Binance Research Explores the Future of Decentralized AI

A new report from Binance Research dives into the intersection of decentralized blockchain technology and artificial intelligence, where it finds the world of decentralized artificial intelligence (DeAI).

What DeAI is not, the report stresses, is a decentralized AI agent that relies upon centralized Large Language Models of the tech world’s AI giants.

“The DeAI movement,” it says, “aims to build networks and infrastructure for creating AI models in a decentralized way.”

The report also quotes Matthew Sigel, head of digital asset research at VanEck, who argues there is a significant likelihood “that public blockchains are the key to unlocking the widespread adoption of artificial intelligence and that AI applications will be crypto’s raison d'être.”

The reason, Sigel says, is that blockchain can provide four elements needed for the foundation of AI, “transparency, immutability, clearly defined ownership properties, and an adversarial testing environment.”

That means blockchains can improve upon AI systems in the same way that they have revolutionized financial systems, claims Binance macro researcher Joshua Wong.

What is Decentralized AI?

Decentralized AI can refer to a system that uses multiple AIs trained by different groups to develop the best response to a user’s input. As opposed to a standard centralized AI, which uses a single input.

These models could include specialized AIs, competing to provide the best output and driving costs down significantly.

A DeAI would have three layers: a top layer with public-facing AI agents, smart contract auditing and identity/privacy tools. The middle layer would have the open source Large Language Models and form the core of the DeAI. The bottom layer would be infrastructure, with decentralized compute resources and decentralized storage.

DeAI Infrastructure

There are three base AI components: data, compute and algorithms, Wong noted.

Data is the most essential part of building AI, and outside of decentralized storage, it is generally centralized by tech giants AWS, Google Cloud and Microsoft Azure. Decentralized storage solutions like Filecoin, built on the Interplanetary File System (IPFS), and Arweave can provide storage.

Compute power is essential, the resource that allows AI models to be trained on that data.

By transforming computational power into a publicly accessible commodity, “projects like Render and AIOZ are among several initiatives developing distributed networks to coordinate the decentralized contribution of compute power for the purpose of AI training,” Wong said.

Algorithms are the hardest part of the DeAI puzzle, as human engineers are required. Wong notes that ChatGPT is estimated to have spent $1.5 billion on labor annually.

Wong looks to Bittensor, a decentralized, open-source platform that allows users to create, train and share AI models. A key, Wong says, is that Bittensor’s Proof of Intelligence consensus mechanism, “evaluates and rewards participants based on the quality and utility of their AI models, thereby promoting the development and sharing of high-quality models.”

Together those three parts may offer a road to a decentralized, Web3 alternative to the tech giant AIs that currently dominate.