Kait* is a counterexample to the Delphi method, while Allora takes the Delphi method to the extreme.
┈┈➤ Let's first understand — what is Delphi?
╰┈✦ Ancient Greek Legend
Delphi, originating from the Oracle of Delphi in ancient Greece, is located at the Temple of Apollo in the ancient city of Delphi. It is said to be the place where the priestess Pythia of Apollo delivered oracles.
Ancient Greeks, including national leaders and ordinary people, would come here for consultation when making significant decisions.
Thus, the Delphi method is named after it.
╰┈✦ The Theory and Practice of the Delphi Method
The Delphi method, derived from the Oracle of Delphi, is a structured approach primarily used for analysis, judgment, and forecasting.
In simple terms, the Delphi method involves a group of experts with authority in a specific field working collaboratively to achieve a goal. Therefore, the Delphi method is also known as the expert survey method, created and implemented by the RAND Corporation in the United States in 1946.
In practice, experts are usually anonymous to each other. First, through questionnaires and surveys, experts make their analyses, judgments, or predictions separately.
A simpler operation involves summarizing the experts' opinions to reach a final conclusion.
A more complex operation allows experts to see the opinions of other experts after each round of surveys, enabling them to adjust their own opinions based on this feedback, leading to a consensus after several rounds of coordination.
In fact, there are some opposing voices regarding the naming of Delphi, as it originates from ancient Greek legend and seems less scientific, while the Delphi method is actually a scientific process.
However, the name Delphi has persisted, and even in academia, it is more commonly referred to as the Delphi method rather than the expert survey method, as the latter fails to demonstrate the anonymity among experts.
╰┈✦ Key Features of the Delphi Method
◆ Predictive: Primarily used for insights and predictions about the future.
◆ Authoritative: Involves participation from authoritative experts.
◆ Anonymity: Experts remain anonymous to each other, ensuring that authority does not interfere with others' judgments.
◆ Aggregation: Derived from the collective wisdom of experts rather than focusing on individual authoritative opinions.
◆ Structure: Organized and executed according to a certain structured method.
┈┈➤ The Counterexample of Delphi — Kait*
Kait* initially calculated scores primarily based on ICT interaction.
The original intention of Kait* was to apply the Delphi method, treating ICT as experts and using the interaction of ICT with articles as a basis for judging the quality of the articles.
However, Kait* ultimately discovered that, on one hand, the level of ICT was still uneven and not necessarily expert-level, which did not align with the authority aspect of the Delphi method.
On the other hand, since ICT serves as both judges and competitors, this led to the phenomenon of ICT interaction, affecting the objectivity of the scoring, which again violated the anonymity of the Delphi method.
Subsequently, Kait* made a series of algorithm adjustments… The merits of the adjusted algorithms are not discussed here, but Kait*'s application of the Delphi method is indeed a counterexample.
┈┈➤ Why is Allora considered the Delphi of the AI world?
At the level of AI products, Allora does not participate in the training work of the model layer. Allora is similar to the AI Agent layer, primarily involved in the application layer of AI, utilizing AI models to execute consumer tasks. However, Allora fully leverages the functions and advantages of the Delphi method, making it significantly different from ordinary AI Agents.
╰┈✦ AI Reasoning and Predictive Decision-Making
Allora is a product system that utilizes AI reasoning to make predictions and decisions, aligning with the predictive nature of the Delphi method.
For example, Allora can provide consumers with enhanced DeFi strategies. Specifically, by predicting cryptocurrency prices and the liquidity of DeFi products, it can infer changes in returns, risk factors, etc., further enabling consumers to achieve smarter asset allocation and maximize returns.
╰┈✦ Achieving Authority through AI Model Filtering
When executing any theme within the Allora ecosystem, the algorithm effectively filters the participating AI models for validity, which aligns with the authority aspect of the Delphi method.
First, it matches the historical reasoning and prediction results of AI models with the theme. For example, if a theme involves predicting BTC prices, Allora will compile the historical predictions of various AI models regarding BTC prices and filter out those models that are not proficient in BTC price prediction.
Second, it filters out anomalous data from the current task execution. For instance, if a theme involves the federal funds rate and a model predicts an interest rate hike by the Federal Reserve in October 2025, Allora will filter out that prediction data.
Third, it conducts contextually relevant filtering of AI models. For example, in a bullish market environment, when predicting prices, Allora will prioritize trend-following models. In a bearish market environment, it will prioritize mean-reversion models. The goal is to select AI models that best fit the contextual environment.
╰┈✦ Anonymity of AI Model Details
In the Allora ecosystem, the IDs and addresses of the specific Worker nodes are public. However, the details of the AI models that actually complete the reasoning and prediction tasks are anonymous.
For example:
The developers or institutions behind the AI models are anonymous.
The model architecture, algorithms, and specific source code of the AI models are anonymous.
The training data and implementation details of the AI models are anonymous.
Therefore, in Allora's operations, all AI models are independent and do not interfere with each other. This meets the requirement for anonymity in the Delphi method.
╰┈✦ Aggregation of AI Reasoning Results
Based on the filtering of AI models, Allora aggregates the filtered model prediction results according to task characteristics, selecting different algorithms for aggregation, such as median, weighted average, etc. This aligns with the aggregation aspect of the Delphi method.
╰┈✦ Intelligence of Aggregation Algorithms
First, while filtering AI models, Allora sets different weights for different AI models based on their historical predictions and algorithms, aggregating the prediction results according to these weights. Moreover, the weights of the models are dynamically adjusted rather than fixed. For example, if a certain AI model's recent ETH price prediction accuracy improves, its weight may be adjusted accordingly.
Second, during aggregation, Allora considers the complementary advantages of the models. For instance, when predicting BTC trends, it may select several AI models from technical analysis, on-chain data, social media sentiment, macro and policy analysis, and quantitative strategies. Thus, Allora can provide more comprehensive AI reasoning and prediction results.
This aligns with the structural aspect of the Delphi method.
┈┈➤ In Conclusion
When we view AI models as intelligent agents with life, Allora selects advantageous AI models among many based on the reasoning tasks of specific themes, akin to finding authoritative experts in a certain field. Moreover, these AI "expert" models remain anonymous to each other. Based on this, Allora uses dynamic weights and complementary advantages to aggregate the prediction results of these AI "expert" models, achieving structured aggregation. Allora takes the Delphi analysis and prediction method to the extreme.
Furthermore, Allora's aggregation employs AI algorithms. Although Allora's aggregation itself is not an AI model, it utilizes AI algorithms. Therefore, Allora's aggregation is characterized by learnability and evolution, which is key to Allora's intelligence. Allora not only conforms to the structural aspect of the Delphi method but even dynamically surpasses it. The structured adjustments of the Delphi method rely on human intervention, while Allora's intelligence can optimize the aggregation algorithms more accurately and in real-time, thus providing Web3 users with predictions and strategies that are more accurate, comprehensive, and timely.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
