But let's consider the following:
We currently have the ability to generate entirely artificial high-definition images of faces using, e.g., NVIDIA's StyleGAN2 that are for all practical purposes indistinguishable from real faces. We are able to generate simple text postings in constrained environments (such as sports stories and social media posts about politics) using Natural Language Generation tools. Here is a fun Real vs. Fake Faces Survey you can take to see how well you do (thanks Alex Teboul!). Scary!
We have AI tools to help content authors insert fake audio into altered video clips that are thought believable by unsophisticated social media viewers. The current technology is a little rough, and requires many hours of video to build models of real mouth movements sufficient to generate fake movements for a particular person, but this technology is likely to explode in the near future.
While we don't have it yet, we will likely also have in a reasonably near future the ability to generate entirely artificial video of novel talking heads in the same way we can generate novel artificial photos of faces.
It is an easy jump to consider what elections in the United States will look like in the not-very-distant future when most Americans get their news through social media and biased Internet news outlets, and any candidate and supporter can be convincingly shown to say virtually anything the opposition wishes.
Such purely deceitful "news" sources can, and will, be coming not only from internal factions in the United States, but also from foreign actors.
What are the ethics of controlling such fake content on the Internet? What are the dynamics of the arms race between those using AI to generate vast mountains of artificially created fake news—including video and text—and those using AI to detect such fakes?
Will Artificial Intelligence necessarily lead to massive-scale censorship of free speech because so much of that speech is fake?