Recently, we have experienced a significant heat wave in England. The temperature soared well above what is normal for June and air stood still for several days. England is not a country where many residential dwellings have any inbuilt air conditioning, so buying a portable fan or an air conditioner was the only option for those who wanted to be comfortable in the scorching heat. The problem was, however, that most of the shops have completely ran out of these appliances in a matter of days. When the stock was finally replenished and prices were put up, the heatwave was over and shops ended up with piles of cooling appliances that nobody wanted to buy any longer.
Most of the stock management and pricing decisions in retail are made by humans and this is what lead to fans being sold out way too quickly when the demand for them was at the highest. This made me think of how this situation would be managed if such decisions were delegated to an artificial intelligence.
Pro-actively evidence-based stock control
This is how I think the situation would be handled by the AI. High demand for fans and portable air conditioners would be anticipated based on the weather forecast and historic sales records. The shops would have been well-stocked by the time the heatwave would even begin.
Due to anticipated high demand, prices of the units would be brought up. This would generate more profit for the stores; however this is far from the only reason why prices go up in well-functioning free-market economy. High prices play a role as a resource control, which means that as a particular resource or commodity becomes rarer, its price increases, which leads to less purchases being made, which, in turn, makes it easier to replenish the stock. In the context of portable air conditioners, this would prevent those people who don't badly need them from buying them, leaving more stock available for those who either can't sleep due to the heat or are suffering from health issues.
Toward the anticipated end of the heatwave, the prices would be adjusted back down, causing a large proportion of excess stock to be bought. So, at the end, all of the customers who needed a cooling device would have one, shops would have made a noticeable spike in profit and there would be no need to organise a costly logistical operation to get rid of excess stock.
Of course, there are many more problems in retail that AI would be able to solve easily. For example, there was a particularly warm September and I was looking to buy a barbecue. I have visited several high street shops, but all of my efforts to find one were futile. The reason for this was that, according to retail calendars, barbecue season ended on September the 1st and all remaining stock got sent away to make way for Halloween and Christmas stock.
All of this is because, for humans, it is the easiest to work with arbitrarily defined dates, such as the 1st day of a given month. Seasonal changes of stock are pre-planned well in advance. There are just too many variables to make effectively stock-management decisions responsively and the price of getting it wrong is way too high. AI, however, would be capable of working with large variety of data in real time and make dynamic stock-management decisions accordingly. Therefore, a store managed by AI is likely to keep summer stock on display while the weather is likely to stay warm and is unlikely to bring out Santas and raindeers on the 1st of October just because the retail calendar says so.
Forget about signal failures
Retail is not the only sector that would work much better if AI would be allowed to gain a significant amount of control within it. Mass transit systems across the world could certainly use some intelligent automation.
Anyone who lives in the UK is aware how bad the railway transport is. The country-wide railway network has a strong potential to be one of the best public transport system in the world, as the network itself is well-structured and timetables are well-planned. However, the problem is that way too often the system doesn't work according to plan.
Anyone who regularly commutes by train in Britain is familiar with the dreaded announcement of "the train has been cancelled due to a signal failure". Safety rules insist that non-functioning signal on a rail line should be treated by the train drivers as stop signal. Therefore, a broken signal will stop all trains on a given section of the tracks until it is fixed. The problem is, however, that, due to excessive complexity of the network and due to the fact thar repairs are managed by a limited number of engineering teams, most of signal failures have to be managed reactively rather than pro-actively, therefore delays are unavoidable.
The process can be made much more efficient by introducing the AI. Initially, there would be no requirement to even connected every signal to AI-controlled network or replace human engineers with robots. All that would be needed is to continuously collect the data on when each component has been replaced and which components malfunctioned. Based on this data, learning algorithm of the AI would be able to pro-actively determine which components are the most likely to break next and send the maintainance teams to replace them before they break. Due to the nature of learning algorithm, the accuracy of such predictions will increase over time and eventually there will be a minimal amount of interruptions for the railway passengers.
What about jobs being lost to the AI?
This is the question that worries the AI skeptics the most. The fear of losing jobs to the AI is so real, that trade unions are even prepared to call strikes to prevent any increased automation from happening, regardless of how gradually the changes are being introduced. In England, for example, workers of Southern rail company have organised a series of strikes to oppose a relatively minor step towards increased automation - introduction of driver-only trains where no conductor is required. Striking workers cite safety concerns, despite the fact that the technology on newly introduced trains allows the driver to operate all the doors on the train safely. The train drivers have even rejected a substantially large pay rise offer, which would bring their base salaries to more than double of the average pay within the country while their work week is only 4 days long. So, the strikes are certainly all about opposing any sort of increased automation and not about improving the working conditions. If the job of conductor is slowly being made obsolete, it will be only a matter of time before the drivers are also replaced by the machines, so the train workers are determined to prevent even the first stage of automation from ever happening.
However, loss of jobs is not something that people should be afraid of when more processes are automated. Previously, I have written how recent technological advances have created numerous non-traditional ways of making a living for those who are willing to think out of the box. Anyone these days can stars a blog or a Youtube channel and make money without having a formal job or running a traditional type of business. However, the good news for those who want to remain in the comfort zone is that traditional jobs will not go anywhere when more processes are delegated to the AI.
It is not only that automation has been gradually happening for several centuries and made professions like blacksmith obsolete. Based on recent technological advances, we can confidently predict that automation will result in more jobs being created than jobs lost. For example, the term "computer" was originally applied to a person who was hired to perform complex calculations. However, it doesn't matter how good someone is with maths when modern computers can perform much more complex calculations than any person can handle and do it millions of times faster. With the invention of computing machines, not only the profession became obsolete, but also many new, previously inconceivable, uses of such an enormous and cheap computing power were soon discovered. This lead to creation of whole new industries. Now, many offices across the globe are staffed with an army of people using computers to perform jobs that wouldn't even exist if there was no such thing as a computer.
Any system controlled by the AI would still require some humans to operate it. Taking this into account, let's consider a hypothetical scenario where AI-based system that requires one specialist to operate has replaced a process that was previously performed by 5 people. In the short term, 4 work places have just disappeared, but this has only happened in the short-term. What happens next, however, is interesting. The whole process becomes much more efficient and cost-effective, as even if the salary of the AI operator is double of what previous workers were paid, it is still cheaper to hire this person compared to what it costed to hire 5 workers. This leads to increase in productivity and an opportunity to substantially reduce prices for the end users. This, in turn, leads to the surge in demand and prompts the organisation to introduce even more AI-controlled systems to increase its production, each of which is operated by a specialist professional. So, within a certain amount of time, efficiency brough by the AI helps the organisation to expand and creates more jobs than were originally lost.
Introduction of self-service checkouts didn't cause checkout operators to lose their jobs. However, it enabled fast movements of queues and allowed more people to visit supermarkets at a given time. Introduction of self-driving cars may be on its way to eliminate the need for the driver, but it also created many IT jobs in the in industry that used to be very far removed from the IT.
Of course, some types of automation do result in people being made redundant and some people are not in position to easily recover from such a job loss. However, every other process within the economic and political systems produces winners and losers as well. It is just an inevitable fact of life and nothing can be done against it. Compared to other processes, automation usually produces far more winners than losers.
Aren't software developers making themselves unemployed?
Many software developers fear that their profession is one of many that will, one day, be replaced by the AI. The concerns are not unfounded, as AI is already capable of performing tasks that developers perform. For example, there is an intelligent software that is able to analyse an image of a user interface and generate a markup in a declarative language, such as HTML.
However, based on the evolution of software development processes, there is no need to fear. What developments in AI really mean is that programmers of the future will no longer have to perform mundane tasks, such as writing their own web page markup.
Early software developers did not write software in high-level programming languages like Java. Instead, they wrote low-level instruction sets for a specific processor architecture in assembly language and had to manually translate these commands directly into the machine language consisting of ones and zeroes. The process was tedious and highly error-prone and the code wasn't easily readable by those who haven't written it. Today, a compiler software will translate a relatively easily readable code into assembly instructions and runtime environment will translate those instructions into processor-specific machine language commands.
When the job of a software developer mainly consisted of writing low-level assembly commands, it was easy to argue that introducing high-level programming languages and an automated compiler would leave many of programmers jobless. However, the opposite happened. There are now more software developers than ever and the software that is being produced is far more complex than anything that could have ever been written in assembly language directly. Modern software developers don't need to worry too much about the processor architecture and are able to fully concentrate on the high-level functionality. The salaries of the developers remained high due to ever-increasing demand for their work, but, at the same time, it became cheaper to higher them in terms of productivity increase.
There are many tasks in both software and computing hardware industries that are completely impossible to perform without a high degree of automation. For example, due to an excessive complexity of modern CPU chips, it is impossible to design one by hand. Hardware definition language is used to define the architectural logic, which is then compiled into a definitive processor design.
As advanced AI will automated many more processes than are automated already, developers will gain freedom to invent new technologies that aren't even conceivable at present. Those technologies will require new trained professionals, which means that neither the jobs are going to disappear, nor the salaries are likely to go down.
Of course, there is a catch, but not a huge one. Any good software developer knows that, to succeed in the industry, you need to constantly adapt to new technological developments and never stop studying. Anyone who follows this formula will be fine when AI takes over the tasks that they currently perform on daily basis. The only types of programmers who are usually made redundant are the ones who have learned COBOL 30 years ago and only ever wrote software in it, until their employer decided to replace all the legacy software and their skills were no longer needed.
With AI, there will be no shortage of programming tasks, but modern programming technologies may become obsolete. Therefore the best bet is to keep your knowledge up to date. Although technological advances happen fast, the changes are incremental rather than revolutionary. Therefore, if you keep studying, it will be unlikely that you will be caught by surprise.
Written by Fiodar Sazanavets
Posted on 30 Jun 2017