{"id":77,"date":"2020-10-11T18:36:37","date_gmt":"2020-10-11T18:36:37","guid":{"rendered":"https:\/\/marshallbrain.com\/wordpress\/?page_id=77"},"modified":"2020-10-11T18:36:37","modified_gmt":"2020-10-11T18:36:37","slug":"second-intelligent-species5","status":"publish","type":"page","link":"https:\/\/marshallbrain.com\/second-intelligent-species5","title":{"rendered":"The Second Intelligent Species"},"content":{"rendered":"\n

Chapter 5 – How Computer Vision Systems will Destroy Jobs<\/strong>
by Marshall Brain<\/a><\/p>\n\n\n\n

If you look back at the description of self-driving cars in the previous chapter, notice that computer vision does not really play a role. Current self-driving cars do not have two eyes on the roof or the hood looking out at the road and deciding what to do based on visual input. Self-driving cars do have an optical camera, but it plays a small role. For example, it helps the car decide if a traffic light at an intersection is red or green.<\/p>\n\n\n\n

This might seem odd to many people. When humans drive a car, visual input through our eyes is essential. Why don’t self-driving cars do it the same way? Why doesn’t a self-driving car use optical cameras and binocular vision in the same way that human beings use their eyes to sense the world?<\/p>\n\n\n\n

Instead of cameras, a self-driving car uses different sensors to detect the world around it. LIDAR and radar are the two most essential sensor packages on a self-driving car.<\/p>\n\n\n\n

There is a simple reason for this difference: the computer vision systems that exist in production today (2015) are still fairly primitive. Computer scientists still have a ways to go when it comes to perfecting general vision systems. Yes, there are simple things that computer vision systems can do (for example, this video<\/a> shows a simple camera system to detect pancakes on a conveyor belt). But at this moment in history, there is not a computer vision system that can look at a common scene of a farm and say, “that is a barn, that is a horse, that is a man, that is the man’s hat, that is grass, that is a tree, etc.” A five-year-old human child can do that easily, but computers are not there yet except in special situations.<\/p>\n\n\n\n

In the same way, it is not currently possible to put a camera on the front of a car and have a computer use the pictures from the camera to identify other traffic, lane markings, bicyclists, pedestrians, dogs wandering into the street, etc. Computer scientists simply have not created the algorithms yet for computer vision at that level. But research in this area is occurring on many different fronts, both for the general case and specific situations. In the same way that Chess-playing computers eventually beat human players after several decades of research, and a Jeopardy-playing computer beat the best human players, there will eventually be computers running algorithms that are better than human beings at seeing the world. We simply haven’t arrived there yet.<\/p>\n\n\n\n

The thing to understand is that we will arrive there eventually. As this computer vision research bears fruit, a surprising thing will happen. It turns out that there are many sectors of the economy that will come under new pressure from robots and automation once robots can see the world. Once robots can see things like human beings do, it opens up whole new areas for robots to take over human jobs.<\/p>\n\n\n\n

If you think about it, you realize that certain human jobs – jobs that are not particularly difficult to perform otherwise – have been protected from automation because of their dependence on vision. Imagine that you have a time machine and you are able to travel back in time to the year 1950. If you walk into a restaurant, hotel or store in 1950, it would be nearly identical to a restaurant, hotel or store today from an employee perspective. If you go to a store or restaurant in 1950 and a store today, people do everything: people stock the shelves, prepare the food, serve the food, help customers, man the cash registers, clean the toilets and sweep the floors today in much the same way as they did in 1950. It’s the same on any construction site. In 1950, guys with circular saws and hammers built houses. Today it is guys with circular saws and nail guns. No big difference. Similarly, a hotel today has clerks at the front desk and people doing all of the cooking and cleaning just as it did in 1950. An amusement park in 1950 looks much like any amusement park today, with people operating the rides, selling the concessions and keeping the park clean.<\/p>\n\n\n\n

Industries like these are, by and large, untouched by automation today. These people-powered industries represent at least half of the jobs in the American job pool.<\/p>\n\n\n\n

Think about the people who work in a typical retail store like Wal-Mart, Target, PetSmart, Home Depot, and so on. The jobs these people are doing are not particularly demanding, but they do require vision. Think about tasks like:<\/p>\n\n\n\n