Author: Lao Bai

From the movie "Mercy" to real-world reflections

This year, a popular sci-fi movie "Mercy" tells the story of a detective accused of murdering his wife who must prove his innocence to an AI judge within 90 minutes, or he will face immediate execution. The film is highly polished and definitely worth watching.

Last month during TOKEN2049, I attended a BBQ hosted by JE Labs and met a founder of GenLayer. GenLayer is a decentralized AI consensus protocol positioned as the "adjudication layer" for the agent era. After listening to his introduction, I immediately thought of this movie: What would the world look like if "judgment" were handed over to AI?

Upon further contemplation, I increasingly believe this direction has strong practical feasibility. Especially in the Agent Economy that both the AI and Crypto industries envision, as well as in the more distant silicon-based civilization, the forms in which "law" and "judgment" will exist is a core topic worthy of serious discussion.

Technological waves are reshaping finance and law

When talking about the United States, the public often has a deep impression of three pillar industries: first, the financial sector represented by Wall Street and NASDAQ; second, the legal industry represented by lawyers, judges, and juries; third, the tech industry represented by software, AI, and the seven giant companies of U.S. stocks (the Crypto industry can also be included here).

In the current wave of global technological advancement, the tech industry is comprehensively reshaping these two traditional strongholds of finance and law.

In the financial sector, stablecoins are undoubtedly the most significant institutional and technological innovation in recent years, and the public chain networks that support stablecoins form a new payment base. Practitioners in the Crypto industry are the first to use stablecoins for various remittances, transfers, and commercial settlements, becoming the first batch of "pioneers."

The current consensus in the industry is that when the Agent Economy truly arrives, its financial system will most likely not follow the old structure designed for interpersonal society by Visa, Mastercard, or traditional commercial banks, but will firmly establish itself on Crypto and stablecoins: global instant transfers and micropayments are the rigid foundational demands of AI agents.

Agent economy calls for a new legal paradigm

Since finance has been reshaped, how will the legal field evolve?

It is well-known that the legal industry in the U.S. is highly developed, with lawyers consistently ranking among the highest income groups. Works such as "Better Call Saul" and "12 Angry Men" are widely recognized. However, the prosperity of the industry and the high incomes of practitioners come with the other side of the coin: the high barrier of litigation costs.

In reality, the hourly rate of a typical commercial litigation lawyer is often several hundred dollars, and partners in large law firms can charge up to $1,000 to $2,000 per hour. Once a case enters the formal trial phase, costs often start at several hundred thousand dollars, and complex cases easily exceed a million dollars, with a timeline of 1 to 2 years being quite common.

Therefore, it is not uncommon in reality to see a fraud lawsuit involving millions of dollars, where after a two-year tug of war, the winning party may end up with only tens of thousands of dollars, with the majority of the compensation funds used to pay high attorney bills.

This is merely the state of human society. If this dispute resolution process is placed within the context of the Agent Economy and even the silicon-based civilization, its inefficiency and outdated structure may be even worse than that of traditional bank card organizations.

In human society, a significant number of disputes have been unable to enter legal relief channels due to high judicial costs, such as minor disputes and small insurance claims. In the era of Agent Economy, transactions will only become smaller and more frequent: one Agent may only spend tens of dollars to hire another Agent to write a piece of code, conduct industry research, or book a flight.

Once disputes arise during collaboration over issues such as "whether the delivery meets acceptance criteria" or "whether the flight constitutes a serious delay," agents cannot afford to spend tens of thousands of dollars hiring lawyers for a two-year lawsuit, but must complete a fair ruling within minutes at an extremely low cost of a few dollars.

GenLayer's core solution: Optimistic democracy adjudication layer

What GenLayer constructs is such an adjudication layer for the agent era. Its core operational mechanism can be summarized as “AI preliminary review + multiple-party review + on-chain appeal,” which includes the following four-layer design:

First, randomly form a multi-model committee: The protocol randomly selects 5 validators from the verification nodes to form an adjudication committee, with each validator running different large language models locally, covering heterogeneous model matrices such as GPT, Claude, Gemini, Llama, etc.;

Second, the Leader node executes intelligent judgment: One of the validators takes on the role of Leader for this adjudication, responsible for executing the intelligent contract, autonomously reading webpage contexts, reviewing relevant on-chain and off-chain evidence, and delivering adjudication judgments;

Third, independent review and equivalent semantic principle: The remaining validators independently run their respective large models for review. If the wording of the answers differs but the semantic essence is consistent, it is also regarded as reaching a consensus (equivalence principle). The voting process uses a "encrypt first, reveal later" cryptographic mechanism to thoroughly prevent collusion among nodes;

Fourth, stepwise deposit appeal mechanism: Any party with objections to the preliminary results can pay a deposit to initiate an on-chain appeal, increasing the size of the adjudication committee stepwise from 5 to 7, 11, or even more, with the economic cost of malicious behavior and disputes increasing exponentially.

This entire governance mechanism is named "Optimistic Democracy," which bears a philosophical resemblance to "Optimistic Rollup" in Layer 2 scaling. In conventional simple judgment scenarios, the cost of a single intelligent adjudication is only about $1.

Why must judgments be made collectively by a multi-model committee?

As for why not directly submit to a single centralized model (like a single GPT-4) for judgment? The reason is very clear: a single AI not only has the potential for hallucinations and errors but also faces the risk of being misled by specific adversarial samples.

This design aligns closely with the basic logical system of the American judicial system: a jury collectively determines whether "core facts have occurred," while the judge is responsible for ensuring that "the legal process is compliant" and "sentencing discretion is determined."

Earlier this year, a16z published a long research article titled "How AI judges can scale prediction markets: The case for locking LLMs into the blockchain to resolve the hardest contracts." The core proposal suggested using large language models as on-chain oracles for prediction markets, which essentially resembles an AI adjudication structure similar to GenLayer.

I previously commented in a public post that we should not rely on a single large model oracle, but rather a decentralized form of voting by multiple heterogeneous LLMs serving as a review committee. Remarkably, within less than a year, this legal vision for future silicon-based civilization has already welcomed genuine engineering practices in the Web3 field.