Will Robots Take my Job?
Will a robot take your job? This is an understandable question given the current hype about robotics and artificial intelligence. Many people are asking “will my job be automated?” or “is my job to go altogether?” as companies and industries become redundant.
I could ask, will a robot take my job as the writer of this article?
My answer is no.
Will robots take our jobs? It’s very unlikely to happen anytime soon unless you have a job that is in every aspect a real world equivalent of playing a computer game. I will explain.
Many dynamics are driving the current narrative around robots, chatbots for business, and AI taking over jobs. A large part of the discussion is just an extension of an old and ongoing economics discussion around the impact of automation and productivity improvements.
The more important question about jobs that will be automated is whether there will be new jobs to replace them. Essentially working people are asking firstly whether their job will be replaced and secondly if their job is replaced will there be other, preferably better, jobs created that they can reasonably do. The worst case scenario is automation leads to a class of permanently unemployed people because new jobs fail to be created after old jobs are automated.
The reality is that until artificial intelligence is as smart as humans, it is going to change jobs in the way that all innovative technologies change jobs, it eliminates some tasks but adds new tasks to be done. Workers in an affected industry will need to update their skills, but their jobs won’t be eliminated. As is the case with all innovation, the bigger risk is that less people are needed to do the same job (or achieve the same or better outcome) so from that point of view the job is eliminated for some people.
Since the dawn of time tools, including machines, have been invented to increase human productivity. This has meant that over time many jobs have disappeared, but that has open the door to new jobs. It has also meant that people have become richer as the cost of the old essential products and services has dropped and new products and services, some now essential, have been invented.
It’s basic economics, and most people know this. A middle class person today is much better off on most dimensions than a king living a few hundred years ago. When a manager of a construction project boasted that he had saved jobs by having the men use spades and not machines, he was asked “why not take away the spades and give the workers tea spoons?”. It’s hard to argue that forcing people to be less productive helps anyone except those directly affected by the relevant innovation (and only in the short term).
Of course, people argue that what we are now facing is a singularity. Once artificial intelligence (AI) reaches human levels of capability on important dimensions, there will be no work for humans left to do.
There are many people with a powerful vested interest in hyping this perspective. Warning about the coming dangers of or making incredible claims about AI generates clicks and likes, it creates publicity for individuals and companies. It is a topic that people will pay to know more about because it is at the same time fascinating and threatening. These kinds of moonshot ideas motivate employees and generate sales. But it is a real concern?
The answer is yes and no. If it were true that we were able to create a general artificial intelligence then this would definitely be a concern. And in this case the concern for jobs would rank lower than other concerns (such as about AI dominating humanity) given that productivity would explode and it would be a world of abundance.
Even if general intelligence is far away (which I believe it to be) it is also correct that people need to understand the implications of what they are doing with AI to make sure there are no unforeseen consequences of how the technology is implemented. If you are relying on computers to write the rules (i.e. not explicitly programming them) then you need to make sure that if these techniques are applied to any mission critical systems, people understand and account for the risks (as needs to be the case with every technology used for a mission critical system). This can a justification for hyping AI and the need to regulate aspects of AI, however without achieving general intelligence for AI there is no binary threat to jobs. More on this later.
It should be noted that concerns about job automation and fears about the end of work is partly due to an predictable failure of imagination. We can see the jobs being lost but cannot imagine what might replace them. Who could have foreseen information technology as an important job category before computing became mainstream? Who could have foreseen all the jobs around social media before social networks became mainstream?
There is an element of a leap of faith in believing that productivity gains will lead to a better life for all, understanding that any changes will impact different sets of people, so not everyone will be better off after the change but the majority will be.
No one can dispute that computers have created more jobs than they destroyed and we are all better off for it. Jobs have not been destroyed by computers, workers have been able to accomplish more with the use of computers. This will be the same for how AI, based on how bots are programmed.
When will robots take over?
Yes, we do need to consider the special case where AI reaches human levels of comprehension, but is this going to happen anytime soon? If AI reaches human levels of comprehension then the impact is going to be massive. That is true. If AI is simply an improved automation technique the impact will be much reduced and will lead to great prosperity for all. I should mention that if AI did reach human levels of comprehension this could just as well lead us to a heavenly future as a dystopian one.
There are some people who argue that it will reach human levels of comprehension by 2029 such as Ray Kurzweil who works for Google. There are others that argue that it will never happen with the current silicon based technology such as George Gilder. Obviously we know that intelligence and consciousness is possible, because it exists in humans, but it’s likely we are underestimating the complexity involved, the novelty of our intelligence and the ability of our silicon based systems and technology to replicate biological processes. This applies even if we are assuming exponential progress on the data, algorithms and processing power. The answer to when will robots take over is not anytime soon.
If we exclude the case where the AI reaches human levels of intelligence or at least say that this won’t happen in the next 20 years, we can address the more pressing question as to whether an AI using today’s rapidly advancing technology will rob you of your job.
- The first mistake is the framing of the question. The question should be what “tasks” will a robot take rather than what “jobs” will a robot take. Of course when we talk about robot here we are referring to the physical robot as well as the AI software that can provide software based services.
- The second mistake is not asking how will AI change how I do certain tasks. In many cases it won’t replace a task in full but just improve the way it’s done in collaboration with the human.
- The third mistake is not asking what new tasks / jobs will it enable?
The real question here is what is the essence of the job, and do people face unnecessary friction in accomplishing tasks. The answer is undoubtedly yes. Imagine you could say to an AI “prepare a presentation on X, that is 8 pages long and has a chart on Y” and it instantly creates it for you and you can tell it the modifications you want. The process takes 5 minutes instead of 3hrs. An even bigger time saver could be an AI that helps you decide which tasks are important. How much time is wasted spending time and effort doing something well and later on finding that what you did was not needed?
The points above apply to any new automation gain of course. To answer specifically the questions about what tasks will change and what tasks will be enabled where AI is concerned, we need to understand the way in which AI algorithms work.
Essentially AI algorithms are ways of getting computers to do certain tasks without explicitly programming them to do so. The AIs are trained to make the connection between the inputs and the required output without explicitly programming what the connection is (or by partially programming and having the AI extrapolate from there). If you want an AI to identify cats in photos, you don’t need to manually program in the cat features such as ellipse shaped eyes, pointy ears, whiskers, but you simply show the AI millions of pictures of photos with and without cats and it will figure out how to identify cats.
How it does this, compared to human intelligence, is not very “intelligent”. It is a brute force algorithm that needs a lot of data. What it does is weight the significance of groups of pixels in the picture in relation to each other to find a pattern that identifies the cat. By iteratively testing layers of these weights (called a neural network) or using other similar techniques it can create a calibrated algorithm that can accurately identify cats, even edge cases of cats where some important feature is missing. This is powerful as it would be impossible (or extremely time consuming) to try to program something like this manually. With huge amounts of data and lots of processing power, it’s possible to create a kind of brute force intelligence.
These kinds of algorithms are very useful when plenty of data (preferably highly structured) data is available. In order to train the algorithms it’s also needs to be clear whether a given iteration of the algorithm improved the result or not versus previous iterations. If the level of relative success is not easily or instantly measurable (ambiguous) from one “guess” to the next then this can be a difficult problem for AI. This often is the case for human tasks where there is no right answer.
If data is scarce or the solution to the problem lies outside the data these are also difficult cases for an AI. The problems that an AI finds difficult however are exactly the problems that humans are good at solving.
For example, even though there is plentiful data on human conversation, everything uttered by a human has a potentially unique context in terms of the history of the specific relationship, the history of the conversation and the situational context. The further you go into the history, the more dimensionality and the more difficult it is to train the AI. That is why AI solutions are best focussed on a narrow situational context for chatbots (for applications other than one off superficial answers). Imagine deciding what to say next based on the fact that the previous five things you said were similar to five things you said consecutively in another conversation two years ago. This gives you some intuition of the problem.
To figure out what tasks are “at risk” you need to figure out to what extent they can be automated using the above described techniques.
What is clear is that in the vast majority of cases AI will improve productivity most commonly by being a complement to humans. The human + AI combination will be much more powerful than either human alone or AI alone.
It is true there might be certain tasks, such as truck or car driving on certain routes, that may be completely automated, however even in that case a human presence may be required for the unanticipated case, such as in the case of a breakdown, an accident or a security incident. It may turn out that driverless trucks without any human presence may be extremely easy to rob.
A plumber may have an app that helps diagnose issues, but the plumber will likely have to fix the issues themselves.
Automation has already created a world where there is more focus on experiences and entertainment than there was in the past, and that trend is likely to continue. People go to restaurants, have bigger weddings, more exotic vacations and consume experiences and entertainment more than they did in the past, and AI will perpetuate this trend. More and more jobs will be created in the “Experience” sector.
While the effects of increased productivity are positive, it is undoubtedly true that more automation and globalization with result in more winner take all effects and will increase inequality in the world. Unskilled and semi-skilled workers will have improved lives on some metrics, but will likely continue to lag versus highly skilled workers. This can have a negative impact on communities and politics unless the issues are addressed.
Accelerating productivity improves the lives of everyone on the planet, as even those who will benefit the most financially and get very rich from the new innovations will only be able to capture for themselves a small percentage of the value of the create for society as a whole.
Transitions to the new economy will need to be managed to ensure that no group of people experiences a catastrophic decline in their living standards as productivity accelerates.
It is important to note that everything said above applies to all innovation, not just innovation associated with AI. All innovation has an impact on the labor force and requires that workers get additional education to stay in employment. Educational institutions need to adapt curriculums including university curriculums to the changing requirements of the job market. We have seen this process play out continuously as automation has transformed the world in particular over the last 200 years.
For the foreseeable future, science will progress and machines will continue to compliment human workers and make them more productive. A machine will not be able to replicate the novel intelligence of humans to deal with unique situations or devise surprising solutions. Machines also cannot replace the feeling of a human connection that is important in many industries, from health to recreational industries.
AI will be able to effectively eliminate drudge work and remove friction where the right conditions exist (lots of appropriate data). Like historic productivity gains, the productivity gains from AI will continue to improve the lives of everyone in the world directly or indirectly and will create new products, services and jobs not yet imagined.