Many times at work, I wonder about the future of my job, about how it will eventually be taken over by robotic labor. It may be strange that I think about how to eliminate my own livelihood, but it still fascinates me. I am a checker at a grocery store. My duties are to check out groceries, collect money, stock the deli case (dairy, cold juices, butter, cheese, etc), and ice cream when I close. These are deceptively simple tasks since we would generally expect anyone to be able to do them. But automating my job means overcoming some of the toughest obstacles in robotics
For one, I have to able to interact with and help customers. That’s no easy task for a machine. I require a mastery of the English language with very minimal margin for error when doing this. We have voice-command systems today, but these are fairly limited (for example, when checking various features on my cell phone’s plan, I can solely use my voice, but maneuvering through menus essentially consists of the system telling me “to do X say Y”. If I stray from that, it easy gets confused. Fortunately, it’s good enough that I never had to go through a tutorial or anything to “train” it to recognize what I’m saying). In order to interact with customers, a computer
I also have to walk around the store, maneuvering among other customers and employees, which, of course, requires accurate vision processing, balance, and mobility. In addition, when ringing up groceries, I have to be able to recognize the many different types of produce. When buying groceries, you may have noticed funny little stickers that have numbers on them. These are called Produce Look-Up codes, which checkers punch into the check-stand to correctly charge the customer. One thing you may notice is that many types have no sticker at all. We checkers have to memorize the rest. It’s easy to teach a computer what PLU# goes to which type, the hard part is getting it to visually recognize what the produce in question actually is, preferable without having to stop every time to ask the customer what it is.
Bagging groceries is also no simple task. You have to be able to recognize which items should go on top, which should go on the bottom, that only meat should be bagged together, only cold items together, etc. It might be useful to be able to feel temperature by touching it. That way, only cold items go in one back, hot items go in another, and so on.
All these things require a formidable computational capacity to achieve. Not only that, we also have to be able cram all that capacity into approximately the volume of a toaster. I mean, modern-day supercomputers already approach the capacity of the human brain. They, in theory, could be used to drive a bipedal humanoid robot that could directly replace a human being, but that robot would be horrendously expensive and impractical. Plus the computer system running it would struggle to fit in a semi-truck, not to mention a humanoid frame.
But, we only need to look to Moore’s Law to see that the main requirement (a massively powerful computer system), is rapidly become cheaper and cheaper. Within 15 to 20 years, the average desktop computer will exceed the intelligence of its human operator. About ten years after that, the required programming to create multi-tasking humanoid and non-humanoid machines will be commonplace.
I’m not worried, though. I highly doubt I’ll still be a checker in 2030. Maybe, if I choose to stay with the company, I’ll be a store manager or even executive. My children, however, looking for their first job, won’t be so lucky…