Watch all the Transform 2020 sessions on-demand here.
With open source and third-party tools to guide them, businesses should be able to incorporate artificial intelligence into their work without having to start from scratch, right?
Not if the industry in question has no data, as is the case with parts of the food industry. AgShift CEO Miku Jha has a mantra to sum up the problem: “Older is the industry, tougher is the battle.”
“No matter how deep your bank account is, if you wanted to bring AI to solve the issues in the food supply chain, you will have to start from the absolute bottom. Because no one has images stored anywhere, anywhere in any facility,” Jha said at VentureBeat’s Transform 2019 conference.
Her company, which automates food inspection, has to create data for each new commodity it works with before training its tools to assess food quality. AgShift does this by shadowing its clients and working with experts in the domain of food defects. “We just camped in the processing facilities of our customers day in and day out,” Jha said, in explaining how this hands-on approach works. “We were eating their food, we were picking strawberries from the farm. And in the process, we were collecting a data set.”
June 5th: The AI Audit in NYC
Join us next week in NYC to engage with top executive leaders, delving into strategies for auditing AI models to ensure fairness, optimal performance, and ethical compliance across diverse organizations. Secure your attendance for this exclusive invite-only event.
Once AgShift has created a data set for one commodity, the company can use that data as a baseline for training AI tools to work on the same item for new clients.
The end result for food processing facilities is a kiosk-like device, Jha said. “You put in the commodity you’re trying to inspect, [and] the system takes the image, [and] gives you the results back in under 20 seconds.” Traditional human inspection takes considerably longer and is less thorough.
Agriculture might be an extreme example of an industry that lacks data, with so much work done painstakingly by human hand. But factories in other industries can also lack sufficient quality data to train AI software, said Janet George, chief data officer at data storage company Western Digital.
George said a good solution to this problem is to use GANs (generative adversarial networks) to create synthetic data.
“This helps us a lot, because we can go in there and start creating data that we want for training purposes. And so I was able to go into [a] factory, get all the domain experts, and really standardize what bad data looked like, what good data looked like,” she said. Based on that, companies can create data sets to train AI tools.
