In late post-revolutionary France one man was tasked to map out the country. Gaspard de Prony, a mathematician and engineer, decided to approach the task by creating logarithmic and trigonometric tables. These tables, which would come to be known as Tables of de Prony, were destined to speed up the trigonometric calculations needed to complete these cartographic task.
In handling the vast amounts of data, de Prony asked for help. His team was divided in three levels of hierarchy: besides a couple of highly skilled mathematicians, several mathematicians with less sophisticated skills, he also hired sixty to eighty hairdressers.
These were the people who lost their jobs with the massacre of aristocracy. Now that their clientele was there no more, they were employed as some of the first human computers.
What de Prony showed was what happens when you apply a little bit of code and a little bit of engineering to a problem. Since that time, we’ve made fantastic improvements with this: we’ve replaced tables with billions of silicon circuits; we’ve got better and better with algorithms, and the result we’ve got now in 2017 is that there are many domains where computers are catching up with human abilities.
One of the key milestones with this was in 2011/12 when deep learning came of age. Alex Krishevksy and colleagues put together a convolutional neural network to identify objects in images. This state-of-the-art network was eight layers deep and ran it on 2 GPUs.
In 2011 computers were getting one in four images wrong; there were improvements where the red line is (chart ↓), and it only gets better afterwards: image recognition catches up with human abilities.
And it does so with lip reading:
And again with speech recognition (which is particularly interesting considering this wasn’t possible for 20 years of using hidden Markov models; ASR only improved when scientists started implementing deep neural networks):
And with sketching, among other things.
For the longest time, one question has lingered: what will automation do to human workforce as computers get better and better?
Frey and Osborne found that 47% of US jobs are at ‘high risk’ of automation in the next one or two decades. This year Deloitte published their findings for the UK, where 35% of jobs are at ‘high risk’ of automation in the next decade or two.
Acemoglu and Restreppo have shown that industrial robots bring decline in both employment and wages. Robots are perfect substitute to human labor, and in addition work for free, day and night. Flooding the market with these bits of labor will reduce the prices that other types of labor can command, and will take some out of work.
But it’s not all black and white. Looking at the implementation of automated teller machines, a less gloomy and sensationalistic trend appears: the rising number of ATMs didn’t oust bank clerks completely. People were redirected to complete other tasks in banks that haven’t been automated.
If we look at the US economy, there is a clear decline in agricultural employment over 140 years, parallel to the rise of some other opportunities — such as accounting or hairdressing.
The question is what happens to millions of people who move from certain classical jobs and don’t have the flexibility to move into new jobs?
We can see the image of automation if we look at what it takes to support $5bn of revenues of Blockbuster vs. Netflix. In fact, even as a declining business (“Blockbuster (2)”; (“BlockbusterB(1)” is the business in its primetime), this company employed more than 50k people more than Netflix does in its bloom.
There will always be some things that computers can’t do.
There are more and more domains captured by computing. But no matter how well we improve the computer systems, the one thing computers won’t be able to do — is to be human. I’m not talking about some advanced Turing test, but the substance of being human, on a biological level even.
In the world of the future, automated perfection is going to be common. Machines will bake perfect cakes, perfectly schedule appointments and keep an eye on your house. What is going to be scarce is human imperfection.
One of the few bits of economics that tends to work is the macroeconomic analysis of demand and supply. If you have a world where the amount of perfect products we can produce increases almost infinitely by using AI, robots and clean energy, we’ll end up with a surfeit of supply, which will push the supply curve far to the right. It will come along with demand curve and ultimately the price will decline.
What will be plentiful will be the perfect product. What will be rare will be imperfect products; the products that got touched by the human hand.
A little bit like Persian rug which are produced with error in them. The weavers believed that only god can produce perfection. We might start to value the things that are less perfect, from the ones that are, from the less scarcity value.
Human-made will be valuable — I can imagine going into a supermarket and seeing on the shelf products labeled “not touched at all by a robot or machine.”
This will be one part of the artisan economy—the economy in which humans will have the space to excel in experiential, discretionary, and intimate. Robots will take all those things that are of high risk, seek reliability, and are repetitive.
De Prony took humans and employed humans to do mind-numbing tasks. Here we are 200 years later — we have robots to do those tasks, while humans can take over the artisanship.
If you enjoyed reading this, you might want to check out my Sunday newsletter on the impact of exponential technologies on businesses and society. Sign up here. ⭐
This article resulted out of a presentation I delivered at The Europas Conference. The presentation is available on SlideShare. Thanks to Hung Lee for recording: you can view the video here. I’m grateful to @auerswald for his insights into de Prony.