Voters Want AI Companies to Prevent Their Products From Being Discriminatory
By Tenneth Fairclough II
While artificial intelligence (AI) prevails in sectors like health care and finance, there are growing concerns over these systems using incomplete or unrepresentative datasets that reinforce certain biases against particular people and groups.
New polling from Data for Progress and Accountable Tech examined voters’ attitudes toward AI models being discriminatory and whether AI companies should prevent this from occurring.
First, voters were given a short description of how AI models work and then were asked how concerned they are with these models using incomplete data to make discriminatory predictions. Seventy-seven percent of voters report being “very concerned” or “somewhat concerned” about discriminatory predictions by AI, including 77% of both Democrats and Independents and 76% of Republicans.
Additionally, Data for Progress asked voters whether they believe AI companies that develop these tools should or should not be required to clearly show their products are nondiscriminatory toward certain people and groups. Seventy-eight percent of voters believe AI companies should be required to show their products are nondiscriminatory. Voters show strong consensus across party lines: 83% of Democrats, 76% of Independents, and 74% of Republicans believe AI companies should be required to clearly show that their products are nondiscriminatory toward certain people and groups.
These findings highlight that voters are concerned about AI models using incomplete data to make discriminatory predictions, and they want AI companies to develop methods that clearly show their products are nondiscriminatory toward certain people and groups.
Tenneth Fairclough II (@tenten_wins) is a polling analyst at Data for Progress.
Survey Methodology
From March 6 to 8, 2024, Data for Progress conducted a survey of 1,227 U.S. likely voters nationally using web panel respondents. The sample was weighted to be representative of likely voters by age, gender, education, race, geography, and voting history. The survey was conducted in English. The margin of error is ±3 percentage points.