Hacker News Reader

Comment by burtonator | Hacker News Reader

I've been working on some algorithms I think can identify fake news easily and reliably.

The key part is you have to have a massive dataset of historical information.

With Datastreamer (my company) we have about 1PB of news, blogs, and social media from the last 4-5 years.

I'm curious if it would make sense to ship an open/public API that would allow people both to report AND check on whether something is fake news.

Sort of like Hacker News but a browser extension that allows you to flag whether you think something is fake.

View on HN

What could possibly go wrong?

False flagging, FUD based on groupthink and cognitive biases, accidental suppression of newly created websites in favor of older ones. Just to start off a list.

Not everything is solvable through algorithms. Especially if your initial goad is ill-defined. There are about half a dozen completely separate things that are being labeled as "fake news" today, from websites impersonating news agencies and pushing lies, to actual news agencies pushing lies, to viral hoaxes on social media.

Ultimately, whether a news article is fake or not is determined by its relation to reality. You could use context as heuristic to filter out complete garbage, but this is something that's actually being abused by many subcategories of "fake news" to begin with.

People should be able to use their own judgement to spot nonsense, without algorithms telling them what to think. If they can't, the first questions should be "why?" and "why now?", not "how do I make an app to fix it?".

Personally, I think the underlying assumption behind current social media are profoundly broken and can't be fixed just by sprinkling more tech on top of it all. Obvious example: if you release a news-checking app and someone else releases a fake news-checking app, how do you expect your users to choose the right one?

> People should be able to use their own judgement to spot nonsense, without algorithms telling them what to think. If they can't, the first questions should be "why?" and "why now?", not "how do I make an app to fix it?".

Sort of.

People should also be able to use their own judgement to spot scams.

And yet there are spam filters. Both to detect garbage, and also to hide garbage.

Obviously, the email client that classifies my bank loan documents as "spam" is not going to be popular.

What would make more sense is to build a certificate system around this that news organizations can use to declare an article is Fake News Free. Might be a nice business to start riding the Fake News Boom.

Taking this idea. What would be the hypothetical flow for the recent Bloomberg chip article? I’m sure it would have been published with the cert since it’s from Bloomberg but later when it turns out to be less than true or more inflammatory than intended. Do they lose their cert? Or maybe just the article has to persist with the cert rating?

I love the idea, just curious how it works logistically.

Maybe something like:

Bloomberg registers with cert auth. They publish articles and each has a pending cert item (image?) on their page that is loaded from 3rd party cert auth. The cert auth uses algos to determine if it's fake or not and published results back to the placeholder.

Edit: Forgot to mention, please do take the idea. And good luck!

Thx for the info. So could this possibly run as a browser plugin?

That way the site doesn’t need to assert its own trust? And you wouldn’t have to deal with the heavy lift of updating every news site.

Just thinking out loud typing... you could integrate this with an ad blocker so that only cert content could show an ad. This would create an incentive model not only for the publisher but the advertising partners to make sure content is good and certified. Sites and content with no cert or lost cert would block ads for that page and for repeat offenders you could escalate to say all ads are blocked for the sesssion.

While I’m not a fan of ads I understand their purpose and maybe with something like the above, the market could find equilibrium once again.

Hi I am interested in this sort of service. I have 2 concerns. First is I don't install browser extensions unless I really trust them. Too many are pumped up and then sold to some malware company, one update later and you're owned. My second concern is that if this works well eventually the fake news producers will become aware of it, reverse engineer it and then attempt to game the detection.

Fakespot has been popular for years, and spam reviewers haven't been able to game it.

I launched something like that as a side project https://www.quod.us (to let people report misinformation) and I was going to add an API to let people report things from anywhere, but I got busy with other projects.

What I found was that it's really hard to get people to report this stuff, I never really found my ideal user. A lot of people signed up, but only a few people ever actively reported things. I never finished validating the concept, but my general feel is that the only way to keep the project afloat is via grants and donations, and that's hard to do.

Email me if you ever want to collaborate.

> The key part is you have to have a massive dataset of historical information.

Why is this historical information not susceptible to the exact same issues today's news is?

How does an algorithm short of strong AI do that?

Reply on news.ycombinator.com