The Cloud, through bringing vast processing power to bear inexpensively, is enabling artificial intelligence. But, don’t think Skynet and the Terminator. Think cucumbers!

Artificial Intelligence (A.I.) conjures up the images of vast cool intellects bent on our destruction or at best ignoring us the way we ignore ants. Reality is a lot different and much more prosaic – A.I. recommends products or movies and shows you might like from Amazon or Netflix learning from your past preferences. Now you can do it yourself as one farmer in Japan did. He used it to sort his cucumber harvest.

Makoto Koike, inspired by seeing Google’s AlphaGo beat the world’s best Go player, decided to try using Google’s open source TensorFlow offering to address a much less exalted challenge but nonetheless a difficult one: sorting the cucumber harvest from his parent’s farm.

Now these are not just any cucumbers. They are thorny cucumbers where straightness, vivid color and a large number of prickles command premium prices. Each farmer has his own classification and Makoto’s father had spent a lifetime perfecting his crop and customer base for his finest offerings. The challenge was to sort them quickly during the harvest so the best and freshest could be sent to buyers as rapidly as possible.

This sorting was previously a “human only” task that required much experience and training – ruling out supplementing the harvest with part-time temporary labor. The result was Makoto’s poor mother would spend eight hours a day tediously sorting them by hand.

Makoto tied together a video inspection system and mechanical sorting machines with his DIY software based on the Google TensorFlow and it works! If you want a deep dive on the technology check out the details here. Essentially the machine is trained to recognize a set of images that represent the different classifications of quality. The challenge is using just a standard local computer required keeping the images at a relatively low resolution. The result is 75% accuracy in the actual sorting. Even achieving that required three days of training the computer on recognizing the 7000 images.

Expanding to a server farm (no pun intended) large enough to raise that accuracy to 95% would be cost prohibitive and only needed during harvest. But Makoto is excited because Google offers Cloud Machine Learning (Cloud ML), a low-cost cloud platform for training and prediction that dedicates hundreds of cloud servers to training a network with TensorFlow. With Cloud ML, Google handles building a large-scale cluster for distributed training, and you just pay for what you use, making it easier for developers to try out deep learning without making a significant capital investment.

If you can do this with sorting cucumbers imagine what might be possible as cloud power continues to increase inexpensively and the tools get easier to use. The personal assistant on your phone will really become your personal assistant and not the clunky beasts they are today. In your professional life they’ll be your right-hand minion taking over the tedious aspects of your job. Given what Makoto achieved perhaps you should try your hand at it. Who knows what you might come up with?

Share This