"10 years of automation in 1 year"
What the automation acceleration from COVID19 actually will look like. What did robotics look like a decade ago and is this step feasible?
Over the last few months, my media subscriptions have been flooded with phrases about how robots are taking over and the future is here.
Tweets saying that we are going to get “10 years of automation transition in 2020” (that I sadly could not find).
Coronavirus is bringing the fourth industrial wave - Arizona State University.
Companies are investing in automation more than ever - Wall Street Journal.
Seed money is increasing from venture capital to automation startups - TechCrunch.
Some countries are using self-driving cars at scale - MIT Technology Review.
I did some digging into reporting on actual financial movements in automation software or physical robots.
Pre-COVID19 reports predicted a 25% annual compounding rate of the robotics market (buying robots). With the power of compounding, a 25% increase year-over-year is pretty wild (triples in size in 5 years).
Amazon recently bought another warehouse robotics company.
Companies do a pretty good job obscuring investments on a short time scale.
Will the predictions of these publications come true? What would a dramatic acceleration in automation (potentially) look like?
Past, Present, and Future of Automation
A Brief History of robotics
Robots are enthralling to most people. They’re awkward but obviously valuable.
The term "robot" was coined by the Czech Karel Čapek in R.U.R. (Rossum's Universal Robots), 1921. The earliest industrial robots we designed in the 1930s-1950s to replicate human motions in manufacturing. Asimov coined the Three Laws of Robotics in 1942. Late in the 1900s, the accuracy of the robots improved and funding increased in military and medical applications.
The robots of the 21st century (I would call them “modern” robots) are rapidly expanding in scope. The first personal robot emerged in 2002 to vacuum floors. Designers were giving robots personality. All constraints are off, and robots are expanding in scope, personality, and effectiveness. This article is examining how.
Robotics in 2010:
Robots a decade ago were either toys thrown at hard problems or simple agents thrown at toy problems. The former showcased how far we have to come and the latter highlighted the potential for autonomy to solve simple problems.
Take this example of the DARPA Grand Challenge (autonomous cars in the desert 2004, 2005; wrote more about self-driving cars here).
The multi-agent ground robotics competition (MAGIC) was a nationwide research push. Now, UC Berkeley has its own local delivery robot fleet (they’re cute, a little reckless, and collecting a very valuable local transit dataset).
This was also the year(s) of First Robotics Competitions from NASA. This competition scaled down to grade-schoolers like myself and served as an introduction to robotics.
We had Robonaut, which was cool but not super useful, and a lot more. Robots a decade ago are not impressive to look back on because of the rate of progress. A decade of technological progress is hard to predict, and exponential, so the past gets rapidly dwarfed.
Everyone wants personal robots for housekeeping. It’s a question I get all the time. It’s fun to retort that there is already a financially stable, consumer robot company doing this - iRobot.
They make the Roomba. State-of-the-art as the robotics researchers would say.
That is where home robots are right now and where they have been for years. Although, now we also have high-speed pick and place robots accelerating manufacturing (founded by a Berkeley AI Research member, Professor Abbeel). This was not remotely possible a decade ago (see this paper). This application is automation in a controlled environment.
Tesla is doing a really good job. I’m not going into the details, but the corner cases still exist and are very scary (but compare this to the DARPA Grand Challenge video).
The dual landing of rocket-boosters never gets old. Space-X is carrying the torch for the inspiration of incredible engineering these days.
All of these innovations are enabled by highly specific engineering progress. Teams of experts working on one problem. (Yes, they are enabled by new tools like deep learning, but the problem of robustness generally needs more human design).
Robots in 2021
The robots will be mostly the same, just with more adoption in areas where applicable. The areas that robots work now are enabled by an extreme level of specificity and structure, not by flexible algorithms and random tasks. This is why there are no household robot-housekeepers that can unload the dishwasher and make your bed (although thousands of dollars of research showed that robots can slowly learn to fold towels).
The robots of the future decade will likely look similar and accomplish tasks of similar difficulty, but there’ll be way more of them.
A drastic increase in investments does not equate to a dramatic increase in effectiveness overnight.
Autonomy & Sci-fi
Sci-fi has enabled the aims of technology for many years. I have written about this in the past, so re-including it here.
Radio-controlled automobiles were first demonstrated in 1925, and only 5 years later, science fiction predicted full autonomy in Miles J. Breuer’s 1930 book Paradise and Iron (source Robin R. Murphy’s “Autonomous Cars in Science Fiction,” Science 2020). Paradise and Iron predicted that every moving system on an island paradise was fully autonomous — there wasn’t even a steering wheel. This extends to more than cars, this means that cranes are autonomous, construction is person-less. Full autonomy systems will touch more than self-driving cars.
Maybe we will see a new wave of sci-fi, but I don’t expect robots to change overnight (side effect, this automation wave will benefit big tech companies disproportionally).
I expect the applications of 2030 to be as big of an increase in awe of the last decade, but the coronavirus will be driving adoption, maybe not innovation.
What’s new with me
I am reading (newsletters/blogs):
SARS-CoV-2 and the host response: psychological stress - Peter Attia. Take care of yourself (TCOY) - cjk87.
Why Facts Don’t Change Our Minds - James Clear. A good reminder of how our brains work.
How much did AlphaGo Zero cost? - Dan H. Insight into how much it costs to do at-scale machine learning research.
Books (this was a big one with a cross-country drive):
The Gatekeepers: How the White House Chiefs of Staff Define Every Presidency - Chris Whipple. A great insight into the inner structure of The White House intertwined with American history.
The Most Dangerous Branch: Inside the Supreme Court in the Age of Trump - David Kaplan: How the Supreme Court is becoming a political and powerful entity in the age of Trump.
Human Compatible: Artificial Intelligence and the Problem of Control - Stuart Russell: Why we need more future-conscious AI designers.
I am listening to:
A history and future of deep learning - AI Podcast 94 - Ilya Sutskevet.
There is a valid defense to Facebook’s position on policing speech - Exponent.fm 166, Speech and Systems.
Going back and hearing both sides of net neutrality - Exponent.fm 133, Two Terrible Options.