Information
Artificial Intelligence and the Truth Problem
AI can now generate a convincing photo, voice, or news article of something that never happened, in seconds and at scale. A society that cannot agree on basic facts will struggle to do anything else together.
It used to take real effort to fake a photograph convincingly, and a faked one could usually be caught by someone who looked closely. That is no longer reliably true. Generative artificial intelligence can now produce a photorealistic image, a cloned voice, or a fluent news article describing an event that never occurred, cheaply enough that anyone with a laptop can do it and quickly enough that it can spread before anyone fact-checks it. The tools are neutral; what people do with them is not.
Americans worry about this from more than one direction. Some worry chiefly about deliberate deception: fabricated video of a candidate saying something they never said, synthetic voices used to scam grandparents out of their savings, entire websites of AI-generated propaganda posing as local news. Others worry chiefly about an overcorrection: that fear of AI-generated falsehood will be used to dismiss inconvenient but real evidence as fake, or to hand a small number of platforms and governments enormous power to decide what counts as true. Both worries have already happened, sometimes in the same news cycle.
Why this cuts across every other issue
Nearly every subject on this agenda depends on citizens sharing some baseline of fact before they argue about what to do. Elections, courts, local news, and public health all become harder to reason about together if a growing share of what people see and hear cannot be trusted at first glance, and there is no unpartisan winner in a world where nobody believes anything.
A shared civic stake, not a fight over one platform
This is not a question of which party's misinformation is worse; fabricated content and genuine confusion have already touched every side of American politics, and every side has both spread it and been victimized by it.
What we ask of members
Common Ground does not endorse a specific AI regulation or content-moderation regime. We ask members to slow down before sharing anything that seems too perfectly damning to be true, to support tools and norms that help people verify what they see, and to resist using "it might be AI" as a reflexive excuse to dismiss evidence that is simply inconvenient.
Civic principles
How we approach the topic, before any policy.
Verify before you amplify
The speed of sharing has outrun the speed of checking. We treat a brief pause to verify a striking claim or image as a basic civic habit, not paranoia, especially when the content confirms what we already wanted to believe.
Skepticism should not become a universal escape hatch
Real evidence exists alongside fabricated content, and treating every inconvenient fact as possibly synthetic is its own form of dishonesty. We hold both a healthy skepticism and a willingness to accept evidence that survives scrutiny.
Power to define truth is dangerous no matter who holds it
Whoever controls the tools for labeling content as true or false gains real power over public debate. We favor transparent, contestable processes for that judgment over trusting any single company or government to get it right unchecked.
Where agreement stands
- Broad agreement
AI-generated impersonation of real people should carry legal consequences
Broad agreement exists across the political spectrum that using AI to fabricate a real person's voice or likeness for fraud, harassment, or deceptive political advertising should be illegal, regardless of who is depicted or who benefits.
- Broad agreement
Platforms should label AI-generated content
Most Americans support requiring clear labels on images, audio, and video that are substantially AI-generated or altered, particularly in political advertising. Support for the goal is strong even where the technical means of achieving reliable labeling remain unresolved.
- Emerging agreement
Media literacy and verification tools deserve public investment
Interest in funding public education about how to evaluate digital content, along with tools that let ordinary users check an image or clip's origin, has grown among educators, technologists, and lawmakers across the spectrum. Which institutions should build and maintain these tools is still being worked out.
- Contested
How aggressively AI development itself should be regulated
Some Americans believe the technology is advancing faster than society's ability to manage its risks and want binding safety rules and oversight before harms compound. Others believe heavy-handed regulation will hand the technology's future to a handful of already-dominant companies and to other countries with fewer scruples, while doing little to stop bad actors who ignore rules anyway. Both sides want the technology's benefits without its worst harms; they disagree about whether caution or speed better serves that goal.
- Emerging agreement
News organizations should disclose when AI plays a substantial role in producing a story
As newsrooms adopt AI tools to draft, summarize, or translate stories, there is growing expectation across the political spectrum that readers should be told when a substantial part of what they are reading was AI-generated or AI-assisted rather than written and verified entirely by a reporter. Outlets accused of quietly publishing error-filled AI content have drawn criticism from readers and media critics of every persuasion. Exactly what counts as substantial AI involvement, and how prominently it should be disclosed, is still being worked out newsroom by newsroom.
- Contested
Whether AI chatbots that answer questions as fact should be held to publisher-like accuracy standards
Some Americans believe that when an AI assistant states something confidently as fact, the company that built it bears real responsibility for that answer being accurate, much as a publisher bears responsibility for what it prints. Others believe treating AI companies as publishers for every generated response is technically unworkable given the scale and variability of what these systems produce, and would either stifle a useful technology or push companies to make their tools far more evasive and hedged than users want. Both sides want people to be able to trust what they are told; they disagree about where responsibility for AI errors should sit.
Open questions
We state the tension honestly and do not pretend to resolve it.
Can reliable technical methods to detect AI-generated content keep pace with the tools that create it?
Detection and generation are in a continuous arms race, and some researchers doubt detection can ever fully catch up. If it cannot, the long-term answer may have to rely less on catching fakes and more on verifying authentic content at its source.
Who should decide what counts as harmful misinformation, and what happens when that judgment is wrong?
Governments, platforms, and independent fact-checkers have all been accused of getting this call wrong in ways that happened to favor one political side. Every proposed arbiter carries its own risk of error or bias, and errors in this domain can silence true speech as easily as false speech.
Does the danger lie mainly in believing false things, or in a growing unwillingness to believe anything at all?
Some researchers argue the bigger long-term risk is not that people will be fooled by fakes but that the mere possibility of fakery will let people dismiss real evidence they find inconvenient. If that is the deeper problem, the solutions look quite different from simply getting better at spotting fakes.
Should a small number of companies whose AI systems answer questions for hundreds of millions of people have that much influence over what a society believes is true?
As more people ask an AI assistant a question instead of searching for sources themselves, the judgment calls those systems make about contested topics reach an enormous audience by default rather than by any public decision. Whether that concentration of influence is meaningfully different from the influence past information gatekeepers held, and what if anything should be done about it, is unresolved.
Does it matter whether AI-generated disinformation originates from a foreign government or a domestic actor, and should the response differ?
Foreign influence operations using AI-generated content raise national security concerns that domestic misinformation does not, and different legal tools are available to address each. But audiences encountering a fabricated video rarely know its origin at the moment they see it, and some worry that focusing on foreign sources lets domestic actors doing the same thing escape the scrutiny they deserve.