How to beat “fake news”
Most observers hold several or all of the views:
- “Fake news” and the like are severe problems.
- Algorithmic solutions have not worked well to date.
- Neither have manual ones.
- Trusting governments to censor is a bad idea.
- In light of the previous points, trusting large social media corporations to censor is a bad idea too.
- Educating consumers to evaluate news and opinions accurately would be … difficult.
And further:
- Whatever you think of the job traditional journalistic organizations previously did as news arbiters, they can’t do it as well anymore, for a variety of economic, structural and societal reasons.
But despite all those difficulties, I also believe that a good solution to news/opinion filtering is feasible; it just can’t be as simple as everybody would like.
1. When people think about these problems, they’re probably most focused on social media platforms such as Facebook, YouTube or Twitter. But before getting to those, let’s consider the simpler case of search engines. In essence, what search engines do is:
- Assign a relevance score to the relationship between a site (or particular page) and your query.
- Assign a quality score to a site.
- Combine those two scores into an overall ranking, and serve up results accordingly.
How well does this work? I’d say that search engines:
- Are good at directing you to information that is generally related to what you want to know. This is the technological core of what they do.
- Are good at shielding you from the worst cheating/spammer/hacker sites. That’s also a major technical focus for them.
- Are poor but not horrible at distinguishing between good and bad sources of information, opinion or advice. Generally, they do this via some kind of popularity contest, whether via Google’s venerable PageRank or by more directly observing which sites users seem to go to and stay at.
- Don’t even try to filter sites according to leanings such as political bias.
Lessons from that start:
- Huge technology companies can actually do pretty well at the parts of the problem that technology alone can solve.
- A lot of the challenge boils down to adversarial information retrieval, where the adversaries range from somewhat honest polemicists or hucksters to completely awful hackers and spammers.
2. When defending against bad actors, scale helps a lot. In my favorite example:
- A significant fraction of all the world’s email goes through Gmail.
- Thus, it is very hard for email spam blasts to escape Google’s honeypots.
- Informed by those honeypots, Google does what in my opinion is a very good job of fighting spam.
Similarly, as the publisher of multiple blogs, I can tell you that much the same is true of WordPress’ Akismet’s fight against spam comments. Akismet isn’t perfect; indeed, I’ve stopped adding new content to the blog where this post would fit best — Text Technologies – because of a multi-year spam attack. But on the whole Akismet works very well.
Thus, in contradiction to many observers, I believe that the huge scale of social media companies is NOT the root of the problem.
3. Of course, concern is really focused on social media, and especially on the concern that people communicate things they (supposedly) shouldn’t, where:
- “Communicating” includes words, pictures, videos, etc.
- “Shouldn’t” covers outright lies, great factual distortions, hate speech … or just opinions that the would-be censor doesn’t think should be spread.
And even if you don’t worry so much about those problems, some kind of censorship, filtering or gatekeeping is inevitable anyway, simply because there’s vastly more information in the world than any one person can consume.
So what are the main options for censorship and other gatekeeping? My opinions start:
- Having governments be in charge of censorship is a terrible idea.
- Having large, non-journalist corporations be directly in charge of censorship is also a bad idea, because ultimately they’ll just succumb to government or other political pressure.
- The traditional modern gatekeepers are journalistic organizations, who both deserved and received trust that they’d do the job responsibly. But that model no longer suffices in its old form, for several sets of reasons:
- New requirements. Traditional journalistic gatekeeping boils down to organizations vetting the content they themselves produce. It doesn’t work nearly as well for third-party content, worthy efforts such as fact-checking columns notwithstanding.
- Trust. Walter Cronkite is long dead; journalists aren’t nearly as trusted as they used to be.
- Deliberate bias. Opinion and bias are now part of many “news” organizations’ business models, to a much greater extent than they were a few decades in the past.
- Money. Most journalistic organizations have much slimmer news budgets than they had at their peak.
4. So if we need gatekeeping, and no natural kind of gatekeeper can on its own be effective or safe, what’s left? In simplest terms, we need gatekeeping by (technological) committee. Mainly, what I propose comprises:
- Different kinds of gatekeeper for different aspects of the problem, including at a minimum:
- Human-led filters to deal with various issues in credibility and bias.
- Technology-led filters to deal with pure fakery and false provenance.
- Further filtering of the kinds that would be needed even in a more benign world.
- Multiple choices for at least the human-led filters.
- Good, simple (!) user interfaces for combining those filters’ results.
Above all, people must be able to choose their own censors.
5. What I envision for the “human-led filters to deal with various issues in credibility and bias” is something like:
- An organization maintains slowly-changing whitelists and blacklists of information sources.
- The same organization fact-checks or other vets specific stories, claims and content in near-real-time.
- The results scale to other stories, claims and content via very-rapidly-retrained machine learning models, whether those are based on a single gatekeeper’s hand editing or, more likely, on multiple hand-edited training sets and other collaborative inputs at once.
Here an “organization” can be anything trusted by enough people to be economically viable, for example:
- An offshoot of an existing journalistic organization.
- An offshoot of an existing political party or advocacy organization.
- A successfully started-up new entity.
Obviously, there would be business issues, notably:
- Costs. Who pays, in an economy where news is commonly “free”?
- Chicken-egg adoption. Which gets developed first: Human-led filtering services that can’t yet be integrated into actual social media filtering, or technology to integrate human-led filtering services that don’t yet exist?
But given the importance and visibility of the problem, optimism about solving the business issues is appropriate. The hardest part is the technology itself. Can machine learning models be retrained on a sub-hour or even sub-minute basis? Sure. That’s been confirmed many times. But what I’m suggesting is a pretty complex case, with global scale, intermediate results passed among organizations, with plenty of adversarial elements, all done at very high speed.
That is not yet a solved problem. But it certainly seems solvable. Further, it’s a problem that must be solved, lest liberal democracy be as doomed as some people fear it actually is.
Related links
- I wrote a bit about adversarial analytics in May, 2016.
- I outlined my views about the “War(s) on Truth” in February, 2018.
- Earlier this month, Cory Doctorow offered a hard-hitting column on the dangers of expecting internet companies to do our censorship for us.
- More sedately but with more explanation, so did Will Oremus.
Comments
8 Responses to “How to beat “fake news””
Leave a Reply
[…] there are less dangerous ways to address the same challenges. I expect to make as much fuss about this issue in the upcoming […]
Fantastic and Knowledgeable Blog…Thanks for Sharing!!!
Best photographer in Allahabad
Demon Slayer Merch
How to beat "fake news" | DBMS 2 : DataBase Management System Services
An intriguing discussion is worth comment. I do think that you should publish more about this topic,
it might not be a taboo subject but typically folks don’t discuss these topics.
To the next! All the best!!
From designing simple websites to developing
robust software applications or crafting the most effective digital marketing strategy with SEO, AdWords and social media at the core,
we are the perfect web design company in Srilanka for businesses of all sizes.
You’ve provided so much clarity. Much appreciated!
Wonderful write-up! Your explanation on topic was spot on. Have you considered exploring another aspect?
Thanks for sharing this article! Do you think
software solutions will simplify this process in the near future?