Automation's Future Depends on Social Choices Not Technology

The Robot in the Boardroom

Here is a fact that should unsettle anyone who has read a headline about artificial intelligence taking jobs: the machines are not the ones making the decisions. People are. And those people are not neutral.

Debra Howcroft and Phil Taylor, sociologists at the University of Manchester and the University of Strathclyde, spent years watching the automation debate unfold. They noticed something strange. Nearly every public conversation about robots and jobs treated technology as a force of nature. A tsunami. An inevitability. The machines are coming, we are told, and we must adapt or be swept away.

Howcroft and Taylor decided to test that framing. Their 2022 paper in New Technology, Work and Employment does not ask whether automation will destroy jobs. It asks a more interesting question: who is shaping automation, and why? The answer, drawn from decades of research on how technologies actually get built and deployed, is that automation does not fall from the sky. It is designed, funded, and implemented by people with specific goals, specific budgets, and specific ideas about who should win and who should lose (Howcroft & Taylor, 2022).

This is not a comforting thought. It is a useful one.

The Myth of the Autonomous Machine

The dominant story about automation is what scholars call technological determinism. The idea is simple: technology evolves on its own trajectory, and society must rearrange itself to accommodate it. The printing press changed religion. The steam engine changed cities. Artificial intelligence will change work. The technology is the cause. Everything else is effect.

Howcroft and Taylor reject this entirely. They draw on a body of theory called the social shaping of technology, or SST, which has been around for decades but has been largely ignored by people who write about the future of work. SST argues that technology is not an external force acting on society. It is a product of society. The choices made during design, development, and deployment reflect existing power structures, economic incentives, and cultural assumptions (Howcroft & Taylor, 2022).

Consider the self checkout machine. It is often presented as an inevitable response to labor shortages and customer demand. But the real story is more specific. Retailers chose to invest in self checkout because it shifts labor costs from employees to customers, because it weakens the bargaining power of cashiers, and because it allows for more granular surveillance of purchasing behavior. None of this was inevitable. It was a set of decisions made by people in boardrooms, shaped by the political and economic context of the last thirty years.

Howcroft and Taylor are not saying technology does not matter. They are saying technology does not make its own decisions.

What Actually Shapes Automation

The authors identify five forces that determine which automation technologies get built, how they get deployed, and who benefits. Each one is a choice, not a law of physics.

Existing Technology

Automation does not emerge from a blank slate. It builds on whatever is already in place. A factory that already uses barcode scanners and inventory management software is more likely to adopt warehouse robots than a factory that runs on paper logs and hand signals. This creates path dependency. Early decisions constrain later ones, not because of technical necessity but because of sunk costs, training requirements, and compatibility issues (Howcroft & Taylor, 2022).

This means that the automation we see today is partly a function of the automation we built yesterday. It is not the best possible automation. It is the automation that fit into existing systems.

Economics

This is the obvious one, but it is worth stating clearly. Automation is expensive. Companies do not automate for fun. They automate when it promises to reduce costs, increase output, or open new revenue streams. But here is the twist: what counts as a cost reduction depends on what you value. Labor is expensive in countries with strong worker protections and cheap in countries without them. The same robot that makes economic sense in Germany might make no sense in Bangladesh, not because the robot is different but because the social context is different (Howcroft & Taylor, 2022).

The authors point out that automation is often sold as a response to rising wages, but wages themselves are a social choice. They are determined by minimum wage laws, union strength, and labor market policies. When a company says it must automate because labor is too expensive, it is really saying that the social arrangements that make labor expensive are not acceptable to the company.

Social Relations

This is where the analysis gets sharp. Workplaces are not neutral spaces where efficiency is the only goal. They are sites of conflict between workers and managers, between different departments, and between different levels of hierarchy. Automation is often deployed not to make work more efficient but to make workers more controllable.

Howcroft and Taylor cite examples of warehouse management systems that track every movement of every worker, not because this data is needed for production but because it gives managers leverage. The technology is designed to break the informal autonomy that workers have carved out. The stated goal is efficiency. The real goal is control (Howcroft & Taylor, 2022).

This is not a bug. It is a feature. Automation that reduces the need for skilled labor also reduces the bargaining power of that labor. Companies know this.

Gender

The authors argue that automation is not gender neutral. The kinds of work that get automated first tend to be male dominated and relatively well paid, like manufacturing and logistics. The kinds of work that get automated last tend to be female dominated and poorly paid, like care work and cleaning. This is not because care work is harder to automate. It is because the people who make automation decisions are mostly men, and they tend to focus on problems that matter to men.

There is a deeper point here. The division of labor in society is gendered. Women are overrepresented in jobs that involve emotional labor, physical care, and social interaction. These jobs are systematically undervalued in economic terms. When automation decisions are made, they reflect these valuations. A robot that replaces a male factory worker is seen as progress. A robot that replaces a female nursing assistant is seen as cold and inhuman. The technology is the same. The social meaning is different (Howcroft & Taylor, 2022).

The State

This is the force that is most often ignored in popular discussions of automation. Governments shape automation through regulation, tax policy, research funding, and direct investment. The authors note that the state has historically played a massive role in developing the technologies that underpin automation, from the internet to GPS to artificial intelligence itself. Private companies did not build these things alone. They built them on top of public investments.

The state also shapes automation through labor law. A government that weakens unions and cuts unemployment benefits makes automation more attractive to employers, because the cost of displacing workers is lower. A government that invests in retraining and social safety nets makes automation less threatening, because workers have alternatives. These are political choices, not technological imperatives (Howcroft & Taylor, 2022).

The Study Behind the Argument

Howcroft and Taylor did not run an experiment. They did not survey workers or build a model. Their method is theoretical synthesis. They reviewed decades of research from sociology, labor studies, and science and technology studies, then applied that framework to the current automation debate.

This is a different kind of rigor. The authors are not claiming to have discovered a new fact about the world. They are claiming that the existing facts have been misinterpreted because the wrong framework has been used. By switching from technological determinism to social shaping, they argue, we can see patterns that were previously invisible.

The strength of this approach is that it accounts for complexity. The weakness is that it does not produce simple predictions. Howcroft and Taylor cannot tell you how many jobs will be automated in 2030. They can tell you that the number depends on choices that have not been made yet.

What the Research Does Not Prove

This paper is not a denial that automation is happening. It is not a claim that technology does not matter. It is not a prediction that everything will be fine if we just make better choices.

The authors are careful to note that technology has real effects. A robot that can pick and pack warehouse items faster than a human will change the economics of that warehouse, regardless of who designed it or why. The point is that the robot itself is a product of prior decisions, and its deployment is shaped by ongoing decisions.

The paper also does not prove that all automation is bad or that all automation is driven by bad motives. Some automation genuinely improves working conditions by removing dangerous or tedious tasks. The authors acknowledge this. Their argument is that the outcome depends on the social context, not on the technology alone.

The biggest open question is whether the social shaping framework can predict specific outcomes. Howcroft and Taylor are good at explaining why things happened the way they did. They are less clear on how to use that knowledge to shape the future. This is not a flaw in their analysis. It is a reflection of reality. Social systems are complicated, and prediction is hard.

Why This Changes Everything

If the social shaping view is correct, then the entire public debate about automation is built on a false premise. We have been asking the wrong question.

The question is not: what will technology do to us? The question is: what do we want technology to do for us, and how do we build the social and political structures to make that happen?

This shifts the conversation from passive waiting to active choice. It means that unions, regulators, and voters have power, even if they do not realize it. It means that the design of automation systems is a political act, not a technical one. It means that the future of work is not written in code. It is written in policy, in collective bargaining, and in public investment.

The authors are not naive. They know that power is unequally distributed. The people who design and deploy automation are usually rich and well connected. The people whose jobs are at risk are usually not. But the social shaping framework reveals that this asymmetry is not inevitable. It is the result of specific decisions that could be made differently.

What This Actually Means

• ▸The automation debate needs to shift from technology to power. Every time a company says it must automate to stay competitive, ask who makes that decision and who benefits. The answer is rarely the workers.
• ▸Unions and worker organizations need to engage with technology design, not just technology deployment. If the design phase is where social choices get embedded, then that is where labor needs a seat at the table. Waiting until the machines arrive is too late.
• ▸Government policy is the largest lever. Tax incentives for automation, research funding, labor law, and social safety nets all shape which technologies get built and how they get used. Voters and advocates should treat these policies as automation policy, because that is what they are.
• ▸Gender and race are not side issues. The automation of work reflects existing hierarchies. If the goal is a future where automation benefits everyone, then the people designing and funding automation need to look more like the people who will be affected by it.
• ▸The future is not written. This is the most important takeaway. The technology exists. The economics exist. But the social choices are still open. That is terrifying. It is also liberating.