Microsoft wants to make sure your AI-powered factory can go the distance.
The company filed a patent application for AI agent design for “long-term reliability and stress testing.” Microsoft’s system aims to test the durability of a machine learning agent that operates a control system in charge of mechanical or electrical components, such as an industrial machine.
Essentially, the system tests how long a machine learning agent can go before going haywire by simulating an accelerated use environment to test its capabilities.
Microsoft’s system first analyzes the mechanical components and the control system of the machine, checking factors such as projected failure rate and hours of continuous operation. Then the system derives the parameters that the machine learning agent uses to interact with this control system, essentially looking at what actions the agent takes and how often it takes them.
Using the lifespan of the control system’s components, and the “what” and “how often” factors of the machine learning agent, this system creates an “accelerated test procedure.” This is to test how reliable the AI agent will be at controlling this machine in the long term, even as the components of the machine itself ages. From this, Microsoft’s system comes up with a long-term “reliability score” for the machine learning agent under different stress levels and conditions.
For example, if Microsoft wants to deploy a machine learning agent to control a conveyor belt, this system would test how long that agent could successfully control it despite the conveyor belt itself aging with use.
Microsoft has sought patents for myriad AI-based knick-knacks, from email tone detection tools to AI backpacks to a computer vision-assisted TSA checkpoint. And its patent activity follows suit with its public commitment to AI, touting AI-enabled products like 365Copilot and Bing Chat, partnering with OpenAI several times over, and potentially even looking at nuclear reactors to power the data centers that keep its AI running.
What may be less expected about this patent is its focus on manufacturing environments. Microsoft has some prowess in manufacturing, but is far from a power player in the industry. But as we saw with Nvidia’s recent robot arm patent, factories are ripe for innovation with AI — though until now, the industry has “traditionally been a little bit of a laggard and have not adopted the latest and greatest technology,” said Vinod Iyengar, head of product at AI training engine ThirdAI.
While Microsoft’s patent hones in on the factory floor, the concept could be applied to a number of different industries, Iyengar noted. For example, in a logistics or supply chain setting, this could simulate when its systems would fall apart if demand were to increase. Even in a hospital setting, it may be able to tell when certain machines are at the end of their life, and simulate how long they would last in the event of an influx of patients, Iyengar said.
“Any interconnected, complex systems, where a single part or a single component could break the rest of the system, that’s a great use case for something like this,” said Iyengar.
Because of the broad use cases, Microsoft could get its hands on a patent for a powerful, stress-testing multitool that predicts when just about anything could break down. That may prove to be a bit more useful – and more lucrative – than just putting it to use in a factory.