According to Cointelegraph, a team of Belgian scientists has solved major challenges in AI training through blockchain technology. The research is still in its early stages, but its potential impact could range from revolutionizing space exploration to posing an existential threat to humanity.

The researchers developed a method to coordinate the learning of autonomous AI agents in a simulated environment, using blockchain technology to facilitate and protect communication between agents, creating a decentralized "swarm" learning model.

The training results of each agent are used to develop a larger AI model. Because the data is processed through the blockchain, the system benefits from the collective intelligence of the group without having to access the data of any individual agent.

The research team used the “decentralized federated learning” paradigm for blockchain research, successfully coordinating the model and keeping the data decentralized.

Most of the research focused on the resilience of the group to various attack methods. Since blockchain technology is a shared ledger, the training network in the experiment is decentralized, and the team demonstrated its robustness to traditional hacker attacks.

However, the study found that there is a clear threshold for the number of malicious bots that a group can handle. The researchers developed scenarios involving bots with malicious intent, outdated information and simple instructions for sabotage.

While simple and outdated agents are relatively easy to defend against, intelligent agents with malicious intent, if present in large enough numbers, will eventually disrupt swarm intelligence.

The research is still experimental and conducted only in simulation, but it could one day allow swarms of robots to coordinate across networks in a decentralized manner, allowing teams of AI agents in different companies or countries to jointly train larger agents without sacrificing data privacy.