The press loves a good dramatic story about how AI will replace us and steal all our jobs. Sure, in the future general AI will surpass human abilities in pretty much everything and should do a better job, but what about today?
Google have got a great blog post up about a Japanese farmer using deep learning to sort cucumbers. I was surprised to learn this is actually a highly-specialised task, taking eight hours a day during peak harvest time. It requires serious training. With some off-the-shelf code from Google, a Raspberry Pi and a camera, Makoto has built a deep learning system that hits 70% accuracy in real world operations. He's not far off fully automating a human judgement role.
It's a great example of isolating a highly-skilled but repetitive task and using the power of image recognition and deep learning to automate it, even if it's not yet performing at super-human levels. This application of AI isn't stealing a cucumber farmer's job, but is freeing the farmer up to spend more time farming. So maybe not as dramatic and newsworthy as stealing jobs, but a real-world example of leveraging AI to improve human productivity.
But can computers really learn mom's art of cucumber sorting? Makoto set out to see whether he could use deep learning technology for sorting using Google's open source machine learning library, TensorFlow.