Home

We Need Better Data on Workplace AI

Owen Davis / Oct 2, 2024

Yutong Liu & Kingston School of Art / Better Images of AI / Talking to AI / CC-BY 4.0

When surveyed at the end of last year about their AI fears, 30% of US workers expressed concern that AI could eliminate their position. Yet when asked about AI’s role in hiring and promotions, a much larger 71% of respondents voiced worries—more than double the share concerned about their own jobs.

These survey results underscore an important yet often overlooked point: AI is poised to change not just what work we do, but how it is done. As AI tools advance, workers can expect to be managed increasingly by algorithms. Even now, employers have access to AI-powered tools that recruit, compensate, monitor, direct, and evaluate workers. Although the ultimate effects of these tools are unknown, they pose real risks of eroding job quality and worsening income inequality.

Unfortunately, researchers and policy-makers currently lack the data necessary to understand how workplace AI affects workers and businesses. Existing AI data collection efforts tend to focus on the automation of tasks and the productivity of new technologies, missing a critical aspect: AI’s role as a manager.

The Rise of the AI Manager

Throughout history, new technologies have found their way into management systems. In Charlie Chaplin’s 1936 film Modern Times, celebrated for its dystopian caricature of the factory age, the first technological marvel shown is not some industrial machine, but a two-way television used by the boss to monitor workers.

Sure enough, CCTV systems soon took on the role of production-line supervisors. Punch cards gave way to RFID and biometric scanners. Factory managers hung up their stopwatches and let assembly lines set the pace.

Today, employers are adopting AI tools to take on a wide range of management functions. Algorithms target job ads to job seekers, screen resumes, calibrate wage offers, and suggest raises. Monitoring software tracks drivers’ eye movements, records desk workers’ app usage, and infers the moods of call-center workers. In high-tech warehouses, AI-powered systems guide workers’ movements and score them in real time.

Leading the shift towards AI-powered management are platform-work companies like Uber and TaskRabbit, where algorithms touch nearly every aspect of the job. Machine learning tools match customers to workers and set pay levels. Hiring and firing occur largely via app.

Yet gig work is hardly the only sector where AI tools play a role in staffing and management. In a 2024 survey of HR professionals, 26% of respondents reported using some form of AI, largely for recruitment and hiring. Among these, 10% said their firms employed AI-enabled monitoring systems. Another survey, this one targeting recruitment and hiring managers, found 60% of respondents reporting the use of AI-powered hiring tools like resume screeners or applicant scoring systems.

How AI Upends the Story of Automation

If management technology has been around so long, does AI really change anything? The short answer is yes. The standard view in economics is that automation historically took hold where routine tasks could be broken down into codifiable steps: weaving thread, stamping metal, adding numbers. Where work was non-routine or only tacitly understood, as in creative work and some manual labor, automation posed little threat.

AI upends these old patterns. AI systems can decipher handwriting, interpret X-rays, and compose original sonnets. They do this without a set of detailed instructions, relying instead on training data and machine learning algorithms. With the right data, AI systems can also be trained to detect when a worker may be loafing or predict the wage a job seeker might accept. This is something new under the sun.

While economists have developed robust theories about AI’s impact on employment, they have grappled less with AI in management and HR (especially compared to scholars from other disciplines). This gap arises partly because economists typically use a so-called task model, viewing jobs as bundles of tasks with varying degrees of susceptibility to automation. This focus on the content of work abstracts from the context of work, particularly how workers are managed.

The economic implications of AI in management and human resources could be substantial. In my research, I explore how AI might affect worker power, focusing on AI used to staff, monitor, direct and evaluate workers. In a range of theoretical models, I show how AI tools could reduce worker pay or exacerbate inequality. To use the academic jargon, AI may enable employers to capture a greater share of “economic rents” associated with a job. Translation: Workers end up with a smaller slice of the pie.

This is all to say that the question is not only “will AI take my job?” but also “will AI take my boss’s job?”

It may not all be bad news, of course. AI-powered job recommendations could steer workers to better opportunities. Hiring algorithms may be less biased than the humans they complement. Even an AI monitoring system may be preferable to a capricious office tyrant.

Measuring Workplace AI

The broad effects of AI tools in workplace settings are unknown. The theoretical concerns spelled out above have some basis in case studies and popular reporting, but they remain untested quantitatively. For that, we need data.

On the worker side, compelling new findings indicate that millions of workers feel the weight of digital management tools in the workplace. As Columbia University professor Alexander Hertel-Fernandez documents in survey-based research for the Washington Center for Equitable Growth, more than two-thirds of US workers report some form of digital monitoring on the job, while nearly 40% report that algorithmic assignment of tasks or schedules. More intense management technologies were associated with lower job satisfaction and well-being. (The tools reported on in the survey may or may not involve AI.)

In order to complement and extend these findings from the employee side, we also need data on the employer side—and, ideally, data that researchers could link to government-collected tax and administrative records. Outside of a few private-sector efforts, we have a limited understanding of AI’s prevalence in management and HR. Ideally, the US Census Bureau would include questions on this topic in its surveys of private businesses, particularly the Annual Business Survey (ABS), which collects data from 300,000 businesses every year.

In some years, the ABS has included questions about AI used by businesses in producing goods and services, excluding other AI applications like hiring and firing. And while fielding new census questions will be arduous, it’s necessary. Enhanced visibility into AI's role in the workplace will support better decision-making and policy development, from the factory floor up to the halls of Congress.

Other statistical agencies around the globe already have some experience in querying firms about AI used throughout the organization. The EU’s survey on new technologies has asked about AI use across a variety of business processes, including human resources management and recruiting. Israel’s statistical agency has asked similar questions. Their experiences could provide valuable insights for the US.

As AI continues to transform the economy, it is crucial for researchers and policymakers to have access to comprehensive data on AI’s scope, scale, and impact. Only then can we begin to unpack the performance of AI managers, and make decisions about what we want our workplaces to look like.

Authors

Owen Davis
Owen Davis is a labor economist and post-doctoral research fellow at Siegel Family Endowment, where he explores the workforce impacts of artificial intelligence and emerging technologies. His current research focuses on the impacts of AI on worker power, job design, and the nature of work. He receiv...

Topics