Jan Hochadel

Senator Hochadel Supports Passage of Bill That Regulates Artificial Intelligence in Connecticut

Today, state Senator Jan Hochadel (D-Meriden), supported State Senate passage of a bill that will work to regulate Artificial Intelligence (AI) in Connecticut. The bill will set several requirements for state agencies’ development and use of automated systems for “critical decisions.”

Senate Bill 1103, ‘An Act Concerning Artificial Intelligence, Automated Decision-Making, And Personal Data Privacy.” Today it passed the Senate with a unanimous vote and now heads to the House of Representatives.

“I am hopeful this legislation will address and drastically reduce pre-determined online bias,” said Sen. Hochadel. “I know when people apply for a job, artificial intelligence algorithms will completely toss aside a resume submitted online just because of a few key words. This will eliminate discrimination for several folks who have encountered unfair treatment while applying for a job.”

Senate Bill 1103 will require assessments ahead of the implementation of AI in specific high-risk incidences. It will create policies and procedures to govern the state’s use of AI, and it will create a task force to work toward creating a Connecticut AT Bill of Rights.

SB 1103 will: establish an Office of Artificial Intelligence; exempt air carriers from certain provisions concerning data privacy; make sure a controller cannot process a consumer’s personal data for purposes of targeted advertising, or sell the consumer’s personal data without the consumer’s consent; establish a task force to study artificial intelligence and develop an artificial intelligence bill of rights; require the Office of Policy and Management to designate an AI officer to develop and adopt procedures for using automated systems; require the Department of Administrative Services (DAS) commissioner to designate an AI implementation officer.

In written testimony submitted by the Commission on Human Rights and Opportunities, they stated one of the biggest problems they have seen is algorithms are trained on biased data. Police departments across the nation – including in Connecticut – use predictive policing algorithms to determine where to deploy officers to deter and investigate crime. They do so by looking at crime rates in the past. Many communities of color have been overpoliced in the past, however, due to bias such as racial profiling. The computer will therefore tell a department to send more police to an overpoliced area in order to address issues in what it identifies as a high crime area. As more police officers go to that area, the officers are more likely to arrest people there, while the same crimes in a different community go unaddressed.

Hiring algorithms have been shown to discriminate based on age. Some algorithms have given higher interest rates for loans based on race, and many government used algorithms in other states, ranging from provision of SNAP benefits to deciding when to investigate reported incidents of child abuse, have been shown to discriminate based on income.

The online world has the increased capacity to store data online that can relinquish unwanted results. AI can produce ethical challenges including lack of transparency and un-neutralized decisions. Choices made through AI can be susceptible to inaccuracies, discriminatory outcomes, and inserted bias.

Digital discrimination is the next frontier in the fight for civil rights. When most people think about discrimination, they think of intentional decisions made by humans against others based on their protected class. The danger of digital discrimination is that it is too often invisible both to the victim and the perpetrator. A few examples of how digital discrimination works will show why bills like HB 1103 are so important.