Although I am genuinely intrigued by AI running out of things to ingest, and moving onto AI generated content. Is the snake starting to eat its tail?
AI will produce whatever evidence a person needs in order to shore up their own "skeptical" beliefs. Same problem as social media, where finding 1000 other people who share your fringe beliefs looks like compelling evidence for your belief being true. Just now it's on-demand, hyper-personalized, responsive to your own doubts, and there's no chance of you realizing "ah, my comrades on the other end of the tube are actually idiots!"
I'd bet an AI product that tells a flat earther that "it's an open question" will be much more successful than an AI product that attempts to dispel that myth. That is, at least until an AI can effectively convince people away from their beliefs, which almost certainly will not happen via "calmly providing solid logical evidence," given that this is not usually an effective vector for persuasion anyway. Now you've got a new problem, which is a technology capable of convincing people of all sorts of insane things.
The past couple of years has convinced me this is 100% untrue.
People are very capable of ignoring evidence, in favour of something that:
a) reinforces previously held beliefs
b) allows them to remain in a social group
Most people believe what they are told at face value.
I stop just short of Feynman questioning the dentist on the evidence for brushing one's teeth. That there is potentially this ritual that goes around the globe as the sun rises of people pointlessly scrubbing their teeth.
Of course, with such a high % of people addicted in a clinical sense to the group think and propaganda engine of social media that is not how most people are going to think or even be capable of thinking.
If Feynman was more popular youtube would ban that video for dentistry misinformation even when the point of the video is to view things from a different perspective.
Writers should adopt two habits: (1) Sign your statements. If you say something, sign it, so that others know you said it. Cryptography is good at this. (2) Hash your citations. If you cite something, include a hash of it. This way, if the thing you cited is altered, readers can tell that that's not what you were citing. Note that this idea can be applied to audio and video, not just text.
The rest of the responsibility falls on readers:
(1) Read (i.e. consider) citations.
(2) Read (i.e. consider) the sources of evidence you ingest.
Don't just find a video and believe it happened. Determine who has claimed to have witnessed those events.
For this, a public database of back-references might be helpful. But even without one, a decentralized solution is possible. Writers, whenever they cite something, could simply send their citation (and its context) to the author of the cited material. If the cited author attaches the back-citation to the content that was cited, then anyone who comes across the content can see who has cited it.
There is of course the problem that some back-citations will be rejected -- if you cite what I wrote to call it stupid, I am unlikely to want to share that fact with the world. But if what I wrote is sufficiently important, then hopefully someone will waht to host a "nemesis" site, which collects negative citations.
A public database of nemesis sites would be helpful.
(3) Read (i.e. consider) the reputations of authors you read.
This is nearly the reverse of the last point. When deciding whether to believe what someone has said, consider what else they have said.
This is of course a hard problem. An author might be qualified in one area and writing about another. An author's reputation might be damaged for extrinsic (e.g. malice) reasons, rather than intrinsic ones.
But a statement's author is too important a context to ignore.
(4) Do cool graph-traversing investigations.
Determine who someone tends to cite. Identify misinformation cliques -- close-knit collections of liars who all cite each other. Identify readership patterns that make people productive.
We have seen how social network information can make a corporation money. As a society, I suspect there is a similar amount of value to be extracted from them.
The parent comment, mentions that an encyclopedia page can be modified by a bot. That holds true for wikipedia, but we can create encyclopedias strictly edited only by humans.
Just have a prominent individual issue a top ecdsa identity, with correspondence to the real person's name info known only to him. He publishes that ecdsa identity somewhere, let's say on a blockchain to be always available and secure from deletion. Let's say this prominent individual is the Ronaldo football player. He publishes 1000 ecdsa identities to a public digital highway somewhere, all of the real names connection known only to him. That set of 1000 identities is called Ronaldo's social graph.
From then on, each child identity derived from the top identity, when they edit a wikipedia page, they are pseudonymous if they like. No need to reveal their name, only Ronaldo knows that, but we know they are human, because Ronaldo has met everyone in person in order to issue the top identity. But pseudonymous is only as far as they can go, because someone will always know their real name. A.I. actually spells the end of anonymity on the internet.
One more property of an organization structure like that, is that as soon as a person loses his wikipedia account for some reason, he can always get it back, because he can create a new ecdsa child identity, and prove that his older account and his new, match exactly the same top identity. So he can always invalidate older accounts and use the same data, karma etc, with new accounts.
The only downside of that organization structure, is that top identities which belong to the public social graph, have to be absolutely secure. As soon as a person loses his top identity, Ronaldo has to issue a new one, but the encyclopedia cannot invalidate accounts not matching the top identity in an automated way, if the real name is not published. That means a human on the other side has to be involved and boureocracy ensues.
The only two answers I can come up with are
A) independently verifiable facts, for example you can apply the scientific method to the hypothesis that the earth is not flat (make predictions that should follow from that, and test those experimentally); or
B) data provenance. If some crackhead says that the US government is conducting brainwashing experiments you might discard that, if the government answers a Freedom of Information Act request with documentation about brainwashing experiments conducted by the CIA then you have good reason to count them as evidence. And you spread the word about this by showing the proof to a reputable newspaper who write about it, or writing a book and publishing it at a publisher known for fact-checking what they publish.
Anything that isn't independently verifiable or has a chain of provenance is already hearsay. In the age of social media we got used to basing a lot of decisions on hearsay, so maybe we have to dial that back. But AI being better at generating hearsay doesn't mean it gets better at creating evidence.
But as for the case of the Fauci emails from FOIA requests no "reputable" newspaper reported on it. So people still to this day dismiss it since it did not come from a "reputable" source despite the fact you can confirm the provenance and authenticity of these documents by confirming with the agency that released the documents
https://www.washingtonpost.com/politics/interactive/2021/ton...
We're talking about MKUltra right? The classic "crazy" conspiracy theory that turned out to be true but was discredited for decades?
Of course that could be what happens. An entire new industry of "Certifications", companies that "Verify" media.
Just like humans eventually learn that not all input is equally trusted (for example the input "2+2=5"), so should AI eventually learn to sift through.