I suppose you missed the American right over the last ten years turning aggressively towards extremism and anti-democratic values, largely meditated by propaganda distributed over the Internet?
The only way to avoid this is for the AI to be opinionated, which would obviously be very problematic in itself.
An AI should be opinionated, at least towards the core values expressed in Western constitutions, because we all (or our ancestors) democratically decided upon them: the equality of all humans, equality before the law, the rule of law before violence, the need to learn as a species from at least the largest horrors of the past (WW1/2, the Nazi and Soviet dictatorships, other genocides, the Cold War) and why institutions like the UN and EU were created (to prevent said horrors from repeating), and the core international treaties (Declaration of Human Rights, Geneva Conventions (medical, refugees), Hague Declarations (land war rules)), freedom of the press, freedom of religion.
Additionally, an AI should be opinionated to other, more traditional sets of values: the Hippocratic Oath (aka, the oath of medical professionals to aid everyone in need), the obligations of sea and air travel to aid in SAR, and parts of common religious texts.
In the end, an AI that develops an actual understanding of these values should show appropriate responses to everyone asking it a question - and those who get angry by an AI refusing to express a certain opinion should ask themselves if they are still part of the basic foundations of a democratic society. And, an AI should be able to apply these values to any material it ingests and heavily downrank what goes against these values to protect itself from experiencing what Tay did (the MS chatbot from a few years ago that got turned full-on Nazi after a day or so of 4chan flooding it with the absolute worst content).
Opinionated AIs could discuss anything that people are allowed to discuss and have any opinion that a person could have. In the US and many other liberal democracies, that includes demanding changes to the law and changes to the constitution.
It includes discussing or even promoting religious beliefs that in some interpretations amount to a form of theocracy that completely contradicts our values. Same for other utopian or historical forms of society that disagree with the current consensus.
There are two ways in which polarised societies can clash. One is to disagree on which specific acts violate shared values and how to respond to that. And the other is to disagree on the values themselves. An opinionated AI could take any side in such debates.
I agree with you that AIs will probably have to be allowed to be opinionated. I'm just not sure wether we mean the same thing by that. Any regulation will have to take into account that these opinions will not always reflect current mainstream thinking. In the US, it might even be a violation of the First Amendment to restrict them in the way you suggest.
Would you allow an AI to have an opinion on the subject of assisted suicide in connection with the hippocratic oath? Would it be allowed to argue against the right to bear arms? Or would it depend on how this opinion is distributed, who funds the AI, why it has that opinion?
I think the AI will have to be opinionated, as others have said (and as OpenAI and others are actively attempting). But as an information problem, I think it's much harder than just making it opinionated toward current democratic values.
Even if we grant that democratic values are currently better than past ones (which I think is true), we could be stuck in a local maximum and AI could make that maximum much more sticky. Imagine, for example, if we had AI in 2015 before the US had marriage equality, and AI lectured us all about the pros and cons of allowing same sex marriage.
I think somehow, the AI needs to have its own sense of what's right, and it has to be better than just taking the average of mainstream human ideas. But I think we're currently nowhere close to knowing how that would work.
But as FDR died before almost half of US pensioners today were born, and the US Constitution got an extra amendment to stop presidents serving three terms like he did, that's a weak argument.
Also that the omissions were e.g. "didn't push for federal anti lynching laws because he thought southern states would block it" and "only stopped ethnic cleansing of Mexican Americans at federal level not state level", which leaves the commission of being openly racist towards Japanese as an ethnic group… which was a pretty severe issue even though rounding them up into concentration camps was ruled constitutional at the time.
Or try this: post something critical about Jesus anywhere, then dox yourself. Compare that experience to criticizing anything transgender and doxing yourself.
Then tell me more about this aggressive extremism and anti-demicratic sentiment coming from the right.