Curating Content through Prediction Markets and Network Tokens

And now for something completely different: Below I describe an idea I had to use prediction markets and self-issued currencies (network tokens or appcoins) as a way of curating content. Any feedback on this idea is much appreciated (please note that stealing the idea and actually turning it into a product is the best kind of feedback!)

---

Curating Content via Prediction Markets and Network Tokens

As a content curator (e.g. Reddit) and/or publisher (e.g. the Let's Talk Bitcoin network) you want to provide your audience with those articles, videos, podcasts etc that they are most likely to appreciate, and you want that content to be delivered to them fast and in an easy-to-find way. 

This project proposes to use a combination of prediction markets and self-issued network tokens (e.g. LTBCoin.com) to solve two main problems that a curator and/or publisher of content has to deal with in accomplishing these goals:

1. the prevention or minimization of content being submitted that does not meet minimum standards

2. the incentivization, in proportion to its expected and actual popularity, of the production, selection and promotion of original content AND of the selection and promotion of content that already exists elsewhere on the web

Preventing or minimizing the submission of content that is not publish-worthy

The first problem is solved as follows: anybody who submits a piece of content has to put up a small amount of LTBcoin which she loses if and when the item is shown before a certain period of time has elapsed not to meet the minimum standards (e.g. if it is spam, plagiarized, not sufficiently well produced). Else she gets this small amount back. This requirement can be seen as the requirement to bet on her own article, to predict that the article she submits meets the minimum standards.

The funds that don't need to be returned can be used for whatever purpose LTB wants to use them.

The downside of this approach is that it may unnecessarily disincentivize (by slightly increasing the risk of) the submission of publish-worthy content.

Producing, predicting and promoting

The proposed solution to the second problem goes like this: when somebody submits an item of content she receives a voucher from LTB that says: 'If this item is in the top x of most popular items in x period of time, the holder of this voucher receives x LTBcoin.' (this is the simplest form in which the event-condition and the reward can be specified, but it could easily be made more complex by differing the amounts to receive based on e.g. the type of content or the exact position of the item in the top x)

The user is free to sell this voucher to others and she can also sell parts of it. For example, if the voucher entitles her to 10 LTBcoins if the item is in the top x she can sell somebody a voucher for 5 of those coins, somebody else a voucher for 3 of them and keep a voucher for the other 2 herself.

The reason LTB sponsors the prediction market for original content is because the market produces information that LTB can use to select which content it will publish where. It could for example publish on its front page those pieces of content that the prediction market indicates have the highest probability of popularity.

People will buy or sell these vouchers relative to the difference between the current market price and their own estimate of the probability of the event that is the condition for the payout. For example, if I think there is a 70% probability that the item will end up in the top x and if there is a 10 LTBcoin reward if it does, then I will offer <7 LTBcoin for the voucher.

The current market price of the voucher can then be used as a (crude or or more refined (if one adds complexity to the reward structure) measure of the expected popularity of the item of content.

Moreover, others are free to trade derivatives, either of the original content submitted to the LTB network (somebody could offer to pay somebody else 10 LTBcoin if said item of content meets the condition specified in the voucher in return for 6 LTBcoins if it does not) or of content that exists anywhere else on the web (somebody could offer to pay somebody else 10 LTBcoin if any given item that exists elsewhere on the web reaches some metric of popularity in return for 6 LTBcoins if it does not).

The crucial difference between the market for the original voucher and the derivatives market is that in the former case the payout comes from LTB while in the latter case it comes from market participants. In an important sense, by offering a payout LTB is sponsoring / subsidizing a prediction market in the case of original content while it merely facilitates a prediction market for derivatives (of original content and/or content that already exists elsewhere on the web.

The reason LTB sponsors and/or facilitates the prediction markets is that the markets produce information that LTB can use to select which content it will publish where. It could for example publish on its front page those pieces of content that the prediction market indicates have the highest probability of popularity.

Advantages of the model

This system accomplishes the following objectives:

- It incentivizes the production of content in proportion to that content's expected and actual popularity: the producer of content is the one who receives the voucher for free. Everybody else has to pay to obtain the voucher or a derivative.

- It incentvizes the accurate prediction of the popularity of any given item of content: anybody who has knowledge / insight about the probability of the condition-event that is not yet reflected in the market price has an incentive to buy or sell the voucher, with the result that the market price more accurately reflects the probability. Moreover, the requirement of putting one's money where one's mouth is a way in which those with superior knowledge / insight can self-select. Lastly, over time (after 1 or more 'rounds' of this market) people who were better at predicting popularity are rewarded while those who were worse at it suffered losses, causing many of them to eventually exit the market. (these selection effects are often thought to account for the relative predictive superiority of prediction markets in a wide variety of contexts)

- It incentivizes the promotion of that content to the audience: anybody who has a voucher or a derivative is incentivized to help promote the item so that it becomes popular enough to meet the condition for payout.

Problems and challenges

First I list 3 problems with prediction markets in general, then show how at least 2 of these are solved in the proposed system:

1. they are typically illegal

2. they can be expensive to set up

3. it can be hard to find a way of describing the condition-event so as to later be able to measure / agree on whether or not the condition-event has taken place

Network tokens solve problem #1 as network tokens are not real money and prediction markets are only illegal when real money is used. Interestingly and fortunately the use of non-real money (such as LTBcoins) has been shown not to significantly detract from the predictive power of prediction markets.

Bitcoin and related technologies (e.g. Mastercoin, Ethereum, Counterparty, but also multisig) solve problem #2 (Vitalik Buterin wrote in his article on prediction markets that they are easy to implement with Bitcoin)

Problem #3 is the most difficult one: how to come up with a way of defining and hence being able to measure/determine a piece of content's popularity? # of likes? # of upvotes? # of Twitter mentions? All have their problems, one of which is that they are manipulable to a lesser or greater extent.

The temptation may be to try to solve the problem of predicting popularity (expected popularity) and measuring popularity (actual popularity) through one and the same mechanism. This is for example how the (fascinating and admirable) Proof of Tipping system proposes to do it, but it doesn't work:

1. In the 'proof of tipping' (POT) model tipping really is an untenable hybrid of tipping and predicting/investing. Normally tipping is the expression of a personal preference. I tip an article because I thought it was great. Predicting/investing on the other hand is a way of trying to get a ROI by predicting better than other people do some future event. In this case the event is the relative popularity of an article and the reward comes in the form of newly issued LTBcoins.

The problem is that these two different natures can pull in opposite directions. For example, it makes sense for me to predict/invest in an article because I think a lot of other people will like the article even though I myself may think the article is not very good. And conversely, it makes sense for me to tip and article if I think it's great even though I think few others would agree with me.

In the first case the predictive function undermines the expressive function and in the second case the expressive function undermines the predictive function.

2. The system can be manipulated by somebody (or a group of people) tipping their own article as much as possible, undermining both the predictive and the expressive functions.

Prediction markets without an external way (external to the prediction market itself) of specifying the condition-event (the actual rather than just the expected popularity) would sort of be the flip side of this kind of Proof of Tipping system and would run into the same kinds of problems.

Expected and Actual Popularity

As far as I can tell you simply need a different mechanism for measuring expected popularity (prediction markets) and actual popularity ('likes', upvotes, Twitter mentions, page views, a combination of these and/or other things).

Which is not to say that expected popularity cannot or should not influence actual popularity. For example, it seems clear that the prediction market itself influences the actual popularity of the event if items with a high expected popularity are shown on the front page while those with a lower expected popularity are only shown elsewhere on the site/network. By being on the front page these items reach a larger audience and become more popular. This is true, but it is not clear that this is problematic. It may just be efficient. Consider for example the following:

1. The influence of predictions of events on the actual probability of events is common: if a prediction market says that presidential candidate A has 2% chance of being elected while candidates B, C and D each have a 20% chance, A may not be invited to debates, which may very well make it less probable that A would be elected than if he were invited to the debates.

2. The influence of predictions is exactly what incentivizes people to try to change things in the actual world, e.g. what incentivizes holders of vouchers to try to promote the item of content in question.

3. To at least some extent this reduced probability is already reflected in the prediction itself and hence in the incentives, and the larger the difference between the current market price and a person's subjective estimate of the event's probability the stronger his incentives to promote that item.

I don't know whether these considerations are a sufficient answer to the concern though.