Jones-Skiena Public Opinion of Artificial Intelligence Dashboard

2021-04-05 Update

Dr. Jason Jeffrey Jones -- Dr. Jason Jeffrey Jones -- @jasonjones_jjj
Dr. Steven Sol Skiena -- Dr. Steven Sol Skiena -- @stevenskiena

Support for the Development of Artificial Intelligence

We deliver the following item every week to a random, representative sample of 100 Americans:
How much do you oppose or support the development of Artificial Intelligence?

The mean response per week and a linear trend are presented below.

Interpretation

On average, the American public supports the development of Artificial Intelligence. In the data from April 1, 2021, the estimated mean response is 4.86, with a 95% confidence interval of 4.46-5.26. On our 7-point scale (labeled "Strongly oppose" at one end and "Strongly support" at the other) a value of 4 indicates indifference, any value below that opposition and any point above support.

In 28 weeks of running this same survey item, we have never observed a mean estimate below the midpoint of the scale. However, it is useful to observe the raw frequency of responses as a set of histograms. One can see that the modal (most frequent) response is typically 4, the midpoint of the scale, which we interpret as indifference. Americans are generally neutral on this question, but among those who are not, there are more on the Support end of the scale than the Oppose end. This was a rare but not unprecedented week in which the most popular response was 5/7 rather than 4.

Currently, the data point toward stable support. A linear fit of mean estimates over weeks returns a slope not statistically significantly different from zero. Volatility in the weekly estimates is higher in the recent data.

Interested in how support for AI compares to other technologies? Interested in sub-population comparisons on gender and age? Check out the results of our Contextualizing Support for AI Development Survey.

Trust in Artificial Intelligence

NEW DATA! Updated 2021-04-05.

We deliver items of the following form every month to a random, representative sample of 100 Americans:
How much trust do you have in <prompt-item> to do the right thing?

The mean response per week and a linear trend are presented below.

Interpretation

On average, the American public trusts "artificial intelligence algorithms" to do the right thing just a little less than they trust the average American. Americans trust artificial intelligence algorithms more than Congress or the President, but not as much as their best friend. Pooling the data and weighting to match demographics, we may estimate the following 95% confidence intervals.

If you were thinking that these trust survey items would have interesting distributions, you'd be right.

During the period of observation, the magnitude and ordering of trust in these targets has remained stable. Trendlines are shown to aid the eye, but no trendline has a slope significantly different from zero. Trust in Congress is creeping closer, however.

Last month we wondered about the convergence of the trendlines for "the Average American" and "artificial intelligence algorithms." The trend did not continue in the most recent data.

One-Time Special Topic Surveys

Updated 2021-03-29.

Once per month, we run a one-time survey on a special topic. Usually, this is a 10-question survey with 100 respondents. You can request an item or topic by contacting Dr. Jones.

Contextualizing Support for AI Development

AI Fear Scenarios Survey

Confidence in Groups using Artificial Intelligence

Updated 2021-03-22.

We deliver items of the following form every month to a random, representative sample of 100 Americans:
How much confidence would you say you have in <prompt-items> that use Artificial Intelligence systems?

The mean response per month and a linear trend are presented below.

Interpretation

On average, the American public has middling confidence in groups that use Artificial Intelligence. There is plenty of room on the scale for respondents to indicate they have less confidence or more, but the responses point to sustained intermediate levels of confidence, or perhaps indifference.

In our early data (Sep-Nov 2020), scientists that develop new Artificial Intelligence systems appeared to consistently elicit more confidence than other items. The series also indicated it might be increasing (weak positive slope). However, with the most recent three rounds of data, confidence in AI scientists has slumped back toward the pack. No item appears to be steadily gaining or losing confidence. Trendlines are shown to aid the eye, but none has a slope significantly different from zero. (The downward trend for confidence in scientists is closest to statistical significance at p = 0.16.)

The figure below contrasts the means for each prompt item, collapsing across all dates. One can confidently conclude that overall, Americans place highest confidence in scientists to develop new artificial intelligence systems, followed by medical doctors, companies and government agencies to use these systems.

Familiarity with Artificial Intelligence

Updated 2021-03-15.

We deliver items of the following form every month to a random, representative sample of 100 Americans:
How familiar are you with the following computer-related item? <prompt-item>

The mean response per month and a linear trend are presented below.

Interpretation

Are Americans really still on the learning curve for email? It is hard to believe, but that is what the data say. The positive slope of the regression line is statistically significant for Email, p = 0.001. That is the strongest temporal trend present in the results.

Self-report understanding of "Artificial intelligence" is also increasing, p = 0.014. Understanding of "Machine learning" follows a parallel slope, but the variability in the data prevents the coefficient estimate from reaching statistical significance.

On average, the American public self-reports less understanding of Artificial Intellgience terms as compared to other computer terms. That split is apparent by the separation of the datapoints and trendlines above. It can be seen more clearly in the following visualization which pools the data for all survey rounds and estimates 95% confidence intervals for each item.

Americans report very low understanding of 'GPT-3' - an item we added in the December round. Despite the volume and intensity of recent discussion surrounding GPT-3 within AI circles, the broader public is not comfortable expressing any understanding of this item.

These surveys are conducted by Dr. Jason Jeffrey Jones and Dr. Steven Sol Skiena. All raw data is available at this Open Science Foundation repository. Please include proper attribution any time you make use of our work. Read more about the Jones-Skiena Public Opinion of AI Dashboard.