Original author: Vitalik Buterin

Original source: Deep Tide

转载：Koala，火星财经

One of the Ethereum applications that excites me the most is the prediction market. In 2014, I wrote an article about futarchy, a governance model based on predictions envisioned by Robin Hanson. As early as 2015, I was an active user and supporter of Augur. I made $58,000 betting on the 2020 election. This year, I have been a close supporter and follower of Polymarket.

For many people, prediction markets are about betting on elections, and betting on elections is gambling—if it can bring enjoyment to people, then great, but fundamentally, it is no more interesting than buying random tokens on pump.fun. From this perspective, my interest in prediction markets may seem puzzling. Therefore, in this article, I aim to explain why this concept excites me. In short, I believe (i) that even existing prediction markets are a very useful tool for the world, and besides that (ii) prediction markets are merely an example of a larger, very powerful category that has the potential to create better implementations in social media, science, news, governance, and other areas. I will refer to this category as 'information finance.'

The dual nature of Polymarket: a betting site for participants and a news site for everyone else.

In the past week, Polymarket has been a very effective source of information regarding the U.S. elections. Polymarket not only predicted the probability of Trump winning as 60/40 (whereas other sources predicted 50/50, which is not particularly impressive), but also revealed other advantages: when the results came out, despite many experts and news sources trying to entice the audience with favorable news for Harris, Polymarket directly revealed the truth: the probability of Trump winning exceeded 95%, while the probability of taking control of all government departments exceeded 90%.

Both screenshots were taken at 3:40 AM ET on November 6.

But to me, this isn't even the best example of why Polymarket is interesting. So let's look at another example: the elections in Venezuela in July. The day after the election, I remember seeing someone protest the highly manipulated election results in Venezuela out of the corner of my eye. At first, I didn't pay much attention. I knew Maduro was one of those 'basically dictators,' so I thought he would certainly forge the election results to maintain his power, and of course, people would protest, and the protests would inevitably fail—unfortunately, many others did fail. But then, as I was scrolling through Polymarket, I came across this:

People were willing to stake over $100,000 betting on the possibility of Maduro being overthrown in this election at 23%. Now I started to pay attention.

Of course, we know the unfortunate outcome of this situation. In the end, Maduro did continue to hold power. However, the market made me realize that this time, the attempt to overthrow Maduro was serious. The scale of the protests was enormous, and the opposition executed an unexpectedly effective strategy to demonstrate to the world how fraudulent the elections were. If I hadn't received the initial signal from Polymarket saying 'this time, something is worth paying attention to,' I might not have started to pay attention at all.

You should never fully trust the Polymarket betting charts: if everyone believes the betting charts, then any rich person can manipulate them, and no one dares to bet against them. On the other hand, completely trusting the news is also a bad idea. The news has sensational motives and exaggerates the consequences of anything for clicks. Sometimes, this is justified, and sometimes it is not. If you see a sensational article but then go to the market and find that the probabilities of related events have not changed at all, then skepticism is warranted. Alternatively, if you see unexpectedly high or low probabilities in the market, or unexpected sudden changes, that is a signal to read the news and see what caused it. Conclusion: by reading both news and betting charts, you can gain more information than by reading either one alone.

Let's recap what has happened here. If you are a bettor, you can place bets on Polymarket, which is a betting site for you. If you are not a bettor, you can read the betting charts, which is a news site for you. You should never fully trust the betting charts, but personally, I have made reading the betting charts a step in my information-gathering workflow (along with traditional media and social media), which helps me acquire information more effectively.

Information finance in a broader sense

Now we enter the important part: predicting election outcomes is just the first application. The broader concept is that you can use finance as a way to coordinate incentives to provide valuable information to the audience. A natural reaction is: isn't all finance fundamentally about information? Different participants will make different buying and selling decisions because they have different views on what will happen in the future (aside from personal needs like risk preference and hedging desires), and you can infer a lot of knowledge about the world by reading market prices.

To me, information finance is like this, but structurally correct. Similar to the concept of structural correctness in software engineering, information finance is a discipline that requires you to (i) start with the facts you want to know and then (ii) deliberately design a market to optimally extract that information from market participants.

Information finance is a triadic market: bettors make predictions, and readers read those predictions. The market outputs future predictions as a public good (because that is its designed purpose).

Prediction markets are one example: you want to know a specific fact about what will happen in the future, so you set up a market for people to bet on that fact. Another example is decision markets: you want to know which decision, A or B, will yield better results based on some metric M. To achieve this, you set up conditional markets: you ask people to bet on (i) which decision will be chosen, (ii) if decision A is chosen, they get the value of M, otherwise zero, (iii) if decision B is chosen, they get the value of M, otherwise zero. With these three variables, you can determine whether the market believes decision A or decision B is more beneficial for getting the value of M.

I expect that a technology that will drive the development of information finance in the next decade is AI (whether it be large models or future technologies).

This is because many of the most interesting applications of information finance relate to 'micro' issues: millions of small markets, where the decisions have relatively small impacts when viewed in isolation. In fact, markets with low trading volumes often cannot operate effectively: it is not worthwhile for experienced participants to spend time conducting detailed analyses just to make a few hundred dollars in profit, and many believe that without subsidies, such markets cannot operate at all, as there are not enough naive traders to allow experienced traders to profit from anything beyond the largest and most sensational issues. AI has completely changed this equation, meaning that even in markets with trading volumes of $10, we can potentially acquire information of quite high quality. Even if subsidies are needed, the amounts required for each issue have become very affordable.

Information finance requires human distillation.

Judgment

Suppose you have a trusted mechanism for human judgment that the entire community trusts for its legitimacy, but making judgments takes a long time and is costly. However, you want low-cost real-time access to at least one approximate copy of that 'expensive mechanism.' Here are some ideas proposed by Robin Hanson about what you can do: Each time you need to make a decision, you set up a prediction market to predict what result that expensive mechanism would yield if called upon. You let the prediction market run and invest a small amount of money to subsidize the market makers.

99.99% of the time, you actually won’t call the expensive mechanism: maybe you will 'revoke transactions' and refund everyone's stake, or you will simply give everyone zero, or you will check if the average price is closer to 0 or 1 and treat it as a basic fact. 0.01% of the time—perhaps randomly, perhaps focused on the most traded markets, or a combination of both—you will actually run the expensive mechanism and compensate participants accordingly.

This provides you with a credible, neutral, fast, and inexpensive 'distilled version' of your originally highly credible but extremely costly mechanism (using the term 'distilled' as a parallel to 'distilled' in LLMs). Over time, this distilled mechanism roughly reflects the behavior of the original mechanism – because only participants who help achieve that outcome can profit, while others lose money.

Possible models combining prediction markets and community notes.

This applies not only to social media but also to DAOs. A major issue with DAOs is that there are too many decisions, and most people are reluctant to participate, leading either to widespread delegation, with the risks of centralization and principal-agent failure common in representative democracies, or being easily attacked. If actual voting in the DAO rarely occurs, and most decisions are determined by prediction markets that combine human and AI predictions of voting outcomes, then such a DAO might run well.

As we see in the example of decision markets, information finance contains many potential paths to solve important issues in decentralized governance, with the key being the balance between market and non-market: the market is the 'engine,' while other non-financial trust mechanisms are the 'steering wheel.'

Other use cases of information finance

Personal tokens—many projects like Bitclout (now deso), friend.tech, etc., that create tokens for everyone and make it easy to speculate—are a category I call 'primitive information finance.' They intentionally create market prices for specific variables (i.e., the expectation of a person's future reputation), but the exact information revealed by the prices is too vague and subject to reflexivity and bubble dynamics. It is possible to create improved versions of such protocols that address important issues like talent discovery by more carefully considering the economic design of the tokens (especially where their ultimate value comes from). Robin Hanson's concept of reputation futures is one possible end state here.

Advertising—the ultimate 'expensive but trustworthy signal' is whether you will buy a product. Based on this signal, information finance can help people determine what to purchase.

Scientific peer review—there has long been a 'replication crisis' in the scientific community, where certain famous results have become part of common wisdom in some cases but ultimately cannot be replicated in new research. We can attempt to use prediction markets to identify results that need to be re-examined. Before re-examination, such markets would also allow readers to quickly estimate how much they should trust any specific result. Experimental attempts at this idea have been carried out and so far seem to have been successful.

Public goods funding—one of the main issues with the public goods funding mechanism used by Ethereum is its 'popularity contest' nature. Each contributor needs to conduct their own marketing campaign on social media to gain recognition, making it difficult for those who cannot do so or who have more 'background' roles to secure substantial funding. An attractive solution is to try to track the entire dependency graph: for each positive outcome, which projects contributed how much, and then for each project, which projects contributed how much, and so on. The main challenge of this design is to figure out the marginal weights that can resist manipulation. After all, such manipulation has been happening. A distilled human judgment mechanism could help.

Conclusion

These ideas have been theorized for a long time: the earliest works on prediction markets and even decision markets date back decades, while similar discussions in financial theory are even older. However, I believe the current decade presents a unique opportunity for the following reasons:

Information finance addresses the trust issues that actually exist among people. A common concern in this era is a lack of knowledge (even worse, a lack of consensus), not knowing whom to trust in political, scientific, and business environments. Applications of information finance can help be part of the solution.

We now have scalable blockchains as a foundation. Until recently, the costs were too high to truly realize these ideas. Now, they are no longer prohibitively high.

AI as participants. When information finance must rely on human participation for every question, it is relatively difficult to operate. AI greatly improves this situation, enabling effective markets even in small-scale issues. Many markets may have a combination of AI and human participants, especially when the number of specific issues suddenly shifts from small to large.

To fully leverage this opportunity, we should go beyond merely predicting elections and explore what else information finance can bring us.

Special thanks to Robin Hanson and Alex Tabarrok for their feedback and comments.