Poll Shows Voters Oppose Open Sourcing AI Models, Support Regulatory Representation on Boards, and Say AI Risks Outweigh Benefits
NEW YORK, SEPT 13 – A new poll from the Artificial Intelligence Policy Institute (AIPI) shows a clear consensus in favor of public policy interventions that put in place safeguards on AI. The survey—commissioned by AIPI and conducted by YouGov—demonstrates the general public is warm to regulatory measures and skeptical of wider dissemination of powerful AI models.
Among the findings:
- Voters oppose open sourcing powerful AI models by 2:1 margin; 47% oppose it compared to just 23% who support it
- 71% of voters say the potential risks of AI are greater than its potential benefits and that we should err on the side of caution while just 29% believe too much caution will reduce the chances we reap AI’s benefits
- 66% of voters believe AI companies should be required to have regulatory representation on their board
- 68% of voters believe that companies advocating for strict safety controls on AI are trying to do what’s best for for the general public while just 31% believe they are doing so due to their own self-interest
- Almost three times as many voters believe AI models have liberal political bias than those who believe they have a conservative bias
“When it comes to AI, it’s crystal clear that the American public favors leaning toward caution over chasing potential benefits created by advancement and are wary of the proliferation of powerful AI models to anyone who can get their hands on them,” said Daniel Colson, Executive Director of the Artificial Intelligence Policy Institute. “Whether it’s regulatory representation on boards, implementing strict safety controls or open sourcing powerful AI models, voters express sympathy with caution and intervention as they are unconvinced by arguments for noninterference in the name of technological advancement.”
“Policymakers and tech companies should pay close attention to the decisive signals voters are sending by acting with prudence and deliberation rather than reckless abandon,” Colson added.
This poll is the second by the Artificial Intelligence Policy Institute, a think tank dedicated to providing public opinion research and policy expertise on AI regulation. AIPI will continue to conduct frequent surveys to demonstrate where Americans stand on Artificial Intelligence issues, and provide analysis on existing proposals coming out of Washington, D.C. With AI evolving every day and Americans wary about its rapid advancement, it is critical for policymakers to meet this moment with urgency and consideration of their constituents.
AIPI was founded to serve as a go-to resource for gauging American public opinion as it relates to AI and provide policymakers and the media with a fuller understanding of the risks posed by artificial intelligence—and solutions for reining in the worst-case scenarios. By measuring public opinion, AIPI will show policymakers that being active in AI policy is not only the right path forward on the merits, but also that action is necessary to better represent their constituents.
With years of experience in the field, AIPI has a deep understanding of AI’s dangers and opportunities. The threat to humankind is alarming, and policymakers must understand not only the economic consequences but also the potential for the next phases of AI development to bring the risk of catastrophic events. AIPI proposes a way forward where lawmakers and influencers in Washington can be informed by a common, nonpartisan set of resources so they can speak collaboratively, productively, and urgently about the need to control AI development and regulate it to mitigate the most dire risks. AIPI will also advocate for a broad range of policies that impose guardrails and oversight on AI and the development of superintelligence.
About the Poll
From September 2 to 6, 2023, YouGov conducted a survey of 1,118 voters nationally using online respondents. The survey was conducted in English. The margin of error is ±3.2 percentage points.
Note on Open Sourcing
To be a bit more technical, in the context of AI, open sourcing a model typically refers to open sourcing the learned “weights” of the model, which others can then use either to run the model on their local computer, or to further train (i.e., “fine tune”) the model for specific tasks or behavior. When phrasing this question, we considered that most poll respondents probably would not be familiar with the concept of model “weights,” so we chose to simplify the question slightly with the goal of achieving a question that more poll respondents would understand, at a slight cost of technical precision.