Why Don’t Machines Do The Work?
If in Victorian times we took Oscar Wilde at his word, touched on in ‘Soul of Man Under Socialism’, we could have today a base aristocratic Victorian standard of living (at the very least) without people at all, starting with time and motion studies followed by mechanical replacement of labor. Laws would make toil illegal while government treasuries would prioritize the budget toward science, design, medicine, and engineering. The most gifted among us would have life-long personal mentors, PhDs trained in education aligned with the aptitudes and aspirations of the student. For everyone else, the enjoyment of arts and culture.
Fleets of the innovative class might make up less than 5% of the population.
Today about 65% of the population is dependent on the other 45%. In countries like Japan, this dependency is expected to increase with 1/3 of the population being over 65 by 2050. Fortunately, Japan embraces robotics, which will ensure the care of elderly as they outnumber the nation’s working populations into the 2020s.
Unfortunately, there are no signs of a popular uprising to support anything even approaching such ideals. If social media is any measure, much of the political discussion ends with identity politics. Strong AI, it seems, is our only hope.
If it were not for residential rent or mortgage payments, including transport costs, life based on the minimum wage might be of a decent standard, but automation is moving rather swiftly in these types of low skill repetitive jobs. McDonald’s fast food chain, for example, will replace 5,500 cashiers with automated kiosks by 2018.
So how long will it take McDonald’s to fully automate outlets? Thinking linearly, at about 5,500 jobs replaced every two years, only 55,000 jobs are lost in 20 years. Yet, all tasks will become absorbed by information
technology, aligned with robotics, with affordable LIDAR sending real time models of operations using something as yet created by Autodesk. Because an entire global business is not necessarily as straight forward as semiconductor manufacturing, let’s estimate a doubling of automation efforts every six years:
5,500 jobs by 2018
11,000 jobs by 2024
22,000 jobs by 2030
44,000 jobs by 2036
88,000 jobs by 2041
176,000 jobs by 2047
This doubling may itself double sometime around the 2030s with pocket sized computer performance having the computational equivalence of the human brain, and once software can automate the training of deep nets to perform both physical and mental tasks, while also organizing experiences in place of jobs.
An oft-cited 2013 Oxford Martin School study estimated 47% of all U.S. jobs could be automated in the next 20 years. That study looked at jobs that were low skill and repetitive — easy to computerize — but white collar jobs like doctors and lawyers will also come under threat with programs like IBM’s Watson. This McDonald’s estimate for 2036 is conservative to the study’s U.S. average, until the next two doublings, with 346,500 McDonald’s employees being automated from 2018 to 2047. McDonald’s in 2016 employed 375,000 people maintaining 36,900 outlets.
Remaining jobs will relate to service quality, like master chief inspired chemical engineers making food healthier (?!), further reducing costs, while also expanding reach by drone delivery or launching automated food trucks to parks and other recreational areas.
If that scenario sounds terrifying, there is hope. McDonald's will have fierce competition around the 2030s with inexpensive lab grown foods, vertically grown meat and vegetables — delivered by drone — complimented by kitchen robotics to produce exceptional quality meals, relegating tinned food and microwave meals to the gruel of the dark ages.
Transistor size is presently at the 10 nanometer (nm) scale. 5nm transistors are expected commercially by 2020, while Berkeley Labs in 2016
produced 1nm transistors, something to expect commercially by the mid-to-late 2020s. GPUs power autonomous vehicles and we can expect other specialized processors used for high demand AI segments. Eventually Moore’s Law will hit a physical limit, but by such time, wireless computing working in parallel (i.e. cloud computing) with multiple devices should suffice.
The open source Robot Operating System provides a good measure of robotic progress, however slow it may seem, after over 3,000 packages, steady as she goes:
AI will need significant advances to achieve the hypothetical automation doublings mentioned. When IBM’s Watson defeated the Jeopardy Champion, Ken Jennings, in 2011, I was hopeful a question-answering system like Watson would quickly lead to conversive AI sometime soon. And while AI can answer more complex questions or find nearby shops based on specific preferences, an Amazon Echo or Google Home cannot yet converse in a meaningful way. Microsoft, Google, and Facebook have voice over IP services that should provide more than enough data for a smart chatbot. It either seems like any time now or decades away.
AI can identify multiple objects from video and also determine a person’s mood. Verbal languages can be translated from one into another with vocal and video synthesis (not great, but a good start!) now able to emulate an individual’s voice with just a one minute recording.
Google’s AlphaGo has defeated a top Go champion, Lee Se-dol, sooner than expected. Google’s Deep Mind recently demonstrated learning how to walk and traverse obstacles in 3D. Every automotive related company wants to crack autonomous vehicles with Waymo (Google) seemingly in the lead.
Personally, a campfire in the forest sounds really good right about now.
