451 Research – Rise of the machines, Part 2: AI and the labor market

A HAL 9000 run amok entered the public consciousness back in 1968, but as the 451 Research Group notes in its two part research series “Rise of the machines,” technology’s impact on human labour is a subject that has been with us since Luddite attacks on their own version of tech disintermediation – the textile mills of the first Industrial Revolution. Has anything changed, and does AI represent a new threat that, unlike the machine mills, will ultimately replace rather than temporarily displace humans? Debate over the ability of machines to mimic human consciousness has raged for decades, with one side arguing that computers are only as smart as the programmers that feed them, and another pinning true intelligence to the wall with the notion that machines must ‘think’ independently in order to approximate the cognitive capabilities of their creators. Certainly, recent technological development and commercialization of AI solutions would suggest the scale is tipping in favour of the latter, as sheer computing capability reaches new levels of accessibility and affordability, and as complex algorithms create self-learning and self-healing attributes in AI solutions.

A key difference lies in the type of labour that is affected. As 451 researchers Owen Rogers, Nick Patience and William Fellows argue in this series, while the Industrial Revolution replaced physical labour with machines, AI is now impacting white collar or intellectual labour: just this month, InsightaaS ran a story this month on the use of algorithms in the legal profession to mash through thousands of cases in time frames that humans cannot compete with and that produce greater predictive accuracy, and this morning’s media deluge brings announcement of DeepCoder, a joint project developed by Microsoft and Cambridge University that uses program synthesis to write code without break, putting together new and unique code combinations at a speed that developers cannot match. Herein lies the rub; if increasing automation means machines are increasingly doing a better job with white collar work than humans, can education and reskilling keep up to keep human labour relevant? In Part 2 of Rise of the machines reproduced below, Rogers, Patience and Fellows consider this question and others that need to become part of the industry, economic and social discourse around artificial intelligence. (ed.)
From autonomous cars to cancer detection, computers are increasingly matching human performance on both manual and cognitive tasks. In the first report in this series, we examined whether improved technology has benefited the labor market. In this follow-up report, we discuss the impact of improved artificial-intelligence (AI) capabilities on the labor market.

The 451 Take

As AI development progresses, humans will need to improve their intelligence and skills beyond those of computers to ensure there are still decent, well-paid jobs for people to do. This isn’t impossible given the expanded access to education via the internet and improved schooling, but nothing is guaranteed either. The old Luddites’ belief that there was a finite amount of work proved false, but with robots able to both act and think, why won’t many new innovations utilize robots instead of humans? Historically, new innovations needed new labor to build and develop them. But if machines have reached levels of specific task intelligence – where they can learn to build and develop and even self-heal – then what will displaced humans do? Companies and governments will need to support training and development of new skills, and make it easier for employees to change jobs and collect those new skills. Perhaps education should be focused on areas that computers have been unable to address adequately in the past, rather than on jobs computers already excel at – for example, we should be prioritizing critical thinking and creativity over arithmetic. Over the longer term, societies may need to re-evaluate their structure, and new economic and political models such as ‘basic income’ may need to be adopted to handle increased automation without disintermediating those who are ultimately the consumers for these new automated products.

Owen Rogers, research director, Digital Economics Unit, 451 Research

The Industrial Revolution was truly a ‘revolution’ in the sense that it enabled humans to expand beyond their physical means. Before the Industrial Revolution, a single person could only grow as much as he was able to plant manually, based on his own physical energy. With the burning of fossil fuels and the use of water power, though, energy for products and services was vastly expanded, meaning we could do more for less. In a similar manner, AI becomes a challenge for society when it allows us to expand beyond our intellectual means.

The Industrial Revolution, while not providing an immediate benefit for all of those that lived through it, has surely benefited all of society in the long run. Humans have generally had to become more knowledgeable to survive in a world where manual tasks can be automated, and we have never been more educated. Education is vital because we must be able to perform tasks better than a machine that costs less to do similar work – otherwise rational economic behavior means that the machine wins out. If a human can produce a widget at $3, but a robot can do it at $2, the robot is the better economic choice. On the other hand, if an architect can design a house for $1,000, how much would it cost to build a machine to design a similar house? If it’s even possible now or in the future, it would cost a fortune.

The economics here lead to a bifurcation – the very low skilled are likely to keep their jobs, as the technology to replace them may not be worth the cost. The very high skilled leaders are also likely to keep their jobs, since they will often be the ones directing and managing use of the new technology. It is those in the middle of the skills curve who could be left high and dry as machines get more intelligent.

Nick Patience, research VP, Software, 451 Research

The globalization trend has impacted this economic discussion. While robots may win out against humans in markets with higher labor costs, what if a human can produce the same widget for $1 in China or India due to lower wages? Employers will generally pick the lower-cost option if it is effectively equivalent. People often talk about buying local, but realistically the majority of us are slaves to a bargain, and cheaper prices drive growth. Does this mean the low-skilled will keep their jobs, but only in countries where they provide cheap labor?

AI is presenting a whole new challenge, and again it is an economic one, much like the overall replacement of unskilled labor with skilled labor in the post-Industrial Revolution age. Technology might be good at inserting a screw into a panel faster than a human can, but they are not as good at identifying where that screw should go in a car’s new design specs, for example. Ultimately, it is our intellect and capacity for creative thinking that meant society prospered following the Industrial Revolution.

In the short term, the improved skills of machines and services will surely pose a threat to jobs. Why hire a translator for $10, when you can use Amazon’s Lex API to do it for $3? Slowly and surely, some skilled jobs will become redundant. In this example, there is likely to be some benefit in paying more for a translator – a person is likely to understand colloquial language, accents, idiomatic phrases and specific characteristics of languages better than a robot, and the translation will likely be more polished as a result. Value does comes into play in consumers’ and employers’ choices, though – if only a rough translation is needed, why not choose the cheaper Lex API? if you want advanced skills, then hire a translator. Radical shifts are happening in human-computer interaction because of machine learning. If you can talk to a computer or issue instructions and get answers, then you don’t need a screen or keyboard, and it can be embedded in more or less anything. What’s been holding us back from this advancement is the quality of the speech recognition powered by machine learning up until now. But with deep learning, that has changed radically in the past year alone.

The real AI revolution will come when AI can do jobs better than humans, and at a lower cost. For example, driving skills such as instinct, threat detection and awareness have traditionally been stronger in humans than machines, but Tesla’s autonomous cars, with their Autopilot feature, might suggest that machines are closing the gap or even getting an edge. There are many such example use cases. How many web designers will be put out if work if a machine can anticipate what a ‘good’ website looks like based on historic data? How many doctors would be put out of work by IoT monitoring devices and intelligent workflows? If one can pay $100 to see a doctor, or $10 to use a machine, many would likely opt for the machine, at least in the first case. To wit, some technology, such as IBM Watson, is already alleged to recognize tumors in large volumes better than oncologists can.

Some of these examples may seem unlikely today – but in a generation, there are few reasons why these and many others couldn’t be real use cases. Machine learning acts as a force multiplier; it can make individuals and small groups of people considerably more powerful. We are already seeing this in medicine, legal and other sectors. Most jobs, ultimately, involve decisions based on rules and experiences. In the computer age, setting these rules has been the domain of computer programming, while much of the experience that most people’s day-to-day decisions are based on can now be obtained through machine learning and AI. Of course we’ll still need drivers, web designers and doctors, in the same way we still need factory workers today – but we will likely need fewer. Even management isn’t out-of-bounds for the coming age of AI. In a recent study, 32% of US workers stated they’d prefer an unbiased computer program to be their boss rather than a human.

William Fellows, research VP, 451 Research

Meanwhile, the rules and experiences that humans have developed over years sometimes reveal themselves to be suboptimal. Last year Google’s AlphaGo computer beat world-class player Lee Sedol at Go, a Chinese game requiring the use of abstraction and strategy. What is interesting is that the computer chose a different strategy over the expected choice of champion Go players. Jie Ke, the number one Go player in the world wrote, “Humans have been practicing and playing Go for thousands of years, but the computer now tells us that we are all wrong.” Who is to say that computers might, in the long run, actually be better at things we thought we had mastered by addressing them in completely different ways?

Once machines can do things beyond human capability at a fraction of the cost, what jobs can persist? Will we all become programmers of robots? Will we all be cutting-edge innovators? Chances are we will need these roles, but certainly not billions of them. One obvious role for the majority of people will be as manual workers at a pay level that is so low that it’s effectively cheaper than using machines. Is this good for society? Determining how society will operate once computers are better and cheaper than humans at a majority of tasks is a challenge facing civilization in the long term.

The Luddites’ assumption was that machines do ‘easy’ work, and that the definition of ‘easy’ expands as technology progresses – this means the work beyond ‘easy’ typically requires greater brain power, creativity and insight than a computer can deliver. But robots equipped with artificial intelligence are increasingly able to perform work that previously required human intelligence. Google’s Vision API can scan photos at rates of hundreds per second, identifying locations, content and even the emotions of people in the photos. Humans might be able to do the job ‘better,’ but not at the same rate and not for as low a cost. When it comes to image analysis, there are more jobs classed as ‘easy’ for a computer to perform than there were a decade ago.

The cycle will continue. Companies will innovate, processes will improve, prices will fall and demand will increase. AI and a host of other things will play a part in all of this. Cloud will have an impact as well. It is easier and cheaper than ever to take a risk on building something, even if it doesn’t pan out. The mill owners of the Industrial Revolution took big financial gambles investing in steam or water power – nowadays, a developer can build machine-learning applications on the cloud for pennies. This will further drive jobs to machines using AI.

Some people will certainly be left out in the cold as a result of this transition. Companies need to consider how their technologies will impact society as a whole. This isn’t social altruism; this is common-sense economics. Companies need to support retraining and innovation if they want to survive in the long term.

If enterprises are reducing labor requirements through cloud computing, they should consider who will support the next trends and the next wave of technology. It’s better to retrain the staff you have than to pay a premium for staff from somewhere else. If supermarkets lose their cashiers, they will also lose the profits from those cashiers in-store purchases. Supermarkets considering Amazon Go should provide staff retraining wherever possible, even if they move elsewhere – it may pay dividends in the long term.

In 1930, John Maynard Keynes predicted that within 100 years, people in the richest nations would be working only 15 hours a week. Working hours have certainly come down, but Keynes’ prediction has yet to come true, simply because we have found more and more ways to create innovation and growth. But with machines able to beat humans in terms of intelligence and dexterity, there might be a time when not everyone needs to work. This notion is so different from our current day-to-day lives that it’s difficult to imagine, but a ‘basic income’ delivered to every citizen by the government would allow those who cannot find work or choose not to work to survive, while giving those with motivation and talent the opportunity to earn far more. This is a controversial proposal, but it has big-name proponents from the right and left in various countries.

This report and its companion piece have concentrated purely on the developed world. The implications of technology and AI for emerging economies could be even greater. What would happen if automation renders inexpensive labor in China and India obsolete? 3D printing could also be a threat to low-paid assembly-line workers. Mass unemployment and political instability could result from replacing these positions with automation.

The world will continue to turn, and society will continue to innovate. The major long-term challenge is in determining how everyone can benefit from this innovation as robots become more intelligent, and there are ultimately fewer opportunities and incentives for humans to work.