Mar 17, 2024 23:00:00

How does the future prediction community 'Samotsveti' send out highly accurate future predictions?

There is an academic field called ``

superforecasting, '' which uses probability to predict the difficult-to-predict future. It is a discipline that analyzes and predicts the future using various specialized fields and approaches, such as historical trends, mathematical and statistical models, AI training, and micromortes used as a measure of mortality risk. A person with excellent forecasting ability is called a 'super forecaster'. Samotsveti , a group that predicts the future as a forecaster, boasts a high accuracy rate in various fields despite not being a specialized research team, and the online media Vox explains the mystery of Samotsveti. doing.

How do you predict the future? Ask Samotsvety. - Vox
https://www.vox.com/future-perfect/2024/2/13/24070864/samotsvety-forecasting-superforecasters-tetlock

Superforecasting is a concept that became widely known after Canadian researcher Philip E. Tetlock presented it in his book “ Superforecasting: The Art and Science of Prediction ” published in 2015. In some cases, the ideas of super forecasters, who are excellent at predicting the future, are used as a reference for better policy decisions. In addition, a competition called the ``Forecasting Competition'' is held in which participants present difficult-to-predict future scenarios and compete for accuracy.

He won the Forecasting Competition in 2020 and 2021, and in 2022, he bet about $14,000 (about 2 million yen) on the possibility of Russia's invasion of Ukraine and made a bet of $32,000 (about 4.4 million yen). ) was won by Samotsveti from the Forecaster Group. Samotsveti is a name derived from the Soviet band

Samotsveti , which was formed in Moscow in 1971, and they mainly hold meetings using the communication tool Slack .

Dylan Matthews, a senior editor at Vox, describes his experience attending a Samotsveti meeting on Slack on a recent Saturday. The topic of discussion that day was ``What are the chances that China will control at least half of Taiwan's territory by 2030?'' '8%,' said Chinmay Ingalagabi, an economics fellow at Yale University, and Nuño Sempere, a Spanish independent researcher and consultant who led the conference, agreed. Students earning a master's degree in business administration at the University of Chicago predicted ``17%,'' researchers with a doctorate in neuroscience predicted ``15-20%,'' and an anonymous Chinese member predicted the highest rate of ``24%.'' Did.

According to Mr. Matthews, the participants in the Samotsveti conference were researchers and experts from various genres, and they did not necessarily have specialized knowledge on the topic. However, a study published in May 2022 (PDF file) states that ``When predicting the future, it is better to aggregate the opinions of non-experts than to aggregate the opinions of experts.'' It has been shown that 'predictions about geopolitical events are more likely to be true for groups that use standard methods to aggregate information than for groups that have access to confidential government information.' There is also the idea that no specialized knowledge is required to accurately predict the future, as evidenced by the often cited paper (PDF file) that says,

When Samotsveti makes good predictions, including discussions involving Matthews, he is rarely based on facts that are little known to others. However, Samotsveti co-founder Misha Yagudin explains why they are

able to make more accurate predictions than other researchers and groups. The name means a stone that 'glows light/colors'. This is a name that means that prediction is 'finding good nuggets of information.' Even if we are not diamonds, by making ourselves shine well, we can , we can become a great presence that shines a light on the future,'' he said, referring to the group's origins.

Some members of Samotsveti have received invitations to join the group because of their predictive abilities. Molly Hickman, who worked as a forecaster for a government-related organization, teamed up with her father and friends to take on Infer , a competition to see how accurate future predictions can be. Although the team itself did not perform well, Mr. Hickman achieved excellent individual results, and Samotsveti sent him an invitation to participate. In addition, Eli Reifland, who primarily works on predictions regarding advances in AI, predicted the explosive spread of the new coronavirus at a time when the number of infections was still underestimated. As topics related to the novel coronavirus were changing rapidly on a daily basis, a variety of items were important to predict, and Lifeland was able to successfully predict many of them with high accuracy. Eventually, Samotsveti contacted him.

However, Samotsveti is more than just a group of people who are good at forecasting. Matthews first cites ``quantitative reasoning'' as the reason why Samotsveti's argument is superior. Indicators such as 'somewhat likely' and 'very unlikely' are easy to understand when it comes to individual predictions, but they cannot be compared quantitatively in discussions. For this reason, Matthews points out that Samotsveti strives to discuss concrete numbers based on computer science, economics, mathematics, etc. According to Mr. Hickman, even if quantitative predictions are not made, some calculations are implicitly made, and making this clear will lead to more fruitful discussions.

Samotsveti also pays attention to what is called the 'base rate.' For example, when predicting the possibility of the New York Yankees winning the MLB World Series, the base rate is calculated as ``22.7%'' since the Yankees have won 27 of the 119 World Series to date. Something to do. In reality, we aim to make accurate predictions by incorporating the current team situation and recent trends, but Samotsveti considers the base rate to be a ``highly calculated number'' and predicts that the actual rate will be much lower than that. There are more things to do. In the discussion that Mr. Matthews participated in, he felt that ``even in cases where there is evidence that the probability of this happening is high, overall the numbers are underestimated.'' The reason for this is basically It is said that it is in the rate processing.

In addition, Jason Matheny, CEO of the US-based think tank RAND Corporation , talks about why Samotsveti's argument is a good one. Although Samotsveti conducts prediction discussions on some topic, his enthusiasm is not only focused on the accuracy of his predictions, but also on how to score the accuracy of his own predictions. Before joining the Rand Corporation, Matheny worked as a government employee funding forecasting work, but Samotsveti's attitude is unusual compared to his experience at that time.

Future predictions are only probability predictions and do not provide clear evidence. But making policy and business decisions without aiming to accurately predict the future is 'sadly shameful,' said Matheny. 'The methods used by most analytical institutions may be thought to be well-regarded, but they cost hundreds of millions of yen, or even trillions of yen, or more,' Matheny said. 'Organizations that make important national security decisions are not really doing a good job of evaluating them.' Masu.