Not Enough Jails
A review of labor economics, robotics data and the sociology of punishment examining what happens when technological displacement exceeds the capacity of labor markets to absorb the workers affected.
Artificial intelligence is commonly discussed as a software story. Robotics is commonly discussed as a manufacturing story. Employment is measured through labor statistics, while incarceration is usually examined within criminal justice. A large body of research suggests those subjects can also be studied as parts of the same institutional system.
The analytical issue is not whether technology creates more jobs than it eliminates in a national total. Aggregate job counts do not identify whether workers displaced from one occupation can realistically enter the employment that exists elsewhere in the economy. A labor market can add jobs overall while particular workers, occupations and communities experience displacement, wage loss, prolonged joblessness or movement into less stable work.
Each relationship in that sequence has to be examined on its own evidence.
Labor surplus and punishment: the established sociological question
Georg Rusche and Otto Kirchheimer's Punishment and Social Structure, published in 1939, placed punishment within the economic organization of society and the position of labor. Their work became the foundation for a long research tradition examining whether penal institutions respond only to crime or also to broader labor-market conditions.
In 1992, Theodore Chiricos and Miriam DeLone systematically reviewed 44 empirical studies of labor surplus and punishment. Their review concluded that, independent of the effects of crime, labor surplus was consistently and significantly related to prison population. They also found relationships with prison admissions in time-series and individual-level studies. The authors emphasized multiple economic, political and ideological mechanisms rather than a single automatic pathway.
Later research qualified the theory by showing that the relationship between labor-market conditions and imprisonment can vary across historical periods and institutional environments. The literature therefore provides neither a fixed unemployment-to-incarceration formula nor a reason to exclude labor-market conditions from the study of punishment.
Automation has produced measurable labor-market effects
The automation side of the relationship is also supported by observed evidence. Daron Acemoglu and Pascual Restrepo examined differences in industrial-robot exposure across U.S. commuting zones. Their Journal of Political Economy study found robust negative employment and wage effects in areas more exposed to robots.
Their estimate was that one additional robot per 1,000 workers reduced the employment-to-population ratio by about 0.2 percentage points and wages by about 0.42 percent. The study also found that more-exposed areas did not show differential pre-existing trends and that the robot effect was distinct from other capital and technologies.
That finding concerns a specific historical period and form of industrial automation. It nevertheless establishes that technological substitution has produced measurable employment and wage effects in U.S. local labor markets.
The physical automation base is already large
The International Federation of Robotics reported on September 24, 2026 that the global operational stock of industrial robots reached a record 5 million units in 2025, up 9 percent in one year and more than double the stock seven years earlier. Factories installed more than 600,000 industrial robots in 2025, an 11 percent annual increase. China alone installed 354,000.
Automation is also extending beyond conventional industrial robot arms. IFR's service-robot data for 2024 recorded more than 199,000 professional service robots sold in its supplier sample, a 9 percent increase, while its robot-as-a-service fleet grew 31 percent. Transportation and logistics accounted for 102,900 professional service units in that sample.
Consumer robotics is already a mass-market category in a different sense. IFR recorded about 20.1 million consumer service robots sold in its 2024 supplier sample, up 11 percent. Robots for domestic tasks—such as floor cleaning and lawn mowing—represented 97 percent of those consumer-unit sales. IFR cautions that its service-robot figures are based on a supplier sample and are not projected to the entire industry.
Humanoid robotics: installed base, production and capacity are different measures
Humanoid robotics requires particular care because a current installed-unit count can describe market penetration while missing changes in manufacturing capability.
In China, Shanghai government reporting stated that AgiBot had reached production of 15,000 humanoid robots by June 2026. The company also conducted continuous prototype validation on an electronics manufacturing line.
On September 12, 2026, a UBTECH industrial humanoid factory began production in Liuzhou. A Liuzhou government source describes the facility as designed for annual capacity above 10,000 units, with a production-line design cadence of one robot every 10 minutes. Those figures describe designed capacity and takt time; they are not evidence that the plant has already sustained that output continuously or delivered that number of robots to customers.
For the humanoid segment, at least four different measures therefore matter: installed productive robots, annual sales, actual production and manufacturing capacity. Conflating them can either overstate or understate the state of the industry.
Software automation reaches another part of the workforce
The International Labour Organization's 2025 Global Index used task-level analysis covering nearly 30,000 occupational tasks. It estimated that one in four workers globally is employed in an occupation with some degree of generative-AI exposure, rising to 34 percent in high-income countries. Clerical occupations remained the most exposed, while exposure expanded into some professional and technical occupations.
The ILO separates exposure from redundancy. Its analysis states that most exposed occupations contain tasks that continue to require human input and concludes that transformation is more prevalent than complete occupational replacement under its methodology.
Exposure, displacement, unemployment, reemployment and economic recovery are different labor-market outcomes and should not be treated as interchangeable measures.
Why net job growth does not measure worker absorption
The latest U.S. Bureau of Labor Statistics projections illustrate the measurement problem. BLS projects total employment to increase by about 5.9 million jobs between 2025 and 2035. In the same projection, office and administrative-support occupations decline by 752,100 positions, the largest numeric decline among major occupational groups.
| Occupation | 2025 employment | Projected net change, 2025–35 | 2025 median pay |
|---|---|---|---|
| Cashiers | 3.11 million | −200,600 | $32,880 |
| General office clerks | 2.60 million | −156,200 | $45,010 |
| Customer service representatives | 2.67 million | −141,800 | $44,770 |
| Secretaries and administrative assistants* | 1.89 million | −114,100 | $47,540 |
| Bookkeeping, accounting and auditing clerks | 1.53 million | −85,600 | $50,670 |
*Except legal, medical and executive. These are BLS projections of net occupational employment change, not a count of current workers expected to become unemployed.
The replacement-opening data make that distinction particularly clear. Although BLS projects cashier employment to decline by 200,600 positions over the decade, it also projects about 521,300 cashier openings per year, all from workers transferring occupations or leaving the labor force. Declining employment and continuing hiring can occur simultaneously.
The composition of projected growth is also heterogeneous. BLS projects large gains in home health and personal care aides, stockers and order fillers, fast-food and counter workers, registered nurses, managers and software developers. Those occupations differ sharply in wages, credentials, working conditions and skill requirements.
National job creation therefore does not tell us which displaced workers enter which new jobs.
Skill distance makes worker absorption measurable
Recent labor economics provides a direct way to study this problem. Claudia Macaluso's 2025 American Economic Journal: Macroeconomics study examines local skill remoteness: the dissimilarity between the skills used in a worker's previous job and those required by other jobs available in the local labor market.
Higher skill remoteness at layoff was associated with persistently lower earnings, a greater probability of changing occupations, a lower probability of returning to jobs with similar skill profiles and a greater propensity to migrate.
This evidence supports separating three forms of worker absorption:
This analysis uses worker-absorption gap as an analytical term for the difference between workers displaced from existing employment and the sustainable replacement opportunities those workers can realistically enter, considering transferable skills, credentials, geography, wages, training requirements and other transition barriers.
It is not an existing BLS statistic.
Economic marginalization and public safety
The next body of research concerns what follows sustained economic hardship. The evidence is differentiated by type of outcome.
A 2025 Annual Review of Criminology synthesis by Jens Ludwig and Kevin Schnepel concludes that policies reducing economic desperation have their clearest effects on property crime, while the best available evidence shows little systematic relationship with serious violent crime. That distinction prevents the economic-crime relationship from being generalized across offenses.
Institutional responses also matter. Robert Apel's 2026 review of research on the social safety net evaluates natural experiments involving public assistance and contributory social insurance, including unemployment insurance. The review concludes that social-safety-net policy has measurable public-safety effects at both individual and aggregate levels.
The evidence therefore places public policy inside the relationship between economic conditions and downstream social outcomes rather than treating policy as an external afterthought.
Labor markets and return to prison
For people already involved in the criminal-justice system, the labor-market evidence is particularly direct.
Kevin Schnepel analyzed administrative records involving approximately four million people released from prisons across 43 states between 2000 and 2013. He found that release into counties with higher low-skilled wages significantly reduced the risk of recidivism. The study reports larger effects for several subgroups and in industries more willing to hire people with criminal records.
The relevant labor-market variable was not simply whether the national economy was expanding. It was the opportunity actually available to people leaving prison.
A modern measurement of labor surplus
The historical concept of labor surplus cannot be reduced to the headline unemployment rate.
A worker may remain employed while experiencing a large wage loss. Another may move from stable full-time work into intermittent employment. Another may leave the labor force. Another may regain employment only after abandoning years of occupational skill or relocating geographically.
A contemporary examination of economically marginal labor therefore requires multiple measures: displacement, duration of unemployment, labor-force participation, underemployment, wage changes, occupational switching, skill remoteness, geographic mobility, work stability and access to economically comparable replacement employment.
A large enough displacement-to-absorption mismatch can create economically marginal labor even during a period of positive national job growth.
That proposition does not depend on whether total employment rises or falls. It concerns the distribution of displacement and the capacity of particular workers and communities to enter the employment that remains available.
The institutional question
When labor-market marginalization occurs, its effects can appear across multiple institutions: unemployment insurance, housing systems, public assistance, healthcare, homelessness services, workforce programs, policing, courts, probation and parole, jails and prisons.
Which institutions experience pressure, the magnitude of that pressure and the conditions under which it occurs are empirical questions.
That is where the historical punishment literature becomes relevant to contemporary technological change. Rusche and Kirchheimer asked how systems of punishment were connected to the economic position of labor. Later researchers tested parts of that proposition and found relationships that persisted in many studies after crime was taken into account, while also documenting historical and institutional variation.
The technological environment is now different. Industrial robots operate at global scale. Service robots are deployed across logistics, professional and domestic uses. Generative AI reaches tasks in clerical, technical and professional occupations. Humanoid manufacturers are beginning to build production capacity measured in thousands of units.
The underlying research question remains measurable:
What happens when people remain available to work but the labor market has less use for the work they know how to perform—and how effectively are those workers absorbed into sustainable replacement employment?
Answering that question requires following workers rather than simply counting jobs.
Sources and research
- Chiricos & DeLone, “Labor Surplus and Punishment,” Social Problems (1992) (opens in new tab)
- Acemoglu & Restrepo, “Robots and Jobs,” Journal of Political Economy (2020) (opens in new tab)
- International Federation of Robotics, World Robotics 2026 industrial-robot release (opens in new tab)
- International Federation of Robotics, World Robotics 2025: Service Robots (opens in new tab)
- Liuzhou government reporting on UBTECH humanoid production facility (opens in new tab)
- Shanghai government / China Daily reporting on AgiBot production (opens in new tab)
- International Labour Organization, Generative AI and Jobs: A Refined Global Index of Occupational Exposure (2025) (opens in new tab)
- U.S. Bureau of Labor Statistics, Employment Projections 2025–2035 (opens in new tab)
- U.S. Bureau of Labor Statistics, occupations with the largest projected job declines (opens in new tab)
- U.S. Bureau of Labor Statistics, Cashiers Occupational Outlook Handbook (opens in new tab)
- Macaluso, “Skill Remoteness and Post-layoff Labor Market Outcomes,” AEJ: Macroeconomics (2025) (opens in new tab)
- Ludwig & Schnepel, “Does Nothing Stop a Bullet Like a Job?” Annual Review of Criminology (2025) (opens in new tab)
- Apel, “Can Social Safety Net Spending Prevent Crime?” Annual Review of Criminology (2026) (opens in new tab)
- Schnepel, “Local labor markets and criminal recidivism,” Journal of Public Economics (2017) (opens in new tab)
Editorial note: OACRA News & Policy Review is published by OACRA LLC, an independent private organization. This analysis synthesizes peer-reviewed research, government labor data and robotics market reporting. Observed findings, market statistics and published projections are identified separately. The term “worker-absorption gap” is used here as an analytical construct, not as an official government statistic.
