# AI and Blockchain Integration Sparks Innovative Projects

The global enthusiasm for AI shows no signs of waning, and this trend is evident within the blockchain industry as well. At first glance, blockchain and AI might seem unrelated, but numerous projects are emerging that combine the two technologies. So, is there really no connection between AI and blockchain?

The answer is that there definitely is a connection. One significant AI concept is federated learning, a subset of machine learning. Machine learning encompasses various techniques, from reinforcement learning to large language models (LLMs) like GPT. Federated learning, a type of machine learning, involves training a model across decentralized networks and devices, using different datasets.

An everyday example of federated learning is the auto-complete function on smartphones. Personal word patterns, common typos, and context-driven word usage are learned individually on each device. However, the actual content used for this learning isn’t sent to a central server; only the information about frequent typos, such as typing ‘ㅋ’ instead of ‘ㄱ,’ is shared.

In federated learning, only the model weights (the learned patterns) are transmitted to a central server, where they are aggregated and used to improve the overall model. This updated model is then deployed back to individual devices, enhancing the auto-complete function. This decentralized data collection method is similar to how blockchain operates. The intersection of federated learning and blockchain has led to the creation of projects like FLock.

FLock is a platform designed to apply federated learning theories, enabling decentralized AI model development, data usage, and learning. It aims to address the centralization of data ownership, which has become a significant issue in AI. In the current landscape, a few companies develop and distribute AI models, using vast amounts of personal information.

Given the rapid focus on development within this new industry, intellectual property and ownership issues have often been secondary and susceptible to opaque and biased operations by single entities. Federated learning, by preserving individual data privacy while enabling model training, brings transparency, trust, and privacy.

FLock offers three primary services: AI Arena, FL Alliance, and AI Marketplace. Here’s how the workflow progresses through these stages:

## AI Arena
AI Arena serves as the platform where new AI models are created and selected for learning. Verified AI developers train the models using traditional methods rather than federated learning. Each developer optimizes the model on their local devices using either personal or public data. Community-requested tasks lead to the selection of a base model that is then further trained and refined. The best-performing and optimized base models proceed to the FL Alliance.

## FL Alliance
FL Alliance takes the base models from AI Arena and further enhances them using federated learning. It ensures data sovereignty, allowing participants to train global models without centralizing the local data. Only model weights, not raw data, are aggregated to form an optimal global model. Thus, it balances data privacy and model performance improvement.

## AI Marketplace
AI Marketplace is the platform where the publicly refined models are deployed and utilized. Data providers and computing resource providers earn revenue based on their contributions. Model owners can continuously fine-tune the models using new data contributions, ensuring ongoing improvement.

Key participants in the FLock process include task creators, training nodes, and validators. Task creators define the desired models and are selected based on their contributions or stakes within the network. This approach ensures a diverse representation and ownership distribution.

Training nodes compete in model training by staking tokens, maintaining network integrity for stability and security. Validators assess the work of training nodes, scoring and submitting results, ensuring accurate and honest validation through a token staking mechanism.

Since FLock’s testnet launch in May 2024, participation has surged. As of September 2024, over 1,400 AI engineers have submitted more than 15,000 AI models to FLock’s network, with over 1.6 million model validations. This has resulted in 37 models being selected as standard models, used in various applications such as AI assistants, trading agents, and health monitoring.

FLock’s success stems from its innovative integration of blockchain and federated learning. Its model demonstrated 95.5% accuracy even with 40% malicious nodes, outperforming traditional centralized federated learning systems in terms of security and performance. The Ethereum Foundation has recognized FLock’s technological prowess, awarding it a 2024 academic grant.

FLock addresses the issues of centralized AI systems and democratizes AI development through a decentralized learning system. By combining federated learning with blockchain, it guarantees data sovereignty while efficiently training AI models, significantly enhancing AI transparency and security.

FLock’s intricate yet essential structure is tailored for stable blockchain operation, especially in integrating blockchain with federated learning within the AI sector. As FLock progresses from its testnet to the mainnet, it remains a critical project to watch.

As the industry continues to release various projects and search for viable business models and platform structures, FLock’s challenge could yield significant results and set a benchmark in both blockchain and AI fields.