While much of the discussion today around artificial intelligence focuses on virtual assistants and chatbots, Mark Hammond, co-founder and CEO of Bonsai, focuses on a completely different set of issues, around the operationalization of AI.
“The company was built on the question: ‘How do you make developers successful using AI?’” Hammond asked. The company’s middleware AI platform was designed to make programming and management of AI models more accessible to developers and enterprises by abstracting away the complexity of artificial intelligence, he said.
The company’s new Early Access Program helps customers through the normal hurdles implementing AI as well as orienting them to the differences in working with AI, Hammond explained. The program is a full immersion, with Bonsai engineers embedding with existing customer development teams.
Hammond wants to utilize AI models to enhance real-time decision support for multidimensional, industrial systems. Think of supply chains, robotics, factories, warehouse operations, HVAC, or oil exploration.
For example, a supply chain might have autonomous trucks that require routing depending on the weather, location or other dependencies. “Is this something where the company needs to make a controlled system or optimize a system that interacts with the physical world?” he said. If so, Bonsai might be worth a look, he argued.
The use cases where there is interaction with the physical world are well suited for machine teaching-centric approach, Hammond said. But when you model that, it fundamentally changes interaction with the physical world. You’re not just moving around a data set.
At scale, such systems quickly outstrip the capabilities of generic AI solutions. But most enterprises lack the resources for coming up-to-speed on the complexities of machine learning fast enough to build application-specific AI models, Hammond said.
The point of the Early Access Program is not just to get an AI system up and running, but to educate the customer team about the differences in working with AI, and to leave them able to run and manage their system on their own, he explained.
Developers interact with Bansai platform in three phases: build, teach, and use.
The first step, outlined in the introductory video above, is to build a high-level model called a “brain.” On-site training and workshops cover the Bonsai product, underlying technology and tools, and relevant AI techniques.
According to Hammond, it’s critical to include subject matter experts (SMEs) in the training workshops and help build the model. SMEs have business knowledge that doesn’t come from data, he said.
These joint sessions include scoping, design, and development of AI models for customized use cases for the industry and the particular business. The brain is built using a proprietary Bonsai language, appropriate training sources like data or simulations, and, of course, specific use cases from the company.
Next, they leverage Bonsai’s AI engine to train the brain. For example, Bonsai uses simulations from Mathworks’ products Matlab and Simulink. Matlab is focused on teaching numeric combinations and Simulink has packages to model mechanical systems, both useful for AI programs interacting with physical objects.
“We get a lot of re-use by tying in simulation models,” said Hammond.
For the last step, they connect the brain to the customer’s existing hardware or software application by using the Bonsai-provided libraries, just like you would connect a database to your application.
The brain sits next to the existing stack and becomes a streaming analytic system with data streaming in real time.
Siemens is part of early access program, and has done a lot of work on building a proof-of-concept to add AI to its business equipment, said Hammond, who has added that the work to date has been “impressive.”
At the end of the day, Hammond said about marrying human intelligence with artificial intelligence, “You need both if you’re going to solve real-world problems.”
Feature image via Pixabay.