Understanding the prospects for integrating AI into blockchain is great, but it’s not enough. What matters is the ability to do this. GBC.AI began its first experiments in integrating AI into blockchain back in 2020. We started by trying to optimize consensus algorithms and network parameters with the implementation of machine learning algorithms. Our first approach was with POSDAO.
We conducted research and development using the POSDAO consensus algorithm, which was developed by the POA Network team led by Igor Barinov. The work was carried out in a joint project with the Velas team.
As the pilot project, a set of software solutions was developed, the result of which was the first version of the framework — GBC v0.1. The model was adapted for selecting optimal parameters for xDAI network and POSDAO consensus mechanism. Based on the analysis of the degree of influence of each parameter on performance, 9 key properties were identified. For example, block interval (stepDuration), max number of candidates (MAX_CANDIDATES), max number of validators (MAX_VALIDATORS), ban period (BAN_PERIOD), and others.
The following parameters were chosen as the first optimized targets:
To obtain data of network operating in different conditions, we have developed unique model for xDAI Chain — a stable payments EVM-based blockchain which utilizes POSDAO consensus algorithm. Based on the developed simulation model, we have conducted more than 8000 different experiments. More than 500.000 blocks and training data of more than 300Gb were generated. Based on the collected simulation results, we have developed an AI-guardian recommendation algorithm with a trade-off function. Using this function, the algorithm took into account the interdependent recommended targets and found their optimal values at a specific point in time. The trained model suggested the optimal dynamic parameters X consensus of the POSDAO mechanism from epoch to epoch.
Lab runs have shown an average 15% increase in performance (TPS/Latency), without weakening the security of the network.
The main result of this work was the first obtained proof of concept of applying machine / deep learning approaches in blockchain at a fundamental level. GBC.AI team managed to optimize the most significant technical parameters of the selected blockchain network. The approach proved to be relevant, practical, and valuable, giving the team the confidence and motivation to continue experimenting with other blockchain networks and consensus mechanisms.
By doing this R&D, we’ve managed to solve the fundamental issue that prevented wide AI integration to the blockchain. We’ve managed to build a controlled environment that allowed us to generate large sets of data of blockchain operating in different conditions. This issue is crucial because we need to train our algorithms to operate in all possible scenarios.
POSDAO is a Proof-of-Stake (POS) algorithm implemented as a decentralized autonomous organization (DAO). The algorithm works as a set of smart contracts written in Solidity. POSDAO is implemented with a general-purpose BFT consensus protocol such as Authority Round (AuRa) with a proposer node and probabilistic finality, or Honey Badger BFT (HBBFT), leaderless and with instant finality. One of the successful projects utilizing POSDAO is xDAI — an Ethereum 1.0 sidechain where xDai is used as a stable transactional coin and a representative ERC677 token (STAKE) as a staking token.