Under capitalism, there is a constant threat that the automation of work tasks will lead to layoffs, the deskilling of work, further alienation of workers from the production of goods or the provision of services, or – sometimes – safer or easier work for those lucky enough to keep their jobs.
Automation is part of the process of capitalism and the turbulence it causes is not dependent on larger economic shifts. However, large economic movements are like tides – they move in spite of the rough water on their surface – but can make their surface effects be felt more acutely.
We are in one of those high-tides now. Terms like automation, artificial intelligence, machine learning, robots, and the hyperbolic threat of the replacement of all work is in the headlines and trending on social media almost daily. As such, we could all benefit from revisiting the reasoned analysis by unions and the democratic positions taken by working people in the face of changing technology across recent history.
There are a few new concepts, a need to update some of the language, and specific issues when it comes to the threats from automation that need to be added, but the basic analysis and response is sound. There are benefits that could be brought if we were under a different economic model and real solutions that could be put forward in organized workplaces or through legislation/regulation.
While the foundation of the analysis is solid, it is easily forgotten by those caught up in the current zeitgeist, who want to believe the dreams of automation put forward by the charlatans and public-relations machines of capitalism.
This is not to say that automation is not real or not a real threat to millions of jobs. It is also not to say that the new wave of artificial intelligence should not be engaged with rigorously. There are threats to society, the economy, and work. But activists must understand the underlying causes of these changes if we are to defend working-class communities and values, and empower workers to advance progressive solutions.
Some of the issues, definitions, and responses are discussed below.
Organic Composition of Capital
The process of replacing workers with machines, robots, and automated processes is constant under capitalism. However the replacement of workers with machines becomes more noticeable to the general public under certain economic conditions (i.e. when debt is generally cheap or local labour is generally expensive). We seem to be in one of those moments now.
Many economic commentators have written about the deindustrialization of historically industrial heart lands; the gutting of employment in industrial sectors; or the replacement of industrial manufacturing with the service economy. In Canada, this has included articles written since the 1980s about industrial southern Ontario. They lament the massive decline in union density, mass layoffs at large plants, automakers, and forges.
However, actual productive output (the amount produced for sale) by private industry in Ontario is greater now than at most any time in the previous 50 years. Employment has stayed relatively flat after a brief decline in the 1970 and early 1980s. Employment has actually grown since 2009. It is hardly a significant de-industrialization if production has grown, but it does feel different from the days when large numbers of workers worked in large centralized production facilities.
The reason why employment numbers have not grown at the same rate as production/output and why large plants have closed is the introduction of new technology and automation in many production facilities. In many ways, the production process resulting from the introduction of machines is often more industrial.
High-tech production has often driven more job loss than the so-called economic condition of de-industrialization or a shift to a service economy. The reality is that large central plants have closed, but there has been a slow proliferation of smaller production facilities that employ fewer workers and are harder to unionize.
However, the replacement of workers with machines on the shop floor does not explain all the reductions to employment (or flat employment growth) in communities and industries locally. Machines, robots, and software that allow for automation are made somewhere. And, the real reason for reduction in employment is the movement of jobs to where the machines are produced that allow for automation related to production.
To understand this process, we must introduce the concept of the Organic Composition of Capital: a measure of the investment of private capital in the replacement of workers with machines – specifically in the context of production.
There is a long-identified general tendency in capitalism towards replacing workers (organic variable capital) with machines/tools (inorganic constant capital).
The reason for this tendency is competition between firms in the attempt to increase the rate of production and efficiency so they may reduce the per-unit costs and increase profit. This is done by investing in automated or machine-supported processes.
The main thing standing in the way of this is firm-level economics. These local economic issues (such as whether the capitalist will make profit or not) drive capitalist decisions to invest in machines to replace their workforces. It takes a lot of money up front to buy machines, and that usually means taking on debt. If the owners of the plant do not think that they will make enough revenue to pay off the debt – or predict a reduction in future profits – then they will not borrow to finance this investment. If debt is cheap, local labour is expensive, and machines are made where labour is inexpensive, then the decision to buy machines is made easier.
If a car assembly spot-welder job is replaced with a computer-controlled robotic arm, then we consider that automation. The job on the assembly line is lost, but the robotic arm, its components, software, and maintenance are usually produced elsewhere in the world. The loss of the job on the assembly line is simply offshored to lower-wage regions that produce the robot and its software. The tendency to replace a worker is made easier when the debt to finance the robot, and the cost of making the robot, are cheap.
Productivity and efficiency in the public sector
There are many confused concepts of productivity and efficiency as applied (or rather misapplied) in the public sector.
The tendency of the increased organic composition of capital applies to industrial production, but this does not apply in the same way to the public sector since competition does not exist in the same way inside government-owned monopolies. Instead, under neoliberalism, the public sector managers have been attempting to mimic capitalist processes with the introduction of private sector methods to the public sector. This includes the automation of some tasks in an attempt to reduce costs, the introduction of fake competitive markets within the public system, and through the privatization of public services.
Efficiency (i.e. the production of more products of the same quality in the same time) is difficult to apply to a public service. The bulk of public services involve public workers helping, caring and supporting other people through education, social services, and health services. A public service can try to increase the pace of work – a characteristic of the neoliberal era – but the result is a decrease in quality of the services instead of increased efficiency.
A classic example is the increase in classroom sizes resulting from cuts to education funding. The result is a type of "efficiency" – increased children graduating per teacher – but the quality of the education that the student graduates with is reduced, so there is no corresponding productivity increase like there is behind the private sector's decision to replace workers with machines. Most public services are weakened this way.
In many situations, the attempt to replicate private sector technology in the public sector leads to massive overruns in cost. Part of the reason for this is the same as in monopoly private sector firms that attempt large-scale automation and fail. The difference is that the government and public services cannot fail, and so we see ever-increasing amounts of money going into "fixing" the botched automation roll-out.
Some recent examples of massive cost overruns in attempts at automation include:
Accenture welfare system "upgrades" in Ontario.
IBM's "Phoenix" payment systems for the federal government.
Accenture Presto Card development for Metrolinx.
Front-line social services computer upgrades in Ontario.
Digitization of health records in the UK.
G4S tracking of inmates in the UK.
Upgrades of telephone services to VoIP systems in most workplaces.
Digital transition of document services in the federal government.
Shared services agreements in the federal government.
All these were attempts to make government cheaper to correspond with budget cuts, which in the end cost the government more than simply delivering the service the original way.
The exception to this rule is the implementation of software in the delivery and/or management of services. If implemented properly, this can result in an increase in the amount of time available to provide care. However, in the neoliberal era, most governments have used this investment in automation to reduce overall funding for public services – resulting in either the same services or worse services. In many cases, it was the cuts to funding that drove the implementation of automation services (like software and mimics of private sector management methods), so increased quality of service was never actually the goal – and certainly not the outcome.
Furthermore, this kind of automation increases efficiency on first implementation only. For instance, when the word processor was first implemented, it was a labour-saving device and replaced some work that clerical workers performed. However, when the word processor is "upgraded" there is no corresponding efficiency increase.
Commercialization of university, government/industry, and corporate research
Much of the research driving innovation of products and services in the private sector is initiated in publicly funded universities and government research labs. While millions of public dollars are used to finance these private sector developments and potential products, the public has little say over the application of these services, or the ability to direct them for the public good.
Staring in the mid-nineties, commercialization offices were set up in universities in an effort to make up for poor investment in research in the private sector. At the same time, government research focused on regulating industry was undermined. The result has been a deeper connection between private investors' desires for profitable products and services, and the direction of research in universities and colleges.
Additionally, commercialized services resulting from this publicly financed research are sold back to governments with little or no additional innovation added by monopoly firms – except for the packaging.
These new "innovative" services promise to be cheaper than current methods used in the public sector through increasing efficiency of work, automating tasks, downloading the work of public services on to the public themselves, replacing public sector workers with private sector workers who are paid less, and outsourcing work into the private sector. In most cases, however, the promises are never realized, or the resulting new methods cost more than any resulting efficiency.
Data, intellectual Property rights, and copyright
The large datasets that allow for automation are made up of data taken from workers and the public. When these datasets are under monopoly private control, it gives huge power to the owners of that data.
Artificial intelligence was invented by the public sector. Most research is carried out and funded by public labs at universities and defence services. However, the commercialization of these methods has resulted in their monopolization through strict intellectual property laws.
Copyright covers the computer code that drives the process of artificial intelligence. Even though the base of that code is developed in the public sector and released under permissive licenses, most of the applications written for the government or end-user are not.
The result is that the public (as consumers and users) have no say over the use of these automation processes, and the exact use of their data is kept from them. Additionally, IP rights keep the broader community from utilizing these datasets, and from regulating the corporations providing the services that allow for automation through software and artificial intelligence processes.
The concept of Just Transition describes a process whereby former modes of work are replaced by new processes and work methods, without negatively affecting the worker who is forced to move from a mode of work that no longer exists.
It is usually applied to large-scale shifts in employment such as a large plant closure, or entire industries closing or being replaced.
The change in the work can be from technological changes, changes to the employer, or even economic and social policy changes.
The principle is similar to bargaining for language dealing with technological change in the workplace, but applied at a much larger scale: Workers and their communities should not be impacted negatively by circumstances over which they have no control– including changes driven by the economic and social system we belong to.
In collective bargaining, protecting employees from negative impacts of technological change – the process whereby the employer implements new processes or methods outside the original job description – is done through protections of work and language for retraining.
Technological change in capitalism is driven by the need to compete in the market against new firms who have no sunk costs in old capital or methods (see the tendency for there to be an increase in the organic composition of capital).
Whenever a new firm appears on the market, they are usually not implementing old technology; they buy new technology and hire newly trained workers. This generates pressure on older companies to implement changes in order to keep up with these new processes and to stay competitive.
In the public sector, implementation of new technologies and methods happen when new managers are appointed, or reviews of business processes and resource allocation are done and compared with what they see as similar private-sector processes.
In both the public and private sectors, technology changes much faster than the workforce. Without a sufficiently strong collective agreement, workers are forced to navigate this shift in necessary skills through the market – usually resulting in layoffs and replacements.
This is highly inefficient at the social level: Both costly to the worker, and a strain on social services that support that individual, it can take many tries before the worker learns, then finds, a new skill that is in demand in the labour market. This can feed discrimination against older workers.
The union solution is to force employers (both public and private) to plan for technological change and fund retraining and upgrading programs for their currently employed workers.
Capitalism drives technological advances through the drive for greater profits. This drive is unending and does not always have positive social outcomes. As such, the broader implementation of technology and the resulting technological change should become a discussion at the level of government policy. A Just Transition for workers – both inside and outside the affected industry – is expensive. However, the costs of that transition can be used to slow the rate of change that has a negative effect on workers – if those costs are passed onto capital correctly.
Automation is a form of technological change in the workplace driven by the same forces that drive the need for retraining for new technologies: Competition between firms and the quest for greater profits. However, automation occurs through investment in production with the explicit goal of replacing a worker's labour (organic capital) with machines (inorganic capital). Retraining in this case will not save a worker's job.
Automation can happen in a variety of ways:
The direct replacement of labour with machines.
The replacement of repetitive labour supported by machines with better machines that require less human intervention.
The replacement of old methods with newer, more efficient methods.
The replacement of work with software that requires less human intervention.
Contracting-out and Outsourcing
Contracting-out of work happens when there is a clear replacement of more expensive workers with less expensive or more easily-managed contract workers.
Outsourcing, a type of contracting-out, is the process whereby production is completely removed from internal production processes, and the work (not just the worker) is moved out to a separate business.
In many cases where machines, robots, or software are replacing workers, it amounts to a complex program of outsourcing that moves the labour to build machines, robots, or software to areas with lower labour costs.
When the machines are not made where the original labour was happening, then it is not just a question of machines versus labour costs. Instead, it is a calculation of labour costs in one country or region versus labour costs in another country or region.
Recent studies (even from those who have previously promoted neoliberal free trade) have shown that economic globalization through free-trade agreements has facilitated the replacement of workers in high-cost (developed) regions with workers from lower-cost regions. The result is not an increase in overall production, but the contraction of economic activity and employment in "developed" industrial regions.
Contracting-out can also include the move of human work to software – software usually produced, maintained, and serviced elsewhere. This process has been referred to as “offshoring” since most software and robotics work is done somewhere far removed from the original labour it is replacing.
Artificial intelligence (AI) is the application of data, statistics, and algorithms to decision-making by computers and machines. In AI, a computer makes a decision (or guess) based on its already "known" data – that data is from previous experiences collected by its programmers.
There are several issues beyond the automation of work including the (re)introduction of biases based on the people who wrote the algorithms. Algorithms that learn – just like people – can make biased decisions when given bad information, or when there are intrinsic biases within the program. Regulations, best practices, and agreed-upon facts are still not mature in this field.
Furthermore, when the development of applications is driven by profit instead of need, then what the applications are designed to do with data, or the outcomes they are seeking, can be affected. There are many examples of this in software such as the recent federal government's Phoenix pay system that was "designed" not to allow basic, everyday changes needed to personal expenses. When these issues within software development are tied with artificial intelligence, you can amplify those intrinsic biases in the underlying software and its design.
Machine Learning is a branch of Artificial Intelligence (AI).
Machine Learning is the process of using large datasets, algorithms, and statistics with the goal of allowing a machine, robot, or software to be able to make a correct decision, and then learn from making correct or incorrect decisions by adding that experience to its dataset.
Issues in this field resulting simply from the way that machine learning works include:
Mass surveillance where the user or the public is unaware that their data is being collected or used.
Privacy violations based on the use of data from surveillance.
Natural monopolies emerging from the centralization of information and data that allow machine learning algorithms to work, resulting in monopoly control.
Perverse incentives for the commodification of personal information that has already been collected (even accidentally or illegally).
Private datasets of public data without regulation or oversight.
Regulation does not exist for most of the ways in which this data is being collected and used, or for the resulting applications produced with this data.
The ongoing debates about the process of upgrading our mobile technology, who owns the technology, what companies can provide it, what the threats are to the public, and the hype, offer a good example on the subject of automation.
The investment in "5G" mobile technology is mostly about facilitating automation. 5G is about reducing latency – lag time – between the two devices connected to each other. Reducing latency is about allowing self-driving cars, controlling robots, and making internet-connected devices used in production facilities more responsive. All of these uses are an attempt to replace workers with machines.
However, the implementation of this new mobile technology requires a massive capital investment and the benefits have been generally oversold to the consumer paying for the upgrades.
Issues that revolve around the current implementation process of 5G include:
Outsourcing of technical servicing jobs which replaces direct physical interaction with components with remote digital servicing.
Investment in automation without specific social benefit guarantees;
Facilitation of mass surveillance of the general public.
A frequency range that interferes with current technology, including weather stations.
The use of shifts in technology by monopoly private employers to outsource to "platform" economies.
Free Software and open source
Software controls the machines that are used to automate work processes.
In many cases in the office service sectors, it is software alone that automates the work.
When it comes to automation through artificial intelligence (AI), our mobile devices, or our use of the internet, it is the collection and use of the data that drives decision-making by the AI algorithm. The only way the public can be informed about, and regulate, this collection and use of their data is by being able to examine the software's source code. The alternative is to trust that the corporations seeking to monetize their monopoly ownership of this data will always have our best interests at heart. Recent history shows that this is not the case.
The best way to have an informed public, increase competition, and regulate the industry around artificial intelligence – especially machine learning and surveillance-driven data – is to make sure the source code driving the devices and network infrastructure for such activities is free software (or at least open source).
The Free Software movement established the framework for the open, collective, and freedom-supporting development of software. The movement is focused on making sure that software is not doing anything that the user has not consented to, and to allow the modification and distribution of the modified software.
Free Software principles applied to the computers and software that control our mobile networks, manage user data, and protect privacy and security would provide more security and privacy for users, as well as more individual and democratic control over the use of personal data.
There are many issues surrounding the implementation of the new generation of mobile networks, and the use of these networks to monitor users and to collect intimate data. The high-level threat of privacy invasion, and the sensitive nature of the devices connected to this network (such as self-driving cars and appliances), have driven these debates about network security.
The debates are focused on the building of these networks with components made outside the country. However, more important than who builds the technology is the ability to audit the software that runs the system.
Regulating this software through free-software rules would solve many of these issues. Making the software's source code freely available would also mean that the software could be hardware-neutral, allowing for more competition and freedom for users in how they connect to this new generation of mobile network – extending this freedom to what information they are sharing, when, and with what entities.
Previous union policy on technological change
Unions have had policy since the 1970s on technology change, how it impacts workers, and solutions to the negative impacts.
Reading through those policies, some major themes and responses emerge. They are summarized here.
All implementation of technological change should be subject to collective bargaining.
Collective Agreements should include language about technological change.
All expected impacts of the technological change should be presented to the union before such negotiations.
Technology should not be used or abused to increase monitoring of employees.
Workers' jobs should be protected and technological change should not be used to reduce employment.
New jobs created to support, or because of, technological change be included in the collective agreement.
(Re-)Training should be paid-for by the employer and/or government.
Reduction of work hours (with no loss of pay) should be negotiated as part of any increase in true efficiency or productivity leading to increased revenue.
Supports for workers, industries, and communities should be developed by all levels of government in conjunction with unions.
The union should be pro-active in identifying jobs throughout the union affected by technological change and automation.
Including education of membership.
Establish technological change committees in each workplace.
Include measuring equity impacts for all impacts of technological change and automation policies.
Combat the unequal application of automation that disproportionately targets certain equity groups.
Increase public and democratic control over implementation of automation.
Reduce dependency on outside contractors for automation and technological change.
Combat both deskilling and use of part-time/casualized workers as a result of technological change and automation.
Protect privacy in the face of increased ease of collection of personal data.
Protect the security and variability of information in transit.
Expand these demands to the developing world and oppose the exploitation of developing world workers in the production of software, machines, and robots that aid automation.
Ensure that the implementation of technology is for the public good, changes lives for the better, and does not degrade workers.