💻 computer science

Support for AI Development -- Automated Daily Measurement with Open Data and Code

This paper advocates for and demonstrates an open-source, automated system for daily public opinion nowcasting, using a web dashboard to track American support for AI development and arguing that such high-frequency, transparent methods are essential for accurately monitoring shifts in attitudes toward transformative technologies.

Original authors: Jason Jeffrey Jones

Published 2026-06-10

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Jason Jeffrey Jones

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to understand how a crowd feels about a new, rapidly changing technology. Traditionally, social scientists act like photographers: they take a single, high-quality picture of public opinion every two or four years. By the time the photo is developed and published, the crowd has already moved, and the picture is outdated.

This paper introduces a new approach: a live video feed.

The author, Jason Jeffrey Jones, built a free, open-source machine called the "Social Science Dashboard Inator" (SSDI). Instead of taking occasional snapshots, this system takes a daily "selfie" of public opinion. It automatically recruits a small group of people every single day, asks them a few questions, analyzes the answers, and posts the results on a public website instantly. No human has to touch the process after the initial setup; it runs on autopilot.

Here is a breakdown of what the paper found, using simple analogies:

1. The Experiment: Tracking AI Support

The author used this "live feed" machine to track how much Americans support the further development of Artificial Intelligence (AI).

The Setup: Every day, 11 random Americans were asked, "How much do you support further development of AI?" They answered on a scale from "Strongly Oppose" to "Strongly Support."
The Scale: The study ran from April 2024 to May 2026, gathering over 8,500 responses.
The Goal: To see if opinions were shifting in real-time, rather than waiting years to find out.

2. The Main Findings: The Rollercoaster Ride

The daily data revealed a story that yearly surveys would have missed:

The Trend is Down: At the start of the study, support was modestly positive. However, over time, support slowly but steadily decreased. It's like watching a balloon slowly lose air; it didn't pop, but it definitely shrank.
The Turning Point: The data showed a specific "flip" happened around April 2025. Before this date, support was actually rising slightly. After this date, it began a steady decline. This is like a car that was driving uphill suddenly hitting a hilltop and starting to roll backward.
Political Splits: Just like how people often disagree on climate change or taxes, the study found that support for AI became politically polarized.
- Initially, Democrats and Republicans had similar views.
- After the 2024 US election (where Republicans won), Republican support for AI actually rose above Democrat support. The author suggests that when people's preferred political party is in charge, they tend to like new technologies more.
Personality Traits Matter:
- Risk-Takers: People who are naturally willing to take risks (like jumping off a high dive) were much more likely to support AI. People who are cautious and risk-averse were less supportive.
- Trust: People who generally trust others were slightly more supportive of AI than those who are suspicious of people.
- Age and Gender: Men supported AI more than women. Interestingly, older people supported AI more than younger people, and this gap grew wider as time went on.

3. What People Actually Said (The "Why")

The survey also asked people to write a sentence or two explaining their answer. The author analyzed these notes and found:

Old Fears: People still worry about "Skynet" (robots taking over) and job loss.
New Fears: As AI grew, new worries appeared, specifically about energy consumption (water and electricity) and the environmental cost of running AI.
New Hopes: Some people mentioned AI as a "friend" or a way to combat loneliness, while others saw it as a tool to make their work easier.

4. Why This Method Matters

The author argues that this "daily dashboard" approach is better than traditional science for a few reasons:

Transparency: Everything is open. The code, the data, and the results are all public. It's like a glass house where anyone can see how the science is done.
Speed: If something big happens in the world (like a major AI breakthrough or a scandal), this system can show how public opinion shifts immediately, rather than waiting two years for a report.
Replication: Because the system is automated and open, other scientists can copy it exactly to check the work, or use the same data to ask different questions.

5. The Future Prediction

Based on the steady downward trend observed in the data, the author makes a simple prediction: Support for AI will continue to drop throughout 2026, potentially reaching neutral or negative levels by the end of the year.

The Bottom Line

This paper isn't just about AI; it's about a new way of doing science. The author built a machine that turns the slow, blurry process of understanding public opinion into a fast, clear, and open video stream. By doing this, they discovered that public support for AI is not a fixed statue, but a living, breathing thing that changes day by day, influenced by politics, personality, and the news cycle.

Technical Summary: Automated Daily Measurement with Open Data and Code

Problem Statement
Social science research often suffers from low-frequency data collection (e.g., biennial or quadrennial surveys) and delayed data release, which hinders the ability to track rapid shifts in public opinion regarding transformative technologies like Artificial Intelligence (AI). Traditional data availability creates a lag between the emergence of societal trends and the scientific understanding of them. Furthermore, the field faces a "replication/reproducibility crisis" and a "predictability crisis," where theoretical models struggle to predict human outcomes despite the explosion of available data. There is a lack of open, high-frequency, and transparent mechanisms to "nowcast" (predict the present and very near future) public sentiment in real-time.

Methodology: The Social Science Dashboard Inator (SSDI)
The paper introduces the Social Science Dashboard Inator (SSDI), an open-source, automated system designed to collect, distribute, and present nowcasted public opinion daily. The system operates on a continuous, autonomous loop:

Recruitment & Collection: The system recruits a small, random sample of respondents daily (11 American adults per day) via the Prolific Academic platform. A short survey (median completion time <2 minutes) is distributed via Qualtrics XM.
Automation & Processing: A series of open-source scripts (R and Python) triggered by cron jobs automate the entire workflow. This includes downloading responses, matching them to demographic data, anonymizing microdata, and updating a cumulative structured data file in a public repository.
Analysis & Visualization: The system automatically performs statistical analyses on the cumulative dataset and updates a public web dashboard to display results, hypothesis tests, and temporal trends.
Transparency: All materials, including survey items, response options, analysis code, and raw microdata, are hosted in a public GitHub repository, ensuring immediate reproducibility and open access.

Demonstration Project: AI Support Nowcasting
To validate the SSDI system, the author deployed a daily survey tracking American adults' support for the further development of AI. The focal item was a 7-point Likert scale question: "How much do you agree with the statement: I support further development of artificial intelligence?" (mapped to -3 to +3). The study ran from April 18, 2024, through May 31, 2026, accumulating 8,551 respondents. The survey also included measures for Generalized Trust, Risk Willingness, political party affiliation, age, and sex, along with optional open-ended text responses.

Key Results
Based on the daily data collected over 766 days, the study reports the following findings:

Temporal Trends: Overall support for AI development decreased over time. A segmented regression analysis identified a trend break on April 18, 2025. Prior to this date, support was increasing (approx. +0.41 points/year); afterward, it declined significantly (approx. -0.98 points/year).
Political Polarization: A divergence in support emerged between political parties. Beginning in December 2024, Republican support for AI development exceeded Democrat support, coinciding with the 2024 U.S. election results.
Psychological Correlates:
- Trust: Respondents with higher generalized trust reported higher AI support compared to those who were more cautious.
- Risk Willingness: There was a strong positive correlation between risk willingness and AI support ( $\hat{\beta} = 0.203$ ). Those more willing to take risks were significantly more supportive of AI development.
Demographics:
- Sex: Male respondents consistently reported higher support than female respondents.
- Age: Older participants reported higher support than younger participants. Crucially, the decline in support over time was steeper among younger respondents, indicating an increasing age-based cleavage.
Qualitative Shifts: Analysis of open-ended responses revealed emerging themes beyond the standard "exciting/useful/worrying/futuristic" categories. New concerns regarding utility consumption (water/electricity) and appreciation for AI companionship appeared. The frequency of mentions regarding environmental costs (water/electricity) increased significantly over time.

Significance and Claims
The paper positions the SSDI not merely as a data collection tool but as a new form of scientific communication. Its significance is framed around three main contributions:

High-Resolution Temporal Insight: The system provides the temporal resolution necessary to detect when and how public opinion shifts, capturing turning points (like the 2025 trend break) that low-frequency surveys would miss.
Enforced Transparency and Reproducibility: By automating the data pipeline and making all code and data public immediately, the system removes barriers to replication and forces adherence to open-science practices. It allows for "parallel dashboards" where fundamental disagreements on measurement can be tested simultaneously.
Principled Prediction: The continuous data stream creates a feedback loop for testing theories. It allows researchers to move from static descriptions to dynamic predictions, holding models accountable against real-time data rather than delayed, static datasets.

The author argues that this approach addresses the "predictability crisis" by providing a mechanism to test whether theories can accurately forecast the trajectory of daily time series. The paper concludes that while the current demonstration is modest in scale (self-funded, single-nation), the architecture is scalable and should be adopted more broadly to improve the efficiency and transparency of social science discovery.