AI and the labour market: opening the black box

ABSTRACTA major focus of sociological research is on the role of the credential as a ‘currency of opportunity’, mediating the relationship between education and occupational destinations. However, the labour market has largely remained a ‘black box’ in sociological and education policy studies. This article draws on ‘big data’ from over 21,000,000 job adverts to explore how employers in the UK describe job requirements, with particular reference to the role of credentials. It challenges existing theories premised upon the notion that higher levels of formal education determine individual (dis)advantage in the competition for jobs. Although they have different views of the relationship between credentials, opportunity and efficiency, these theories assume that credentials largely determine occupational hiring. Our analysis suggests that formal academic credentials play a relatively minor differentiating role in the UK labour market, as the majority of employer’s place greater emphasis on ‘job readiness’. This raises a number of issues for sociological and policy analysis, including the future role of credentials in the (re)production of educational and labour market inequalities. Methodologically, the article highlights how the use of big data can contribute to the analysis of education, skills and the labour market.

This study provides an assessment of the features that, operating inside the black box, shapes the evolution of the Italian economy. In the first section the focus is placed on the functioning of the labour market at the aggregate level. The picture described suggests that the Italian labour market is characterized by high employment rates for adult males and both low participation and low employment rates for other groups of individuals (such as young, old and females) and by a large share of long-term unemployment. The rise in unemployment and its persistently high level have been blamed, both in Italy and in other European countries, on imperfections and rigidities originating mainly in the labour market. In this case, both strong employment protection regulation and rigid system of wage determination appear to have played a major role in Italy. In particular, the existence of high wage floor - either set through statutory minimum wages or by collective bargaining - has the effect of segmenting the labour force, increasing job instability and paradoxically increasing the duration of unemployment. An additional dimension of segmentation also characterizes the Italian labour market, notably the 30 percent's unemployment rate in the South (6 percent in the North) and the presence of a large share of the labour force employed in the underground economy. In the last section, we discuss how the decentralisation of bargaining, the introduction of higher labour market flexibility as well as the reform of social security, could reduce insider power in the labour market and eliminate the vicious cycle that tends to displace workers out of the primary sector, thus fueling the irregular economy.

Nowadays, the number of people and companies using the Web to search for and advertise job opportunities is growing apace, making data related to the Web labor market a rich source of information for understanding labor market dynamics and trends. In this paper, the emerging term labor market intelligence (LMI) refers to the definition of AI algorithms and frameworks that derive useful knowledge for labor market-related activities, by putting AI into the labor market. At the same time, another branch of AI is developing known as Explainable AI (XAI), whose goal is to obtain interpretable models from current (and future) AI algorithms, given that most of them actually act like black boxes, providing no interpretable explanations of their behavior, as in the case of machine learning. In this paper we connect these two approaches, using a graph model obtained through an NLP-based (Natural Language Processing) methodology for classifying job vacancies. We compare the results obtained with those from a European Project in LMI that employs machine learning for the classification task, to show that our approach is effective and promising.

The aims of this report are threefold. The first is to map how much information can be obtained on the labour markets in Central and Eastern Europe (CEE) in order to proceed with analyzing the labour market policies and the growth models in the region. Secondly, while some comparative data can be obtained through international databases, these are often incomplete or not fully up to date. We therefore select a sample of countries (Bulgaria, the Czech Republic, Estonia, Poland, Slovakia and Slovenia) to represent the broad geographic diversity of the region and for these we obtained data that is more detailed and more up to date compared to what is available from existing international databases. Thirdly, we organize some of this information into brief country profiles and a common profile for the region. This provides a handy, up-to-date summary of the state of collective bargaining and other aspects of labour relations and their legal context in these countries and allows us to posit some hypotheses for further research. Why are we interested specifically in the countries in Central and Eastern Europe? These countries seem to be in a “low innovation, low wage equilibrium” (Independent Evaluation Group. 2007). In addition, we might see in a new light the application of some key normative concepts featuring prominently in EU policy debates such as flexicurity and transitional labour markets (TLM). The task here is therefore to open up the black box of labour markets in the new member states by researching labour relations in the region. Overall, CEE countries defy an easy stereotype or model with regard to the labour relations and their wider context. While collective labour relations are quite liberalized, individual labour relations are not, but there is also relatively low presence of atypical forms of labour relations as known from EU-15. However, it can be hypothesized that the role of self-employed and older workers might be to provide flexibility for the labour market. There also seems potential for bifurcation of unemployed, but also parents into “insiders” and “outsiders” – social insurance vs long-term social assistance/paternity benefits. This is worth further investigation.

The youth unemployment in Macedonia has been high and persistent for the last three decades, despite the many active labour policies for young persons. This paper examines the determinants of youth unemployment in the Macedonian labour market using quantitative and qualitative analysis. In order to empirically estimate the relevant determinants of youth unemployment in Macedonia we employ the two-step conditional mixed processes model (CMP) using the school to work transition survey (SWTS) by the International Labour Organization. The quantitative analysis is accompanied by qualitative analysis to further explore and explain the social-economic issues. The qualitative analysis includes semi-structured interviews and focus groups conducted in different regions of Macedonia. The empirical results indicate that gender, age, the wealth of the household, the education level and regional characteristics are the main determinants of youth unemployment in Macedonia. According to the findings of the qualitative analysis, the most of the young people pursue education and live with their families who provide them with financial support to cover their expenses. Consequently, the young persons have a diminished interest to search for a job and this attitude on long run could have a dubious effect and influence on the personal choices. Overall, there is a long school to work transition period, which discourages young people to actively participate in the labor market.

Models of cooperative and noncooperative behaviour opened the household ‘black box’ and allowed for individual treatment of partners in couples. However, labour supply literature has so far largely ignored a broader issue – the distinction of single versus multi-family (‘complex’) households. We propose a method to account for multi-family household structure by borrowing from recent applications of the collective model to identify the degree of sharing. We assume that each household is characterized by a between-family sharing parameter, which is calibrated on estimated preferences, observed labour market status and other characteristics. We apply the method to Polish labour market data.

Previous articleNext article FreeSherwin Rosen PrizePDFPDF PLUSFull Text Add to favoritesDownload CitationTrack CitationsPermissionsReprints Share onFacebookTwitterLinked InRedditEmailQR Code SectionsMoreDavid Deming is the 2022 recipient of the Sherwin Rosen Prize for outstanding contributions to labor economics. Deming is the Isabelle and Scott Black Professor of Political Economy at the Harvard Kennedy School. Deming earned his PhD in public policy from Harvard University in 2010 and served as an assistant professor at Carnegie Mellon University the following year. He has taught at Harvard University since then. He has served as a coeditor of the American Economic Journal: Applied Economics and the Journal of Human Resources.David is a creative and outstanding scholar who has made first-order contributions to a wide range of labor and applied economics topics. His research illuminates important policy-relevant questions that push the research frontier, including articles on higher education, skill development, economic inequality, technological change, the long-run impacts of early and K–12 schooling, and the connections between education and criminal activity.Deming’s most profound impact may come from his work on the labor market returns to skills. In “The Growing Importance of Social Skills in the Labor Market” (Quarterly Journal of Economics, 2017), David illustrates how work is placing greater emphasis on “soft skills” like teamwork, social perceptiveness, problem-solving, communication, and adaptability. The labor market increasingly rewards social skills, while the share of less social, purely cognitive-intensive jobs is shrinking. He develops a team production model where workers “trade tasks” to exploit their comparative advantage. Those with more social skills are advantaged in trade, which enhances specialization and efficiency. He tests the model’s implications for sorting and the relative returns to skill across occupations. He documents the increased wage gain from sorting into social skill–intensive occupations in the 2000s for the 1997 National Longitudinal Survey of Youth (NLSY) cohort compared with the 1980s and 1990s for the 1979 cohort.With Ben Weidmann, he followed up this work with experimental evidence (Econometrica, 2021) that gets inside the black box of the social skills/team performance relation and identifies qualities of “team players.” His research characterizing the returns to experience as a race between on-the-job learning and skill obsolescence with Kadeem Noray (Quarterly Journal of Economics, 2020) similarly brings remarkable data analysis to new theoretical insights. It upends conventional wisdom about the lifetime returns to certain credentials and fields of study. His 2018 Journal of Labor Economics paper with Lisa Kahn presents evidence from a massive database of job postings that cognitive and social skills are complementary.David has made exceptional contributions to the economics of education. His 2009 American Economic Journal: Applied Economics paper uses a within-family siblings design to provide compelling longer-run evidence on the returns to early childhood educational investment. His 2014 American Economic Review paper (with Justine Hastings, Thomas Kane, and Douglas Staiger) documents the long-run implications of school choice and the importance of school quality. Deming has been a leader on policy-related issues in higher education, including the role and regulation of private for-profit postsecondary institutions, the rise of online programs and institutions, the design of student financial aid policies, and public college funding. His 2016 American Economic Review paper (with Noam Yuchtman, Amira Abulafi, Claudia Goldin, and Lawrence Katz) presents evidence regarding the value of degrees and certificates from online programs, for-profit colleges, and nonselective public institutions. Previous articleNext article DetailsFiguresReferencesCited by Journal of Labor Economics Volume 40, Number 3July 2022 Published for the Society of Labor Economists, Economics Research Center/ NORC Article DOIhttps://doi.org/10.1086/721003 Views: 590Total views on this site © 2022 The University of Chicago. All rights reserved.PDF download Crossref reports no articles citing this article.

This article examines the extent to which the increasing ethnic diversity of Northern Europe’s population translates into exposure between natives and migrants (and their descendants) in the workplace, using Danish register data over nearly three decades. Consistent with the expectations of neo-assimilation theory, we find that Danish workplaces have become a common site of mutual exposure, including exposure between natives and two subcategories of culturally distant migrants. At the same time, consistent with segmented assimilation theory, we find increasing segmentation, even in a highly regulated Danish labour market. The distribution of migrants and their descendants in workplaces increasingly deviates from a (simulated) random distribution and we find increasing discrepancies between opportunity pools and exposure levels, particularly at the lower end of the labour market. By opening the black box of workplace composition, this article demonstrates the co-occurrence of increased exposure and labour market segmentation, with its potential to respectively facilitate and hinder integration.

Digitalization brings a number of changes to labor, whereof two are of critical importance: Firstly, an increasing number of the tools used in production will be connected to the internet and/or local networks. That brings an opportunity for gathering data on how those tools are utilized. Secondly, improved and tailored cognitive computing systems have the potential to analyze such data far better than humans. This profoundly challenges how we understand the organization of labor. This is a literature study of, and a theoretical discussion on, the impact of digitalization on the labor market. It addresses concepts as well as concerns and explores if digitalization can be embraced by and integrated in existing labor markets, while avoiding deterioration of labor quality. This study recognized that digitalization is not likely to bring about mass replacement of workers by machines and bots. However, it is clear that transparency of and trust in cognitive computer systems are key factors in the successful digitalization of labor. The Nordic Social Partner Approach is identified as a framework with high potential to overcome these challenges.

This article investigates the hidden gender bias in in-work poverty within the European Union (EU). Using data from EUROSTAT, INE, EIGE, and various surveys, it identifies inconsistencies in poverty rates among employed women due to issues with construction and equivalence scales. The research highlights the "gender paradox," where women face significant disadvantages in the labour market, yet statistical data do not always reflect a gender bias in in-work poverty rates. To elucidate this phenomenon, the study advocates for expanding the analysis of indicators with alternative aggregation methodologies to better understand the household black box. It proposes an alternative methodology for assessing in-work poverty, considering individual incomes and family responsibilities, and suggests including in-kind benefits in poverty measurements. The findings underscore the importance of addressing gender biases in the labour market and their impact on in-work poverty.

The landslide victory of the center-left Moon Jae-In in the 2017 presidential election opened a window of opportunity for progressive reform in South Korea. Elected on a platform of social inclusion and fairness and with support from organized labor, the Moon Jae-In government (2017–22) made at first considerable progress in advancing inclusive labor market reform and an alternative growth model against opposition from business, but the administration quickly lost momentum when facing political headwinds. We show that the government's capacity for progressive reform and social concertation was constrained not only by business interests but also by divisions on the left. Opening up the black box of organized labor, we provide a nuanced analysis of tensions on the left and demonstrate how a counterintuitive coalition of labor market insiders and radical outsiders on the movement's left undermined social dialogue and more inclusive unionism.

This essay is a contribution to the forthcoming Oxford University Press Handbook of Corporate Law and Governance edited by Jeffery Gordon and Georg Ringe. In the 1960s and 1970s, corporate law and finance scholars recognized that neither discipline was doing a very good job of explaining how corporations were really structured and performed. For legal scholars, Yale Law School professor and then Stanford Law School dean Bayless Manning confessed that corporate law has “nothing left but our great empty corporation statutes – towering skyscrapers of rusted girders, internally welded together and containing nothing but wind.” Michael Jensen and William Meckling made a similar comment with respect to finance. The theory of the firm was an “empty box” or a “black box” that provided no theory about “how the conflicting objectives of the individual participants are brought into equilibrium.” The result of Jensen and Meckling’s seminal reframing of corporate law in agency cost terms, and so into something far broader than disputes over statutory language, was that both Manning’s empty skyscrapers and Jensen and Meckling’s empty box began to be filled.The essay proceeds by tracking how corporate law became corporate governance – from legal rules standing alone to legal rules interacting with non-legal processes and institutions – through three somewhat idiosyncratically chosen but nonetheless related examples of how we have come to usefully complicate the inquiry into the structures that bear on corporate decision-making and performance. Part I frames the first level of complication in moving from law to governance by defining governance broadly as the company’s operating system, a braided framework encompassing legal and non-legal elements. Part II then adds a second level of complication by treating corporate governance dynamically: corporate governance becomes a path dependent outcome of the tools available when a national governance system begins taking shape, and the process by which elements are added to the governance system going forward – driven by what Paul Milgrom and John Roberts call “supermodularity.” That characteristic reads importantly on both the difficulty of corporate governance, as opposed to corporate law, reform and the non-intuitive pattern of the results of reform: significant reform leads to things getting worse before they get better. Part II then further complicates corporate governance by expanding it beyond the boundaries of the corporation, treating particular governance regimes as complementary to other social structures – for example, the labor market, the capital market and the political structure – that together define different varieties of capitalism. Next, Part III considers commonplace, but I will suggest misguided, efforts to take a different tack from Parts I and II: to simplify rather than complicate corporate governance analysis by recourse to now familiar single factor analytic models: stakeholder theory, team production, director primacy, and shareholder primacy. Part III suggests that these reductions are neither models nor particularly helpful; they neither bridge the contextual specificity of most corporate governance analysis nor address the necessary interaction in allocating responsibilities among shareholders, teams and directors. As well, these “models” are static rather than dynamic, a serious failing in an era in which the second derivative of change is positive in many business environments and Schumpeter seems to be getting the better of Burke. Part IV concludes by examining the importance of a corporate governance system’s capacity to respond to changes in the business environment: the greater the rate of change, the more important is a governance system’s capacity to adapt and the less important its ability to support long-term firm-specific investment.

The high bar of proof to demonstrate either a disparate treatment or disparate impact cause of action under Title VII of the Civil Rights Act, coupled with the “black box” nature of many automated hiring systems, renders the detection and redress of bias in such algorithmic systems difficult. This Article, with contributions at the intersection of administrative law, employment & labor law, and law & technology, makes the central claim that the automation of hiring both facilitates and obfuscates employment discrimination. That phenomenon and the deployment of intellectual property law as a shield against the scrutiny of automated systems combine to form an insurmountable obstacle for disparate impact claimants. To ensure against the identified “bias in, bias out” phenomenon associated with automated decision-making, I argue that the employer’s affirmative duty of care as posited by other legal scholars creates “an auditing imperative” for algorithmic hiring systems. This auditing imperative mandates both internal and external audits of automated hiring systems, as well as record-keeping initiatives for job applications. Such audit requirements have precedent in other areas of law, as they are not dissimilar to the Occupational Safety and Health Administration (OSHA) audits in labor law or the Sarbanes-Oxley Act audit requirements in securities law. I also propose that employers that have subjected their automated hiring platforms to external audits could receive a certification mark, “the Fair Automated Hiring Mark,” which would serve to positively distinguish them in the labor market. Labor law mechanisms such as collective bargaining could be an effective approach to combating the bias in automated hiring by establishing criteria for the data deployed in automated employment decision-making and creating standards for the protection and portability of said data. The Article concludes by noting that automated hiring, which captures a vast array of applicant data, merits greater legal oversight given the potential for “algorithmic blackballing,” a phenomenon that could continue to thwart many applicants’ future job bids.

Replication data for: Preferences and Biases in Educational Choices and Labour Market Expectations: Shrinking the Black Box of Gender

Compensating Differentials in Teacher Labor Markets