If one were to build an open platform to compete with news media rather than social or blogging media, they would need some sort of accountability/quality control mechanism to mirror the role of editorial review at prestigious news platforms.

The best tool for aggregating information that we have devised is markets. This is especially the case in open systems. The biggest risk to a well-functioning market is collusion. The second major problem with prediction markets for information is that bettors can become more concerned with predicting what people think the truth is rather than predicting the actual truth. Any mechanism that was designed to employ markets in the place of editorial review would need to guard against these failings. With this in mind I propose this system:

  1. A reporter with information on a climate event in Costa Rica writes an article and publishes it to the platform. They must stake a certain TBD amount of money on this article.
  2. The article is published without editorial review (unlike how news media currently works).
  3. Prospective fact checkers (post editorial reviewers) also stake money and state their specialist topics. Some of those that listed Costa Rican current affairs or climate as a topic are randomly chosen from the available pool of people. They are each given guidelines on how to judge an article and use these guidelines to give the article a trust score without colluding among themselves as they don’t who else's been asked to fact check. The article and the guidelines together likely represent the only Schelling point the fact checkers can converge on as the position of the market is unknown prior to settlement. Information about any consensus that might exist outside the truth is unknown. A fact checking assignment is like jury duty. Some of your stake is slashed if you renege on giving a score for a given article in your specialty. This requirement along with random selection should minimise any collusion.
  4. The writer’s payout is determined by the score the fact checkers give him/her. The fact checkers’ payouts are determined by how close they are to the average score from the group. Everyone’s respective rep scores are also updated. Those that drop below a certain trust score will not have their articles listed on the platform or be chosen for post editorial review. Articles from the best performers will be amplified on the platform, minimising the chances of readers being supplied erroneous information.

Would be interested to hear people’s thoughts on this system? One worry I have is that after some time people will be armed with prior probabilty data and will simply strategically pick the high probabilty outcome without doing any fact checking. This could be mitigated by minimising the amount of articles a fact checker gets, leading them to take the utmost care with the ones they are given.

10

0
0

Reactions

0
0
Comments3


Sorted by Click to highlight new comments since:

There's a bayesian truth serum literature coming out of mechdzn/econ/cs, lots of results when you google "bayesian truth serum". What sort of market design principles do you need to elicit honest beliefs with respect to some ground truth? that sort of thing. 

Interesting, thanks. 

I found your post interesting and your proposal compelling. Noam Bardin (founder of Waze) recently founded Post.News, a platform that aims to implement a version of specific parts of your proposal in their mixed new news and social media site. His interview on the Pivot Podcast explained very little details of the product (mainly his "vision"),  although they are still early in development and seemed to have rushed their launch given the happenings at Twitter. Nonetheless, the parts of that podcast episode that I found most interesting and may be relevant to your proposal are:

  • 42:30 - Waze established a hierarchal system of maps editors that have built reputation over time and have jurisdiction over specific geographic locations. A version of this that may be implemented in Post.News sounds like having specific content moderators for specific geographies regions or topics.
    • This sounded similar to the fact-checkers in your proposal, in which fact-checkers list topics they are experts in. A key difference between the Waze and Post.News examples and your proposal, is that your proposal suggests that fact checkers are selected at random for each article to prevent collusion.
    •  The key similarity between Waze, Post.News and your proposal is that these actors' objectives are to optimise for some specific score - whether that is increasing maps accuracy, minimising user reports of community violations or closeness to the average score to other fact-checkers for a given article - with the purpose being that optimising for these scores improves the quality of the service.
  • 53:57 - Establishing a two-tier user system in which:
    • Content on Post.News by verified users (real identity is verified using a paid 3rd-party service) is distributed amongst and beyond your followers using a reputation system to determine degree of amplification. Verified users that cross some threshold of 'bad content' as reported by content moderators or others users, will be removed from the reputation system and their content will only be distributed to their followers. Content on Post.News by anonymous users (real identity is not verified) is distributed to only their followers.
    • I am curious about what happens to content in your proposal that isn't chosen for post editorial review (ever) and what happens to the post while it is in review. Do you think a two-tier user system for content is necessary to attract content, writers, fact-checkers and reader (i.e. there needs to be a certain volume of content for you to attract users and build the news site)?

Question: In your proposal, how would the system onboard fact-checkers and what factors would influence selection of new fact-checkers if reputation scores are 'fresh' broadly on the platform or for a specific topic? Would bad fact-checkers be weeded-out by poor scores from their first few fact-check assignments? 

Question: In your proposal, how / would you enable fact-checkers to list topics of expertise? Again, would topics be removed from fact-checkers' profiles by poor scores from their first few fact-check assignments?

An observation I have about your proposal is the high volume of fact-checkers you would need to onboard to keep up with article content, particularly given that your proposal needs multiple fact-checkers per topic for one article.

My prediction is that future AI systems will emerge that will perform fact-checking activities for a significant volume of content (I am uncertain what amount this would be). These systems will have their own 'knowledge graph' of facts, extracted from a variety of sources, influenced by sources' reputations (that they track), so humans don't have to. Rudimentary versions of this already exist (e.g. Google places an answer in a callout box at the top of search results, Microsoft's implementation of OpenAI's GPT3.5 presents web sources when it produces prompt responses). @quinn's comment about Bayesian Truth Serums are similar to what I envision for mechanistic fact-checking.

Its worth nothing that I've just started an self-direct research project on this very topic and would be open to chatting further if you are invested in this topic too.

Thanks for posting!

Curated and popular this week
trammell
 ·  · 25m read
 · 
Introduction When a system is made safer, its users may be willing to offset at least some of the safety improvement by using it more dangerously. A seminal example is that, according to Peltzman (1975), drivers largely compensated for improvements in car safety at the time by driving more dangerously. The phenomenon in general is therefore sometimes known as the “Peltzman Effect”, though it is more often known as “risk compensation”.[1] One domain in which risk compensation has been studied relatively carefully is NASCAR (Sobel and Nesbit, 2007; Pope and Tollison, 2010), where, apparently, the evidence for a large compensation effect is especially strong.[2] In principle, more dangerous usage can partially, fully, or more than fully offset the extent to which the system has been made safer holding usage fixed. Making a system safer thus has an ambiguous effect on the probability of an accident, after its users change their behavior. There’s no reason why risk compensation shouldn’t apply in the existential risk domain, and we arguably have examples in which it has. For example, reinforcement learning from human feedback (RLHF) makes AI more reliable, all else equal; so it may be making some AI labs comfortable releasing more capable, and so maybe more dangerous, models than they would release otherwise.[3] Yet risk compensation per se appears to have gotten relatively little formal, public attention in the existential risk community so far. There has been informal discussion of the issue: e.g. risk compensation in the AI risk domain is discussed by Guest et al. (2023), who call it “the dangerous valley problem”. There is also a cluster of papers and works in progress by Robert Trager, Allan Dafoe, Nick Emery-Xu, Mckay Jensen, and others, including these two and some not yet public but largely summarized here, exploring the issue formally in models with multiple competing firms. In a sense what they do goes well beyond this post, but as far as I’m aware none of t
 ·  · 1m read
 · 
 ·  · 19m read
 · 
I am no prophet, and here’s no great matter. — T.S. Eliot, “The Love Song of J. Alfred Prufrock”   This post is a personal account of a California legislative campaign I worked on March-June 2024, in my capacity as the indoor air quality program lead at 1Day Sooner. It’s very long—I included as many details as possible to illustrate a playbook of everything we tried, what the surprises and challenges were, and how someone might spend their time during a policy advocacy project.   History of SB 1308 Advocacy Effort SB 1308 was introduced in the California Senate by Senator Lena Gonzalez, the Senate (Floor) Majority Leader, and was sponsored by Regional Asthma Management and Prevention (RAMP). The bill was based on a report written by researchers at UC Davis and commissioned by the California Air Resources Board (CARB). The bill sought to ban the sale of ozone-emitting air cleaners in California, which would have included far-UV, an extremely promising tool for fighting pathogen transmission and reducing pandemic risk. Because California is such a large market and so influential for policy, and the far-UV industry is struggling, we were seriously concerned that the bill would crush the industry. A partner organization first notified us on March 21 about SB 1308 entering its comment period before it would be heard in the Senate Committee on Natural Resources, but said that their organization would not be able to be publicly involved. Very shortly after that, a researcher from Ushio America, a leading far-UV manufacturer, sent out a mass email to professors whose support he anticipated, requesting comments from them. I checked with my boss, Josh Morrison,[1] as to whether it was acceptable for 1Day Sooner to get involved if the partner organization was reluctant, and Josh gave me the go-ahead to submit a public comment to the committee. Aware that the letters alone might not do much, Josh reached out to a friend of his to ask about lobbyists with expertise in Cal