AI tidies up Wikipedia's references and boosts reliability

When Sci-Hub made a lot of papers available to the public, I started clicking through to more references on Wikipedia.<p>My goal was to learn more and go deeper on subjects, but I was stunned by how often the linked citation didn&#x27;t support the claim in the Wikipedia article. There were many times where the linked citation said the opposite of the Wikipedia article.<p>My theory was that overly competitive Wikipedia authors were skimming PubMed abstracts and assuming the paper would support their assertion. Ironically, some of the statements with 5 or more citations were the most incorrect.<p>Trying to correct these articles is some times like going to war with editors who <i>refuse</i> to admit they were wrong.

I feel like <i>this</i> right here is what the singularity <i>actually</i> feels like.<p>With minimal effort, humans hookup AI to do some job, and things &quot;just get better&quot; rather than entropy taking its natural course and many things (without maintenance) trending towards &quot;worse&quot;.<p>Once you have a bunch of this human &#x2F; super-human level doing mundane things on wikipedia, they&#x27;re now there in perpetuity, constantly improving.<p>I suspect this is what&#x27;s going to start to happen across the economy: all of a sudden, the sidewalks seem cleaner, and trains run on time more often, traffic seems less congested, and latency in your favorite software product starts going down (with AI being turned loose on that legacy&#x27;s software&#x27;s code base that it gradually refactoring and optimizing in the background).<p>Effectively, what typically is happening due to entropy (decay, latency, quality, dirtiness) will start to move in the opposite direction due to automation and background AI.<p>This reversal of perceived entropy will start gradual, and then accelerate, and then on a day-to-day basis many things you touch in your daily life will be improving and then... singularity.

&gt; A neural network can identify references that are unlikely to support an article’s claims, and scour the web for better sources.<p>That seems like the wrong approach? The claims of an article should be informed by all relevant sources, not the selection of sources be informed by the claims of an article.

<i>Urman points out that the Wikipedia users who tested the SIDE system were twice as likely to prefer neither of the references as they were to prefer the AI-suggested ones. “This would mean that in these cases, they would still go and search for the relevant citation online,” she says.</i><p>This seems like an optimistic interpretation.

What i dont understand is how we cant just feed papers into some sort of text &gt; logical fallacy analysis which checks for the known fallacies, checks argument logic, checks sources (scores based on study size, and other requirements which qualify good studies) and just stops things right in their tracks before being added the &quot;corpus&quot; of knowledge..?? I&#x27;m just talking out of my ass here - im sure politics and other human factors stop something so seemingly simple from being implemented....<p><a href="https:&#x2F;&#x2F;www.researchgate.net&#x2F;publication&#x2F;340531396_Automated_Discovery_of_Logical_Fallacies_in_Legal_Argumentation" rel="nofollow noreferrer">https:&#x2F;&#x2F;www.researchgate.net&#x2F;publication&#x2F;340531396_Automated...</a>

Perhaps someone who has read the paper itself can address this question:<p>If input a claim that is unsupported, and you search for support, isn&#x27;t that a fundamental error? That is, is the software assuming the claim is accurate?

<p><pre><code> When SIDE’s results were shown to a group of Wikipedia users, 21% preferred the citations found by the AI, 10% preferred the existing citations and 39% did not have a preference.</code></pre> and 30% didn&#x27;t respond?

SIDE looks impressive I wonder how it works and finds these resources.<p>I think the evaluation is flawed these subjective numbers would mean nothing if I did the survey. Looking at some Wikipedia pages in the SIDE demo (linked in comments here), it is clear that they in some cases fail to identify what claims are made in the article, and that the subjective choice of references are too binary.<p>I double check references on wikipedia in subjects where I have basic understanding, it is usually easy to find better references but it takes so much time. So very impressive.

Isn’t this kind of like saying “Adding a blur filter makes people look younger.”

I’m sure I’m not the only person that read it as AI tiddies, so I’ll take the hit here just to let you know it wasn’t just you.

Is there a link to this SIDE project?

AI is the replacement for low-effort work.<p>What annoys me is that all the managers who were accruing technical debt were right. AI will clean it up for them, decades later.

this publicity does not include the MWF internal efforts for the last six+ years? only a META-owned research project?