Artificial IntelligenceCryptocurrencyMetaverseCybersecurityTech Forward

Box’s Deal With Google Involves a Lot of Machine Learning

August 17, 2017, 4:10 PM UTC

People who use Box’s service to store their digital documents could find it easier to deal with files with lots of pictures.

The enterprise software company said Thursday that it’s now using Google’s (GOOG) image recognition technology to power a new service. That service is intended to help its customers who need to identify objects in photos and pictures in their online files.

Box (BOX) CEO Aaron Levie said that the service is suited for companies like retail shops that store images of their department store interiors and what to outline every detail in those photos. Before, those companies would have to use people to view each photo and manually label everything in it, like each individual item on a store shelf.

Levie pitched Box’s service as a faster way to do these kinds of mundane tasks that companies either assigned to low-level staff or bypassed altogether because of lack of resources.

Get Data Sheet, Fortune’s technology newsletter.

“Unfortunately, we are going to have to find some new roles for interns,” Levie joked.

Now, Box is using Google’s machine learning technology to sift through customer documents and automatically label images in their files. It’s one of the ways Box is using the various services of cloud computing providers like Google, Amazon (AMZN), and IBM (IBM) to augment its own service.

As of now, Levie explained, Google’s cloud service seems best suited for the task of image recognition—in part after roughly 15 years of Google honing its technology as part of its Google Image feature. Likewise, Amazon’s cloud computing service may be better suited for sifting and parsing meaning through audio files because of the work the company has done to improve its Alexa digital assistant, which understands and responds to people’s voices, he added.

In the long run, however, Levie expects these companies’ various image and audio recognition technologies to “largely equal out.” At that time, however, it’s likely these cloud companies will have another advantage over one another in some other future service, he speculated.

Levie did not say when the new image-recognition service would be released to the public, only that it’s currently testing the service with certain customers in what’s known as a beta test. Levie said he wants to make sure that the service’s accuracy rate at identifying objects is high enough before it’s ready for prime time.

These types of image recognition and other automated tasks, he continued, are examples of the type of tasks artificial intelligence technologies are currently best suited. “It won’t necessarily show up in some sort of magical robot or a [digital] assistant that can do anything,” he said in reference to AI’s current capabilities.

Instead, Levie concluded, machine learning presents “a lot of extremely pragmatic use cases” to automate and perform certain tasks at an “order of magnitude more accurately than a human can do.”