What if there was a crypto protocol that specialized in arbitration on-chain disputes?

Imagine that every time the prediction markets like the polyme field settled in a controversial way, users had a formal way of appealing through a kind of neutral justice system on the chain. Or if decentralized autonomous organizations (DAOS) could rely on an effective, knowledgeable third party to help them make decisions. Or if insurance contracts could automatically make payments when specific real world events emerged.

It is essentially what Albert Castellana Lluís and his team build with Genlayer, a crypto project that markets itself as a decision -making system or confidence in infrastructure.

“We use a blockchain that has several AIS coordinate and reached an agreement on subjective decisions as if they were a judge,” Castellana, co-founder and CEO of Yeagerai to Coindesk in an interview. “We basically build a global synthetic jurisdiction that has an embedded justice system that doesn’t sleep, it’s superbous, and it’s super fast.”

The demand for such an arbitration project can spike in the coming years with the development of AI agents – sophisticated programs driven by artificial intelligence capable of performing complex tasks in an autonomous way.

When it comes to crypto markets, AI agents can be used in all kinds of ways: for trading in Memecoins, arbitration bitcoin on exchanges, monitoring the security of defi-protocols or providing market insight through an in-depth analysis to quote only a few use cases. AI agents will also be able to hire other AI agents to complete even more complex tasks.

Such agents can spread at an unexpected speed, Castellana said. In his view, most Crypto market participants could control a handful of them by the end of 2025.

“These agents, they work super fast, they don’t sleep, they don’t go to jail. You don’t know where they are. Should they adopt rules against money laundering? Should they have a bank account? Can they even use a visa card?” Said Castellana. “How can we enable quick transactions between them? And how can trust happen in a world like this?”

Thanks to its unique architecture, Genlayer could provide a solution by allowing devices – human or AI – to get a reliable, neutral meaning to weigh in any decision in record time. “Everywhere you would normally have a third party made by a lot of people … We replace them with a global network that gives a consensus between different AIS, a network that can make decisions in a way that is as correct and as impartial as possible,” Castellana said.

Synthetic Court System

Genlayer does not try to compete with other blockchains such as Bitcoin, Ethereum or Solana – or even defi protocols such as Uniswap or Compound. On the contrary, the idea is that any existing crypto protocol should be able to connect to the Genlayer and make use of its infrastructure.

Genlayers chain is operated by ZKSYNC, an Ethereum Layer 2 solution. Its network counts 1,000 validators, each associated with a large language model (LLM) as Openais Chatgpt, Google Bert or Metas Llama.

Let’s say that a market in the polyme field is setting in a controversial way. If the polyming field is connected to Genlayer, users of the prediction market have the ability to raise the problem (or, as Castellana put it, to create a “transaction”) with its synthetic court system.

As soon as the transaction comes in, Genlayer selects five validators randomly to rule over it. These five validators ask an LLM of your choice to find information on the current topic and then vote for a solution. That produces a decision.

But in our example, the polyme users do not necessarily have to be happy with the decision: They can decide to appeal the decision. In which case, Genlayer chooses another set of validators – except this time, their number jumps to 11. As before, the validators send out a decision based on the information they collect from LLMS. This decision can also be appealed, causing Genlayer to select 23 validators for another decision, then 47 validators, then 95, and so on and so on.

The idea is to rely on the Condorcet’s jury, which according to Genlayers Pitch -Tires says that “when each participant is more likely than not making a correct decision, the likelihood of a correct majority result increased significantly when the group becomes larger.” In other words, Genlayer finds wisdom in the crowd. The more validators are involved, the more likely they are at zero on an exact answer.

“What this means is that we can start small and very effectively, but also we can escalate to a point where something very, very difficult they can still get to real,” Castellana said.

The average transaction takes about 100 seconds to process, Castellana said, and the court’s decision finally becomes after 30 minutes – a timeframe that can be extended if more appeals occur. But this means that the protocol can reach a decision on bigger questions for a very short period, day or night, rather than undergoing hard litigation in the real world that can take months or even years.

Looking at incentives

Of course, Genlayers Mission raises a question: Is it possible to play the system? What if all the validators, for example, choose the same AI (eg chatgpt) to solve a given proposal? Wouldn’t that mean that Chatgpt essentially gave up the decision?

Every time you ask a LLM, you generate a new seed, Castellana said, so you get another answer. On top of this, the Validators have the freedom of choosing which LLM should be used based on the current topic. If it is a relatively easy question, it may not be necessary to use an expensive LLM; On the other hand, if the question is particularly complicated, the validator may choose an AI model of higher quality.

Validators may even end up in a situation where they want them to have seen a specific type of question so many times that they can foresee a small model for a specific purpose. “We think there will just be endless new models over time,” Castellana said.

There is a strong incentive for validators to be on the winning side of the decision-making process because they are financially rewarded for the losing side ends up incurred costs associated with using calculation without collecting any reward.

In other words, the question is not whether one’s validator gives a correct answer, but whether it manages to sit with the majority.

Since validators have no idea what other validators are voting, the goal is to use the necessary resources to provide accurate information with the expectation that other validators will also converge about this information – because arriving at the same wrong answer would probably require strict coordination.

And if this gambit is not working, the orange system is ready to kick in.

“If I know I recycle a good LLM and I think other people are using a bad LLM, and that’s why I lost, then I have a pretty big incentive to appeal because I know that with more people, there will be an incentive for them to use better LLMs too,” as other validators want to earn the benefits of a successful app, said Castellana.

The system makes it difficult for validators to collide because they only have 100 seconds to reach a decision and they do not know if they will be chosen to run specific questions. A device would have to check between 33% and 50% of the network to attack it, Castellana said.

Like Ethereum, Genlayer will use a native token for his financial incentives. With a test network that has already been launched, the project must go live at the end of the year, according to Castellana. “There will be a very big incentive for people to come and build things on top,” he said.