AUS
NZ
UK
CA
USA
IND
Researchers in Belgium are exploring a groundbreaking approach to artificial intelligence (AI) through a blockchain-based, decentralized training method. This pioneering research, still in its early stages, may have profound implications for various fields, from space exploration to critical privacy and security applications.
The team has developed a system that coordinates learning among individual autonomous AI agents using blockchain technology. This approach creates a decentralized “swarm” of learning models, where the agents communicate securely and independently. The results from each agent's training are integrated to form a larger, more robust AI model. Importantly, this method leverages {blockchain} secure data handling capabilities, ensuring that the collective intelligence is utilized without directly accessing individual agents' data.
Traditional machine learning models, such as those used in chatbots like OpenAI’s ChatGPT, typically rely on centralized databases to store and manage training data. This centralized approach can pose challenges in applications requiring ongoing autonomous learning or where data privacy is crucial. The Belgian research team has applied a technique known as “decentralized federated learning” to address these issues. Their approach has successfully coordinated the learning models while maintaining data decentralization.
The research also focused on enhancing the security of the swarm against physical and cyber threats. By using blockchain’s shared ledger system in a decentralized network, the team demonstrated resilience against traditional hacking methods. They tested the swarm's defenses against various malicious agents, including those with outdated information and those specifically designed to disrupt the network.
The findings indicate that while simple or outdated agents can be effectively managed, smart agents with malicious intentions could compromise the swarm’s collective intelligence if they infiltrate in significant numbers. Although the research is still experimental and conducted through simulations, it suggests a future where decentralized AI swarms could collaborate across different entities, enhancing both data privacy and operational efficiency.
This innovative approach to AI training and security may soon lead to significant advancements in how autonomous systems are developed and deployed.
