Historically, only trained experts, such as back-end developers, had the tools to configure AIs for users. crystal is changing the game. Its simple user interaction and Self-Service Console makes complex AI technology accessible to anyone. crystal is a machine training machine. Using a limited amount of its users’ usage data, crystal can automatically self train and tailor functions to users’ needs — achieving maximum output with minimum input. No experts are required. Seamlessly, crystal puts users in charge.
crystal’s AI engine (also known as a Business Knowledge Graph) is applicable to any industry. This is due to transfer learning, meaning that crystal transfers and uses its knowledge base to and for multiple applications. For example, if an insurance company is using crystal, it will transfer and use its pretrained knowledge base and then incorporate only new insurance-related training or language on top of its base.
crystal uses another technique, called incremental learning, to maximize usage of its knowledge base. For many AIs, each upgrade requires the entire system to be retrained from scratch and then the additional data sets to be added. With incremental learning, crystal only adds new data on top of its consistent base.
It’s like adding new words to your vocabulary, as opposed to learning the language from scratch.
crystal’s quick setup, transfer learning and incremental learning result in training and operations that, compared to most AIs, require less data, less time, and less computing power. Furthermore, crystal is trained with central processing units (CPUs) as opposed to an industry standard of graphic processing units (GPUs) wherever possible.
CPUs are ten times less expensive and require up to four times less energy as GPUs.
“We cannot ignore the environmental impact of GPUs,” said Michele Pini, Senior Vice President of Technology at iGenius, the developers of crystal, “so we take care to use CPUs whenever we can.”
The result of all of these choices is a truly sustainable AI.