FLock.io, a startup that offers full-stack decentralized training and fine-tuning of artificial intelligence models, today announced it raised $6 million in seed funding to build a blockchain-based platform enabling the co-creation of AI models.

The round was led by Lightspeed Faction and Targus Capital with participation from DCG, OKX Ventures and Volt Capital. The company said that it intends to use the new funds to continue the development of its machine learning and decentralized training platform that places AI governance in the hands of users and avoid centralized corporate oversight.

“With FLock, now anyone can contribute to AI development by offering compute, data,or training scripts; the barrier to participation is much lower,” said Chief Executive Jiahao Sun. “This results in community-owned models built by the many, not just the few, with data contributors being fairly rewarded.”

FLock provides a decentralized machine-learning platform that incentivizes users to supply data in a decentralized manner to train and fine-tune AI models. Using this approach, the company said, it enables scalability, reduces costs and enhances security.

The platform uses a concept known as “federated learning,” a method of training AI models where training data remains decentralized on individual devices or servers, instead of being sent to a centralized server. Using this model, user-submitted data is not exposed to other users or the internet at large, which reduces the likelihood that private data could be leaked or potentially misused.

Unlike large-scale centralized commercial and open-source model training, FLock’s implementation does decentralized learning, local models are trained on the edge nodes and then synchronized with a global model. Users are rewarded with “FLock points” for participating in co-creation and collaboration for AI model customization, knowledge contribution, evaluation and quality assurance. These points open up potential exclusive rewards in the FLock ecosystem, including potential crypto airdrops, benefits and more in the future.

FLock uses zero-knowledge proofs with federated learning data alongside homomorphic encryption, which allows the data to be shared without revealing the information to other parties submitting training data to the model. By combining blockchain technology, another decentralized capability that robustly tracks transactions. It also allows for secure multiparty computation when data is accessed. That prevents data from being shared when a user does not permit it.

According to FLock, the platform is also model-agnostic. Developers can choose between simple statistical models or complex neural networks – including large language models such as OpenAI’s ChatGPT and art-generating models similar to Stable Diffusion.

“The rapid growth of AI has been met with increasing data privacy and regulatory concerns,” said Clara Boh, deal partner at Lightspeed Faction. “By combining blockchain and federated learning technologies, FLock seeks to solve this and unlock broader participation in AI model training.”

In the future, FLock plans to enable decentralized hosting by allowing users to contribute to a distributed computing pool, the hosting of vector databases and inference of models. That will greatly improve the scalability and capability of the storage and computational environment for AI models.

FLock itself is designed to be modular and can also be plugged into decentralized hosting networks – such as io.net Inc., Gensyn and Ritual – which would make AI models accessible to blockchain-based applications.

Image: Production Perig/Shutterstock