11 Lines Of Code Will Replace Millions Of Jobs. What Can We Do About It?
Today, if you work in computer science, it’s hard to go a week without a new doomsday theory about machines either being the death of us or expediting our own destruction by giving us new and more efficient tools to kill each other. It’s been such a frequent occurrence in my field that debunking some of the more exotic predictions with real science was a fun side job for years. But given a recent advance in computer chips, I’m going to reverse course on my trademark skepticism and tell you that I have seen into the future and that future is grim.
For the last six years, public techies have been warning everyone in sight that machines are coming for their jobs. We’ve pointed out that the collapse of blue-collar manufacturing wasn’t caused primarily by outsourcing, but by automation. We sounded the alarm that the countries where manufacturing jobs were indeed outsourced were about to suffer the same fate. And we watched in horror as the Treasury secretary of the current administration said that AI able to take people’s jobs is 50 to 100 years away, much the same way we’d look at someone who was trying to explain to us that human spaceflight is at least a century down the line and the Moon landing was faked.
Don’t get me wrong, automation is already in full swing and I’ve personally replaced what once were, or would’ve been, full-time, white collar jobs in offices with a few hundred or thousand lines of code that run on a server no one aside from IT sees, thus making it seem as if those jobs were still around, just waiting for HR to start the hiring process. But with new computer chips which borrow heavily from the basic processes of the human brain, that automation is about to get a whole lot more sophisticated and replace a whole lot more jobs than we may have even thought possible. And it’s going to catch our leaders with their pants down. Again.
You’d probably forgive me if I didn’t dwell on the details and skipped to the part where we either try to figure out a solution to the steady encroachment of your digital replacements or rue our rotten fates and plan on how best to drown our sorrows. But skipping the details and a lack of understanding of what we’re doing at the cutting edge of computer science and AI is what caused this mess in the first place, and only by understanding what’s out there, how it works, and what it can do, can we come up with workable policy ideas, none of which will be simple or straightforward, and most of which will require a radical rethink of our society in the long run.
If you stick with it for just a little while, you’ll hopefully come away with an informed idea of how modern AI works, why many of the advancements no one is talking about are so critical, and learn some interesting things about your own brain in the process. Yes, this will involve math and science, the things far too many of our politicians proudly tell us they shy away from or giddily misuse, but we’re long past the point where ignoring what they have to say is a luxury any of us can afford. With that in mind, let’s get started.
How Androids Learn What Electric Sheep Are
As we already established, when people talk about jobs going away, they think foreign workers or a distant sweatshop. It’s an idea that has sticking power because the villains in question have faces and are easy to picture. Saying that nearly 9 in 10 manufacturing jobs were lost to automation is extremely abstract. What does said automation look like, other than large robots in factories we’re all familiar with by now? Well, it looks somewhat like this…
This is an algorithm written in a computer language called Python by a Ph.D. student from Oxford. It could be written in any language, however, because what matters is what it does, not how it’s written. And what it does is allow a computer to learn in a way that’s roughly similar to how your brain would, albeit with a few big caveats. Backpropagation is just one of the several ways machines can be trained, and this learning by trial and error resembles how your mind works, though only in the vaguest possible terms.
Your brain learns primarily by repetition. As you’re exposed to stimuli, a mix of electrical signals created by the interactions of sodium and potassium ions, and neurotransmitters, reinforce connections between certain brain cells, called neurons, creating a dedicated path in your mind that will help you execute a task. Or at least that’s the extremely simplified picture on which computer scientists in the 1960s based their ideas of how to make computers learn, realizing early on that entering all the data a machine would need to know to process data for decision making would take centuries.
The result of their efforts was something known as artificial neural networks, which use code like backpropagation to guess their way into doing something correctly. They’re fed a set of inputs and a desired set of outputs, then given tens of thousands of chances to guess what inputs are the most important in coming up with the right solutions so they can calculate future answers. Every time they make a mistake, they calculate their average degree of error, update their guesstimates of the inputs’ importance by this average, and try again until they get it right. This is backpropagation in a nutshell.
Now, the computer doesn’t understand what it’s doing or why, or any of the concepts involved. (That’s a topic in AI known as symbol grounding and it’s far out of the scope of this article.) It’s like a dog learning that shake means raising one of its front paws and letting you take this paw in your hand. It may have taken a while to understand this is what you want when you say “shake” and get it wrong a lot of times. It won’t understand the concept of shaking hands or why you want to teach it this, but it will remember the word and know what you want it to do now in exchange for a treat. Substitute its attempts to learn for statistical formulas, remove the treat because the computer doesn’t actually want anything to motivate it to keep solving the problem, and you’ll end up with a rudimentary AI.
Much, much larger versions of these artificial neural networks power Siri, Alexa, and Cortana, fuel your Netflix and Amazon recommendations and arrange your social media timelines. This algorithm, or some version of it, is behind virtually everything you do in the digital world. And if this is your first time hearing about artificial neural networks and backpropagation, you’ve been none the wiser to the software that’s been running your digital life for close to a decade now.
All Your Jobs Belong To Us
All right, all that’s fine and good, but what exactly does that mean for jobs? It means that if your job involves making decisions or taking actions based on well-defined criteria that rarely change, your job can now be taught to a computer. It can weigh the factors and decide for you, or ping a different piece of code to carry out an action. Different neural networks can even be chained together to make more and more complex decisions and plot more elaborate plans of actions. Best case scenario, you’ll be kept around to keep the software in check as it makes millions of decisions and carries out thousands of different workflows. Worst case? You’re obsolete.
That’s really the reason why you should be worried. It’s not that you’ve just been given some help to do your job faster or will move into a different job. Your job is gone with nothing to replace it planned. During the halcyon days of industrialization, humans were needed to make ever more complicated decisions about how the relevant work had to be done. In the current post-industrial era, the sheer scope and complexity of the work are so vast, we actually need to have computers finding inefficiencies and in the supply chain. Humans alone either aren’t enough or have become the inefficiencies in question.
At least artificial neural networks consume a lot of power and as a result, have a fairly limited mental bandwidth compared to humans. That means there’s hope for us yet, right? Well, not for long. A laboratory in UCLA is developing a hardware implementation of the algorithm we just dove into, perfect for robots, servers, and even household computers, sipping energy while putting the power of AI networks that could emulate the work done by multi-million dollar setups today in a device that will fit on the tip of your finger and consume the same amount of energy as your cell phone.
AI will not only become even more prevalent but get orders of magnitude cheaper. Even a local mom and pop could have the computing capabilities Facebook has today within a decade or so. Millions of what used to be entry-level jobs for humans are being replaced by cheap, intelligent machinery. With these artificially intelligent chips, millions of junior white collar jobs will be the next to go. Whatever jobs will be left will be either highly technical and either maintain and improve this automation, teach future generations how to do this, or stick to law, science, research, and creative tasks where machines will no doubt be used to aid all of them in some way.
None of this would be catastrophic if we actually took the warnings seriously and had a plan on how to transition people to new potential jobs. But we didn’t. Even worse, the party currently in power seems far more interested in saving coal mining, an industry being rapidly automated and employing fewer people than Arby’s, than investing in education and training for future jobs, so it’s unlikely any new ideas are going to come down the pipeline and translate into actionable policy anytime soon.
So this is where we are. Machines are steadily taking over job after job thanks to math that allows them to learn new tasks and can be scaled to meet growing logical and cognitive complexity, and they’re about to rapidly speed up their invasion of the human workplace. In the meantime, our leaders ignored the warnings and have been sitting on their hands for decades while peddling half-baked ideas about the economy, and severely undermining science and education with asinine overhauls and brutal budget cuts. Even worse, they’re pushing us in reverse on the issue.
The current tax plan in Congress will actually sabotage higher education by taxing scholarships and student loan payments, which will especially hurt future scientists and engineers by taxing their average $30,000 in stipends as nearly $80,000 in income by including their tuition waivers. Meanwhile, we’ll institute deductions for outsourcing and private jets, while taking away deductions for teachers’ buying classroom supplies and cutting healthcare to those who won’t be able to find jobs thanks to automation and inability to afford retraining. This lets you know exactly how backwards the priorities held by the majority of our leaders are.
The Problem Downloaded Around The World
Globalization also played a big part in growing the AI threat. Gone are the days when executives of big companies worried about having enough employees to get the job done and making sure they could afford to buy the products they manufactured. Machines can now do more and more jobs, and they’re much cheaper and harder working than any human. And if their employers can’t afford to buy their products, better off foreigners can, and will. Basically, they don’t need to be good corporate citizens if they have a global customer base.
So while globalization is good at creating new opportunities for people with certain skill sets, anyone lacking those skills can be effortlessly cast aside with no side-effects to the bottom line. Without pressure to train new generations of employees, just compete for far fewer people, far too many companies feel little pressure to lobby for real reforms in education and job training, and at any rate, governments don’t seem to be listening. In America, out of touch senior citizens are making policy and it shows in their befuddled inaction and quarter-measures when it comes to AI, on top of their utterly backwards digital strategy.
For the companies that are actually worried, relocating to a nation better equipped to handle the post-industrial world is always an option, although that’s not great news for a lot of countries that are currently awash in manufacturing jobs. The same seismic shifts that rattled factories in the First World can play out exponentially faster in developing nations because the bugs have been worked out, the technology got cheaper, and adoption can be streamlined. This is why the UN predicts that two-thirds of all jobs in the developing world will be gone in less than a generation.
While this may seem like a scathing indictment of the corporate realm, that’s not really the case either. This is just the world we live in, and it’s not a company’s job to educate its future workers or craft policies that will allow us to adapt to the future. That’s why we pay taxes. Governments are running schools and determining budgets for crucial programs we need for future-proofing our workforce, not industry. The day when between half and two-thirds of humans are out of work thanks to AI and lack of investment in the future is still over two decades away, and most businesses are simply not designed to plan on multi-decade time scales. And even if they did, they’d need governments to go along with their recommendations.
But not only is the advice of tech companies which are implementing AI faster than the researchers, routinely ignored, our politicians are pushing for tax cuts and doubling down on failed policies. While this never made much sense, pushing for them right now is particularly bad for two reasons. First is that the cuts are predicated on the notion that companies will hire more people with the money they save, but with machines and AI eliminating jobs at an accelerating rate, for what jobs would they even hire? Second, the proposed tax plan would actually raise taxes on the middle class to pay for the loss of revenue as they’re squeezed by AI. So you see, this is a perfect storm of inaction on one end and ineptitude on the other.
This is where the radical rethink mentioned earlier comes into play. We need to accept that a) many current jobs are gone and are either obsolete now, or will be obsolete soon because they no longer require a human to do them, b) numerous other jobs are on the chopping block, c) we need policies that make education and mobility affordable and easy, and d) the future involves a lot of curiosity and creatively-driven jobs that are only enhanced and accelerated by computers, and we need to start training people for them, not save failing industries.
How To Start Solving Tomorrow’s Problems Today
All this sounds common sense so far, right? Where’s the radical part? Well, the radical part is in the execution of the policies we need to implement based on these four basic new truths of the post-industrial world.
Truth #1: We need to make big changes both to our current system and our funding priorities. Business as usual and gradual change simply aren’t workable anymore, they haven’t been since our leaders twiddled their thumbs for the better part of three decades as warnings about automation and the need to set ourselves up for success in a machine-driven world poured in from experts.
Truth #2: We need to provide a form of universal healthcare, decoupling this benefit from employment. When we need a mobile, flexible workforce, ready to re-educate itself at a moment’s notice for the Next Big Thing, limiting their ability to change jobs, or take time to retrain by trying their benefits to who employs them makes no sense. Otherwise, we will make workers choose between training themselves for some new and desirable skillset and losing access to doctors if they or anyone in their families get sick, or risking going into crushing debt to pay for treatment.
Truth #3: We have to invest in education. We don’t just need to mint more college degrees mind you, but make training for new jobs easier and far more affordable. We need to stop teaching to standardized tests and return to a more discovery-driven, exploratory education style, and consider tracking in high school to help steer students towards careers for which they demonstrate a distinct aptitude.
Truth #4: Finally, we need to pump more money into basic, curiosity-driven research to tell us what’s possible and what new things we can try to invent, expanding government-funded labs and R&D departments in the corporate world. While conservatives seem to hold practical science, like AI, medications, weapons, and gadgets, in much higher esteem than basic science like biology, astronomy, and physics, without basic science, engineers wouldn’t know what they could build and doctors wouldn’t know how they could treat diseases and injuries. One cannot exist without the other over the long term.
Basically, we need to adopt the opposite of virtually every initiative we’ve seen over the past decade because what we’re doing is obviously and clearly not working. It’s enabling the worst actors in the developed world to grab power based on pitching fear and nostalgia, it’s setting up our kids for obvious failure, and it’s hobbling our future right before our eyes, all because the people we placed in power refuse to open their eyes and admit they don’t understand how the world works anymore. How many of them have heard of backpropagation? How many read a primer about AI? How many of them are aware of the artificial neural networks on a chip that will bring Facebook’s and Amazon’s setups to businesses of any size sooner rather than later?
Forget about liberal vs. conservative for a minute. Diving into AI and how it’s changing work for people across the world shows that the divide is more about people who fear the future and want to slam their nations in reverse and people who understand that if we don’t adapt to new challenges, we’ll be worse off while those who do will reap the benefits. No one is entitled to be number one in anything without working to preserve that status, and no amount of pretending that the future isn’t coming for us has ever saved anyone from eventually feeling its full brunt.
Our only two options are to adapt and try to succeed, or hide our heads in the sand and suffer. Americans have always tried to choose the former rather than the latter, so if this time we decide to cling to the past and let history overtake us, it will be on us. And is this really how we want a machine takeover to play out? Not with killer robots with access to nuclear weapons deciding to wipe us out but them taking our jobs and we meekly let them so our graying leaders could rummage through our pockets one last time and mew some conservative buzzwords while reciting what passes in their minds as a motivational speech?