Auckland based FaceMe, a company creating AI-powered ‘digital humans’ to put a human face on customer service chatbots, will work with Auckland based startup Tatau to increase the speed and scope of the GPU-based processing that underpins its AI software.
FaceMe says Tatau’s "distributed supercomputing resources" for AI model training will give it access to greater compute capacity at a more commercially viable price point, enabling it to accelerate its innovation.
FaceMe's head of product, Victor Yuen, said FaceMe's digital humans were engineered to emulate the behaviour of empathetic and socially intelligent people and devleopment was highly compute-intensive.
"The research and development involved with such an undertaking present a heavy processing load, using huge amounts of data and employing complex neural networks that need to be constantly improved upon.”
Tatau will supply FaceMe with computing power through its platform by harnessing GPUs owned by its suppliers, globally distributed companies that operate cryptocurrency mining and high-performance computing operations.
Also Tatau and FaceMe say they will work together to create a template for applying distributed computation techniques to large-scale commercial AI problems.
In August 2018 Tatau launched what it said was the first commercially-viable decentralised supercomputer pre-configured to perform deep learning and other advanced AI operations by harnessing existing GPU hardware, typically found in crypto mining and gaming infrastructure that was often underused or unprofitable.
"Under the hood, Tatau uses a blockchain to handle transactions in a secure and transparent way," it said. "The Tatau ERC-20 token is the exclusive economic unit on the platform and is used to buy and sell compute, and can be 'staked' to reserve future jobs. A proprietary system verifies suppliers' work, guarding against fraud, and ensuring full privacy."