The Trust Layer of AI: How the Mira Network Addresses AI Hallucinations and Bias Issues

Recently, a public test network named Mira has officially launched, aiming to build a trustworthy foundation for artificial intelligence. This raises an important question: why does AI need to be trusted? How does Mira address this complex issue?

When discussing AI, people often focus more on its powerful capabilities. However, an interesting but often overlooked issue is that AI can have "hallucinations" or biases. The so-called "hallucination" of AI simply means that AI sometimes "talks nonsense", seemingly stating unfounded information with certainty. For example, when asked why the moon is pink, AI might provide a series of explanations that seem reasonable but are actually groundless.

The appearance of such "hallucinations" or biases in AI is related to some current technological paths in AI. For example, generative AI achieves coherent and reasonable outputs by predicting the "most likely" content, but this method can sometimes be difficult to verify for authenticity. Additionally, the training data itself may contain errors, biases, or even fabricated content, all of which can affect the quality of AI outputs. In other words, AI learns human language patterns rather than the facts themselves.

The current probability generation mechanism and data-driven model almost inevitably lead to AI hallucinations. This biased or hallucinatory output, if limited to general knowledge or entertainment content, may not cause serious consequences temporarily. However, if it occurs in highly rigorous fields such as healthcare, law, aviation, and finance, it could have significant impacts. Therefore, addressing the issues of AI hallucinations and biases has become one of the core challenges in the development process of AI.

Currently, there are several methods that attempt to address this issue. Some use retrieval-augmented generation techniques, combining AI with real-time databases to prioritize verified facts. Others incorporate human feedback, correcting the model's errors through manual labeling and supervision.

The Mira project is an attempt to address the issues of AI bias and hallucination. It aims to build a trust layer for AI, enhancing its reliability. So, how does Mira achieve this goal?

Mira's core concept is to validate AI outputs through the consensus of multiple AI models. It is essentially a verification network that leverages the collective intelligence of multiple AI models to assess the reliability of AI outputs. More importantly, Mira employs a decentralized consensus mechanism for validation.

This decentralized consensus verification is a strong point in the crypto field, while also fully leveraging the advantages of multi-model collaboration to reduce bias and illusion through collective verification models. In terms of verification architecture, the Mira protocol supports the conversion of complex content into independently verifiable assertions. These assertions require node operators to participate in the verification process. To ensure the honesty of node operators, Mira employs cryptoeconomic incentives and penalty mechanisms.

Mira's network architecture includes content transformation, distributed verification, and consensus mechanisms. First, the system decomposes candidate content into different verifiable statements, which are distributed to nodes for verification, and then the results are aggregated to reach consensus. To protect customer privacy, the statements are distributed to different nodes in a randomly sharded manner.

Node operators are responsible for running validator models, processing statements, and submitting validation results. Their motivation for participating in validation comes from the potential to earn rewards. These rewards stem from the value created for clients, which is reducing the error rate of AI. In fields such as healthcare, law, aviation, and finance, reducing error rates can generate immense value, thus clients are willing to pay for it. To prevent node operators from exploiting the system, nodes that consistently deviate from consensus will have their staked tokens reduced.

Overall, Mira provides a new solution for the reliability of AI. It builds a decentralized consensus verification network based on multiple AI models, bringing higher reliability to customers' AI services, reducing AI bias and hallucination, in order to meet the demands for higher accuracy and precision. At the same time, it also brings benefits to network participants. In short, Mira is trying to construct a trust layer for AI, which will help promote the in-depth development of AI applications.

Currently, users can participate in the Mira public testnet through the Klok application. Klok is an LLM chat application based on Mira, allowing users to experience verified AI outputs and have the opportunity to earn Mira points. Although the future use of the points has not been announced, this undoubtedly provides users with a chance to personally experience the AI verification process.