How Should AI Tools Be Used?
Americans Support Some Uses But Oppose Others

The XandY Poll | U.S. National Adult Population | Nov. 2023

Abel Gustafson, Matthew Goldberg, & Carly wolfer
December 2023

A recent wave of technological advances in AI tools (e.g., ChatGPT, DALL-E) have sparked heated debates about how their capabilities should and should not be used. Some uses of AI tools may seem harmless—such as assisting with writing emails or searching for information. But other uses of AI tools may cross a line that many people are uncomfortable with—such as if they were used to fire an employee or write the news. The uses that people support or oppose can help guide the decisions of technology companies, policymakers, and more. Here, we summarize findings from our recent national survey that show the degree to which Americans support or oppose several specific uses of AI tools.

People Oppose AI Tools Replacing Important Professional Roles

As shown in the figure below, the strongest opposition is when AI tools replace a role typically performed by the judgments of human experts. For example, there is more opposition than support for the idea of using AI to help write the news, evaluate employee’s job performance, and review job applications. People also strongly oppose patients using AI to get medical advice, which is another example of using AI tools as a substitute for professional human judgment.

However, it’s not as simple as concluding that people just don’t want AI to assist important professionals. Our data show a counter-example, where a greater number of Americans (44%) support doctors using AI to help diagnose patients and prescribe treatments than oppose it (31%). 

Also, many Americans (48%) support law enforcement using AI tools to identify suspects, while only 28% oppose it—although this may differ greatly depending on if people are aware of the risks of this use case in terms of privacy and bias.

The AI uses cases with the strongest support are those that involve helping with writing personal communication and artistic uses (musicians, artists). This is already commonly represented in the current abilities of AI tools and are among the most common uses (as we report here).

So What

Companies developing AI tools can use these insights as a strategic tool for assessing consumer demand and market opportunities. Similarly, policymakers can use these findings when developing regulations or oversight that reflects the desires of their constituents. 

However, much more research is needed. These findings describe people’s preferences but they do not answer questions about why people support or oppose these different uses. To design valuable technologies and effective democratic policy, it is important to dive deeper into the reasons and mechanisms responsible for these preferences.

About The XandY Poll

Survey Methodology

The design, data collection, analysis, and reporting of this national survey were performed by XandY, an independent research and strategy firm. Exhaustive details of the scientific methodology of The XandY Poll can be found here. The following briefly summarizes key points of interest.

The survey responses were fielded from November 18 – November 21, 2023 using online recruitment methods to sample adult residents of the United States (N = 1,527). This survey used a nested quota sampling strategy to match U.S. Census proportions of age, income, race and ethnicity, gender, and political party affiliation. To further ensure the insights reported from these data closely resemble the U.S. population, the sample was weighted to match U.S. Census benchmarks.

Percentage points are rounded to the nearest whole number. When sums of two proportions are reported in text (e.g., “63% of U.S. adults say they either “strongly” or “somewhat” support…") we round the total value after summing. Sometimes, this creates an apparent error. For example, 41.4% + 20.4% = 61.8% which rounds to 62%.  But in the figure, the values are individually rounded (41% and 20%) so it might appear that 41% + 20% = 62%.

Margin of Error

Proportion statistics regarding the full national sample have an average margin of error of +/- 2 percentage points at the 95% confidence level. The margin of error in subgroups is determined by the subgroup size.

Citation

This paper and the insights it reports may be cited as:

Goldberg, M. H., Gustafson, A., & Wolfer, C. (2023). How Should AI Tools Be Used? Americans support some uses but oppose others. XandY. New Haven, CT. Retrieved from: www.xandyanalytics.com/how-americans-want-ai-tools-to-be-used