.
“The goal is looking at the safety of artificial intelligence and more transparency,” Sen. James Maroney, D-Milford, said during a brief afternoon debate on the Senate floor. “There are lots of questions. Are we using it? Are we not using it? And I always feel that sunlight disinfects so this will make it clear where we are using and that we’re testing it before it is used.”
The bill, which will now head to the House for consideration, requires the Administrative Services Department to publish a list of agencies using AI and an ongoing assessment of how that technology is used by state government.
Proponents hope the requirements shed light on where automated systems and algorithms are being used by the state. A report published last year by the Media Freedom & Information Access Clinic found that the state employed algorithms at DAS, the Department of Children and Families, and the Department of Education, where an algorithm is used to assign students to schools in Hartford.
Last month, advocates and members of the Connecticut Advisory Committee To U.S. Commission on Civil Rights called for changes to the state Freedom of Information Act to require agencies to provide information on their use of algorithms.
Although those changes are not included in the proposal passed Thursday, the bill does require the state to develop policies to ensure that no discrimination results from the use of automated systems and requires state agencies to conduct impact assessments before implementing the systems. The bill also prohibits the state from employing untested AI after next February.
Maroney, co-chairman of the legislature’s General Law Committee, said the language seeks to provide guardrails preventing automated systems from adopting biases based on the behavior of its users.
“It learns from the way the world is going now,” Maroney said. “So it picks up and it amplifies our existing biases. We want to make sure that we are not perpetuating those biases and making them worse and that we have tested anything before we put it in effect.”
If adopted, the proposal will also create a working group to draft recommendations for a “Connecticut AI Bill of Rights” as well as recommendations for future legislation on the use of AI by private companies.
The bill passed on the Senate’s consent calendar, meaning no senator sought the opportunity to cast a vote in opposition. During the debate, Sen. Paul Cicarella, R-North Haven, said the legislation would provide residents with insight on how technology is used by the state.
“It’s a very complicated process when you’re really understanding what this artificial intelligence can do,” Cicarella said. “Ultimately, this is going to have transparency for everyone to understand it, basically, in plain English and not in ones and zeros or in code.”
Sen. Tony Hwang, R-Fairfield, said the legislature faced a complex challenge in attempting to regulate an emerging technology.
“We’re trying to legislate, manage, anticipate in a fast-changing and dynamic marketplace,” Hwang said. “We are doing our best through due diligence.”
Maroney stressed that the intent of the bill was not to prohibit the use of artificial intelligence and algorithms, just to prevent it from creating any disparate impacts.
“There is great potential [in AI], there’s no doubt,” he said. “We know that there’s so many possible strengths to AI but we also know that there are perils and all we are saying is ‘you need to look at it first.’”