AI Trust Layer: How the Mira Network Addresses AI Bias and Hallucination Issues

Recently, a public beta version of a network called Mira was launched, raising industry concerns about the credibility of AI. The goal of the Mira network is to build a trust layer for AI, addressing potential biases and "hallucination" issues that may arise during AI usage. So, why does AI need to be trusted? How does Mira address this issue?

When discussing AI, people often focus more on its powerful capabilities. However, the issue of "hallucinations" or biases in AI is often overlooked. The so-called "hallucinations" of AI, simply put, refer to the fact that AI sometimes "makes things up" and speaks nonsense seriously. For example, when asked why the moon is pink, AI might provide a seemingly reasonable but completely fabricated explanation.

The "hallucinations" or biases of AI are related to some current technological paths of AI. Generative AI achieves coherence and rationality by predicting the "most likely" content, but sometimes it cannot verify authenticity. Furthermore, the training data itself may contain errors, biases, or even fictional content, all of which can affect the AI's output. In other words, AI learns human language patterns rather than the facts themselves.

The current probabilistic generation mechanisms and data-driven models almost inevitably lead to the emergence of AI hallucinations. While in general knowledge or entertainment content, such biased or hallucinatory outputs may not cause immediate consequences, in highly rigorous fields such as healthcare, law, aviation, and finance, they can have serious impacts. Therefore, addressing AI hallucinations and biases has become one of the core issues in the development of AI.

The Mira project is dedicated to addressing the issues of AI bias and hallucination, building a trust layer for AI, and enhancing the reliability of AI. So, how does Mira achieve this goal?

Mira's core approach is to validate AI outputs through the consensus of multiple AI models. Essentially, Mira is a verification network that ensures the reliability of AI outputs through decentralized consensus verification. This method combines the strengths of decentralized consensus verification, which is well-suited for the cryptographic field, and the advantages of multi-model collaboration, reducing bias and hallucinations through a collective verification model.

In terms of verification architecture, the Mira protocol supports converting complex content into independent verification statements. Node operators participate in validating these statements and ensure the honesty of node operators through an incentive/penalty mechanism based on cryptoeconomics. Different AI models and decentralized node operators collaborate to ensure the reliability of the verification results.

Mira's network architecture includes content conversion, distributed validation, and consensus mechanisms. First, the system breaks down the candidate content submitted by the client into different verifiable statements and distributes them to nodes for validation. After the nodes determine the validity of the statements, the system aggregates the results to reach a consensus and returns the results to the client. To protect client privacy, the statements are distributed to different nodes in a randomly sharded manner.

Node operators earn rewards by running validator models, processing claims, and submitting validation results. These rewards come from the value created for clients, specifically by reducing the error rates of AI across various fields. To prevent nodes from responding randomly, nodes that continuously deviate from consensus will have their staked tokens reduced.

Overall, Mira provides a new solution for achieving the reliability of AI. By building a decentralized consensus verification network based on multiple AI models, Mira aims to bring higher reliability to customers' AI services, reduce AI bias and hallucinations, and meet the demands for higher accuracy and precision. This approach not only provides value to customers but also brings benefits to participants in the Mira network, with the potential to promote the in-depth development of AI applications.

Currently, users can participate in the Mira public testnet through Klok (a Mira-based LLM chat application), experience verified AI outputs, and have the opportunity to earn Mira points. Although the future uses of the points have not been announced, this undoubtedly provides users with a chance to personally experience the AI trust layer.