When we talk about artificial intelligence (AI) in food tech, it’s often about the end result: Cashierless checkout, crop assessment, autonomous vehicles, etc. But one thing that these solutions and any other using AI need is is good data. Sama is a company in the good data business, and it has built a platform that provides training data that other companies can use to speed up the development of their AI models.
I spoke with Wendy Gonzalez, CEO of Sama, this week by video chat. She outlined some of the food tech use cases for her company’s technology, such as fighting food waste. “If you’re in a restaurant or hotel, catering service. A lot of that food gets wasted,” Gonzalez said. Sama is working with a company called Orbisk that provides a device for commercial kitchens that uses computer vision to analyze the food being thrown away. For example, if Orbisk sees a lot of mac and cheese is being tossed because no one is taking it, that kitchen can know not to make as much of it, and by extension save money by ordering less of the ingredients to make mac and cheese. (Winnow is another company that takes this same approach.)
For its part, Orbisk had a thousands of images of different types of food for its system to recognize. Sama came in and provided structure and taxonomy to that data. In other words, Sama labeled all the images of mac and cheese accordingly to train Orbisk’s AI to automatically recognize mac and cheese. The result, according to Sama, is that Orbisk’s system can reduce food waste in commercial kitchens by as much as 70 percent.
Gonzalez said that Sama’s system is also being used in other fields like cashierless checkout. In that setting, Sama is helping train those computer vision systems to recognize packaged goods, which is more complicated than people think. A cashierless checkout system needs to not only recognize a package of Oreos, but a package of Oreos in different lighting conditions, different angles or when the view is partially blocked. Sama provides all of that data.
Sama is also being used to train AI systems on early crop disease detection, automated crop harvesting, and soil condition monitoring.
Sama is among a number of players in the data space including AI.Reverie, which uses synthetic data to create images virtually to train AI models, and Nvidia, which is also using synthetic data to train robots to navigate around a kitchen.
As AI plays a bigger role in our everyday lives, there will be a growing need for more good data to train those AI systems. And the end result is that we’ll all be talking about good data more often.
