7. Social dialogue and collective bargaining in the age of artificial intelligence

Sandrine Cazes

Abstract

Rapid advances in the development and adoption of artificial intelligence (AI) provide new opportunities but also raise fears about disruptive labour market and workplace transitions. This chapter examines the interlinked relationship between social dialogue and AI adoption in the labour market and workplaces. It highlights how social partners can facilitate the AI transition, for both workers and employers, while presenting new descriptive evidence and recent social partners’ initiatives related to AI diffusion. The chapter also discusses how AI adoption may affect social dialogue itself: while AI technologies could be used as an enabler for social partners’ goals and strategies, they may also pose new challenges to social dialogue, such as insufficient AI-related expertise and resources to respond to the AI transition. Based on these insights, the chapter presents key policy recommendations.

In Brief

Key findings

Artificial intelligence (AI) is likely to bring both benefits and risks for workers and employers. Its overall impact in the workplace and the labour market depends on how it is implemented, how it is regulated and the extent to which all stakeholders are involved. In this context, collective bargaining and social dialogue can play a key role in supporting workers and businesses in the AI transition, and in fostering fair and dynamic labour markets. This chapter explores the relationship between social dialogue, collective bargaining and the adoption of AI in the labour market and workplaces. The main findings are:

Social partners can facilitate the use of AI by helping to decide which AI technologies are adopted, facilitating their introduction, and defining training needs. They can also help firms in providing flexible and pragmatic – yet fair − responses to technological changes in the workplace and contribute to enhance the quality of the working environment. Social partners can also shape the design and definition of new rights, such as the right to not be subject to automated decision-making without human oversight, and improve existing ones, such as the right to training. Finally, collective bargaining can also complement public policies in enhancing workers’ security and adaptability.
However, the number of workers who are members of unions and are covered by collective agreements has declined in many OECD countries, and the development of new forms of work and new business models, enabled by AI transition, risks exacerbating the under-representation challenge faced by traditional social partners.
The specific characteristics of AI – such as the speed at which it is spreading, its ability to learn and the greater power imbalances it can create – for instance through information asymmetries or due to complexity of AI − put further pressure on labour relations. New waves of AI technologies, such as generative AI, are likely to significantly exacerbate this risk. At the same time, social partners may be able to make use of AI tools in the pursuit of their own goals and strategies.
Surveys of social partners’ views on AI adoption suggest that their main concerns include a trustworthy use of AI, changing skill demands and the quality of the working environment. While unions identify ethical issues as the biggest threat, employers’ organisations are most concerned about new skills requirements.
Despite these challenges, social dialogue and collective bargaining can play a useful role in promoting fairer AI transitions for workers and a level playing field for firms. New descriptive evidence based on cross-sectional European data shows for instance that the adoption of technologies with AI components in companies which have some forms of workers’ representation, is significantly associated with better working conditions than adoption in employers without such workers’ voice structures. These employers are also significantly more likely to provide social support to workers. In addition, a new cross-sectional survey carried-out in the manufacturing and financial sectors in Austria, Canada, France, Germany, Ireland, the United Kingdom and the United States points to the beneficial role of consulting workers or workers’ representatives for performances and working conditions.
Both unions and employers’ organisations have already undertaken several initiatives on AI at international and national levels. For example:
- They are engaging in outreach and awareness campaigns highlighting the need for new skills and the associated training requirements, as well as warning about a number of issues of concern, such as the trustworthy use of AI, workers’ data privacy and data protection.
- They have also started to provide guidance through framework agreements and by engaging in “algorithm negotiations” (bargaining on the use of AI, big data and electronic performance monitoring), although only a few AI-related collective agreements have so far been signed in OECD countries.
However, most ongoing activities stem from a few very active unions and employers’ organisations. In this respect, the lack of AI-related expertise among social partners is one of the major challenges to support their members in the AI transition. Moreover, the rapid evolution of the technology, including the recent deployment of the latest wave of generative AI, risks making this skill shortage more salient and problematic.

While each country’s situation and labour relations differ, policy makers could consider taking steps to encourage consultations and discussions on the AI transition with social partners and other stakeholders. They could also support social partners’ efforts to expand their membership to forms of work and employers where they are not currently represented, such as those involved in the platform economy. And they could introduce measures to promote the development of AI-related expertise, and digital education more generally, in the workplace for management, workers and their representatives.

Introduction

Artificial intelligence (AI) technologies are likely to have an important impact on labour markets, workers and workplaces. As with any technological changes, AI adoption is likely to bring both potential benefits and risks. AI technologies bring real opportunities to create new tasks and jobs, as well as new business models, for instance. Since AI has the potential to complement and augment human capabilities, it can lead to higher productivity and greater demand for human labour. Yet AI technologies also bring a series of significant risks that need to be urgently addressed (Chapter 2).

Against this background, the way AI adoption and diffusion is regulated (e.g. through national and international legislation and collective bargaining), and the extent to which stakeholders are involved in the design of regulations and implementation in the workplace, are key elements to explore. Previous OECD work has highlighted the instrumental role that social dialogue and collective bargaining can play in technological and organisational changes, by easing transitions and spreading best practices for the introduction of new business methods, training and safeguarding quality, as well as by complementing public policy. It has also shown how collective bargaining, provided it has high coverage and leaves some margins of flexibility, can foster inclusive and dynamic labour markets when systems are co-ordinated (OECD, 2018[1]; 2019[2]).

This chapter concentrates on the relationship between AI adoption and social dialogue. It examines how social dialogue and collective bargaining can shape the AI transition in beneficial ways for both workers and employers, while also looking at how AI is affecting social dialogue itself. On the one hand, AI technologies and their adoption might generate power imbalances between workers and their employers − due for instance to insufficient AI-related expertise or asymmetric information in the context of datafication of work − and challenge further the representativeness of traditional actors of social dialogue (as measured by quantitative indicators such as the number of members or the share of workers covered by collective agreements). On the other hand, AI technologies might offer new tools that could enable social partners to increase representation and improve the way they manage their relationships with members.

This chapter starts by discussing to what extent the AI transition may differ from previous technological changes in terms of its impact on social dialogue, drawing notably on OECD questionnaires circulated to social partners through the Trade Union Advisory Committee (TUAC) and Business@OECD (BIAC) networks 1 (Section 7.1). Based on a combination of literature review, OECD AI surveys of employers and workers in the manufacturing and finance sectors, and new descriptive analysis using European cross-sectional data from establishments surveys (ESENER-3 data) on the role of workers’ representative voice in the workplace, Section 7.2 then presents new empirical insights on the relationship between workers’ voice and risks associated with AI adoption. Section 7.3 describes some concrete examples of social partners’ recent initiatives in providing information, raising awareness or signing collective agreements related to AI adoption. Section 7.4 concludes with some policy recommendations.

7.1. AI transition: What is different for social dialogue?

AI adoption may raise idiosyncratic issues such as the development of new AI management models that could change the nature of the relationship between firms and workers, as well as raise more fundamental ethical concerns that call for specific attention from social partners. Another key distinction between AI and previous technologies is that AI can automate non-routine tasks which extends considerably the potential scope of automation (Chapter 3). Along these lines, many experts hint that AI’s impact on the labour markets are likely to be magnified by the speed and large potential for application across multiple sectors and occupations (Brynjolfsson, Rock and Syverson, 2017[3]). The complexity and opacity of AI technologies and the generation of information asymmetries resulting from AI-based surveillance of workers may also trigger greater power imbalance (De Stefano, 2019[4]). At the same time, AI technologies may bring opportunities to social partners, for instance in helping to strengthen workers’ organisation or voice.

7.1.1. Social partners’ views on risks and benefits of AI

To better understand social partners’ priorities concerning the AI transition, an OECD questionnaire was addressed to trade unions and employers’ organisations through the TUAC and Business@OECD networks across OECD countries (see Box 7.1). This questionnaire complements previous social partners’ surveys on digital technologies more generally − see for example Voss and Riede (2018[5]) and country-specific social partners surveys on AI, such as those by ver.di (2019[6]) and by INPUT Consulting in co-operation with the humAIn work lab (2021[7]) in Germany.

Overall, answers to the OECD questionnaire suggest that social partners’ main concerns relate to a trustworthy use of AI,2 changing skill demands and physical and mental health risks in the workplace. While unions identify ethical issues as their biggest threat, employers’ organisations are most concerned about new skills requirements (Figure 7.1). This does not however prevent heterogeneity within unions’ (or employers’) responses, a trustworthy use of AI, for instance, not being systematically identified by all unions as their main concerns. Compared to previous surveys of social partners’ views on digitalisation or AI adoption,3 social partners’ concerns appear to shift from job displacement risks to more societal ones linked to potential discrimination, excessive surveillance and violations of human rights. This apparent focus shift is also echoed by a survey of German works councils through the network of ver.di, IG Metall and DGB, which ranked changing work content and skill demand as their biggest concern of AI adoption – before job destruction (INPUT Consulting/humAIn work lab, 2021[7]). As for the potential benefits of AI adoption, responses to the OECD questionnaire slightly differ between unions and employers: while the former identify improved job quality and the creation of new tasks 4 as potential biggest opportunities of AI adoption, employers primarily see its potential for productivity gains and higher job safety.

This priority shift is in line with recent waves of evidence on both automation (Georgieff and Milanez, 2021[8]; Dauth et al., 2021[9]) and AI specifically (Georgieff and Hyee, 2021[10]), suggesting that AI adoption so far has not led to job destruction and employment downsizing. The evidence presented in Chapter 3 shows that the clearest effect of AI is the creation of new tasks and jobs, while evidence on the productivity and displacement effects is more mixed. Yet, the fast-moving development of AI latest technologies (such as Chat-GPT) may bring new risks and possibly challenge these results (Chapters 2 and 3).

Box 7.1. The OECD questionnaire on artificial intelligence and social dialogue

While the AI-related literature has focused primarily on the expected risks and benefits of AI adoption in the workplace, less attention has been devoted to social partners’ views and priorities. Filling this gap, the OECD questionnaire sent to social partners focused on: i) understanding social partners’ awareness of what AI entails; ii) reviewing social partners’ assessments of the risks and benefits relating to AI adoption at labour market and workplace levels; iii) collect information on how social partners are responding to support workers and employers in the AI transition; iv) reviewing social partners’ assessment of how AI adoption may affect social dialogue itself, including new challenges for integrating workers or businesses as members, but also new opportunities and tools to facilitate their work (see details in Annex Table 7.A.1).

The questionnaire was circulated to social partners in 2022 through the TUAC and Business@OECD networks across OECD countries. It reached confederations at the national level representing a variety of local unions or employers’ organisations. While providing qualitative and non-representative answers, the results collected by the questionnaire were subsequently discussed in two OECD expert meetings including social partners, researchers, employers and AI developers and complemented with existing social partners’ views and academic literature.

While responses collected through the OECD questionnaire provide interesting insights that complement and update previous social partners’ surveys (Voss and Riede (2018[5]), ver.di (2019[6]) and INPUT Consulting in co-operation with the humAIn work lab (2021[7])), they remain qualitative and are not representative. Notably, responses may be biased insofar as stemming mostly from social partners already active in the area of AI – which may in turn affect the responses.

Figure 7.1. Social partners focus on AI’s impact at workplace levels

7.1.2. What new challenges does AI pose to social dialogue?

In the past decades, social dialogue and collective bargaining have been under increasing pressures. Across OECD countries, trade union density has declined in general from 33% on average in 1985 to 16% in 2019 and the share of employees covered by a collective agreement shrank from 46% in 1985 to 32% on average in 2019. The development of flexible forms of work including platform and gig work, have exacerbated this decline, as workers with such flexible forms of work are 50% less likely to be unionised than standard workers (OECD, 2018[1]; 2019[2]). This underrepresentation by unions is particularly relevant in the context of AI adoption as firms in the platform economy tend to be early AI adopters – (Adams-Prassl, 2019[11]; Liu et al., 2021[12]; Malik, Budhwar and Srikanth, 2020[13])

On the employers’ side, the share of workers employed in a firm that is represented by an employers’ organisation has stayed relatively stable at around 59% across OECD countries – but small firms and those with new business models enabled by organisational and technological changes are also much less likely to be represented (OECD, 2019[2]). This suggests that employers’ organisations also need to improve their representativeness by reaching out to underrepresented or new actors.

Beyond representativeness challenges, AI technologies may also affect social partners’ capacity to support their members through dialogue and bargaining, even if the risk of weakening social dialogue through AI adoption was not identified among the main concerns by social partners (Figure 7.1). AI technologies are expected to diffuse rapidly and for some of them to continuously develop through their potential to self-improve, which will require continuing adjustments from workers and employers (Lane and Saint-Martin, 2021[14]). For social dialogue, this will likely require a shift away from monitoring and agreeing to constant rules towards more regular consultations between social partners and other operating parties as well as new forms of centralised and de-centralised conflict-resolution mechanisms (Albrecht and Kellermann, 2020[15]). While social partners may therefore need to adapt the frequency and way of co-ordinating with each other, collective bargaining may remain the most effective instrument to address AI-related issues as it has the capacity of shaping new rights and implementing existing ones in a flexible and pragmatic – but yet fair manner (OECD, 2019[2]; Aloisi, 2021[16]).

At the same time, AI technologies may also complicate social partners’ capacity to co-ordinate and bargain. For example, the British Trade Union Congress (TUC) fears that the use of AI changes the employment relationship in a way that blurs accountabilities of decisions (TUC, 2021[17]), which may ultimately affect social partners’ capacity to represent workers’ and employers’ interests. Integrating AI into co-determination structures can for example be a challenge, when employers cannot provide the necessary information about AI-influenced decisions to workers or their representatives, because they are themselves detached from AI developers who may not disclose such information (Albrecht and Kellermann, 2020[15]). Accountabilities may also be unclear if knowledge gaps exist about AI between developers, vendors, and contracting authorities, as well as between those negotiating the procurement (Colclough, 2022[18]).

In this respect and beyond blurring accountabilities, AI may also affect social dialogue by changing the power balance between workers, employers and their representatives, for instance when AI-based surveillance of workers generates information asymmetries (Rani and Singh, 2019[19]; De Stefano, 2018[20]). Such asymmetries are likely to reduce workers’ bargaining position (Adler-Bell and Miller, 2018[21]), especially when workers are not aware that they are interacting with AI, or not sufficiently informed about the outcomes of this interaction − for example when AI is introduced through updates of technologies already in place and thus not considered as new technology on which workers’ representations should be consulted (EESC/CFDT Cadres, 2022[22]). Besides, even in the case where AI is considered as new technology, a prior agreement with workers’ representatives before monitoring workers through new technologies is, currently, not necessarily required in all OECD countries (Aloisi, 2021[16]; Salvi del Pero, Wyckoff and Vourc’h, 2022[23]) – see also Chapter 6. Finally, power imbalances in the employment relations may question the notion of workers’ consent to interact with AI or allow the use of their personal data (whether in recruitment, management or other processes), since they can make it difficult for workers to actually deny consent, even in countries where employers are supposed to obtain their consent. (Data Protection Working Party, 2017[24]; Moore, 2020[25]).

Finally, fears exist that the use of AI may limit or prevent social dialogue to some extent. AI-based monitoring of workers can potentially be used to monitor union activity and prevent collective organising, as observed for ride-hailing or delivery platforms (De Stefano, 2016[26]; EESC/CFDT Cadres, 2022[22]). In this respect, AI might be used to analyse information such as the location of union offices, the activity of union officials, the use of union-related vocabulary in emails, and even union activity on social media (TUC, 2021[17]). This risk appears to be higher in non-standard forms of work and in countries where laws do not anchor or support institutionalised forms of social dialogue and collective bargaining, particularly beyond the firm level. The new wave of generative AI is also likely to exacerbate this type of risk.

7.2. Shaping the AI transition: The role of social dialogue

7.2.1. Getting some insights from the empirical literature on social dialogue and automation

As no econometric literature exists on social dialogue and AI adoption (investigating either the role of workers’ representation in enhancing AI adoption or in mitigating its effect), some insights can be obtained from studies on social dialogue and automation even tough, as shown in Chapter 3, theoretical and empirical effects from automation and AI adoption are not the same. In that context, this section presents a short literature review on automation in the form of robots and social dialogue.

Regarding the relation between workers’ representation and robot adoption, the literature finds mixed and only descriptive results that might suffer from reverse causality, i.e. robot adoption may instead impact workers’ representation. Keeping these caveats in mind, Onorato (2018[27]) finds that, at national level, union density is negatively associated with robot adoption in OECD countries, using a constructed panel dataset based on data from the International Federation of Robotics and OECD statistics. Similarly but at the firm level, Genz, Bellmann and Matthes (2018[28]) find that, in Germany, the existence of works councils is associated with a statistically significant lower adoption of automation − and digital technologies in general. However, the authors find evidence suggesting that works councils foster adoption of these technologies in establishments that employ a high share of workers who are conducting physically demanding tasks. In contrast, Belloc, Burdin and Landini (2022[29]) find a positive association between workers’ representation and the adoption of robots and data analytics in management practices in Europe, using cross-sectional data from the European Company Survey 2019. The authors investigate various potential mechanisms driving these associations and find suggestive evidence that workers’ representation influences workplace practices, notably in terms of training intensity and process innovation, in ways that may enhance the complementarity between labour and new technologies.

As for the second issue, namely the link between workers’ representation and the effects of automation, the empirical evidence suggests a positive effect on wages and employment. Parolin (2019[30]) finds for instance that shrinking collective bargaining coverage at the national level is associated with declining relative wage growth for occupations at higher risk of automation. This strand of the literature is further motivated by the paper of Dauth et al. (2021[9]), which finds that early robot adoption in the German manufacturing sector was not associated with increased unemployment but instead with increased reskilling of workers – contrary to findings from the United States (Acemoglu and Restrepo, 2018[31]). The authors’ conjecture that this finding could be due to stronger labour market institutions in Germany like collective bargaining, but do not provide direct evidence on this. Following this pursuit and using a random effects regression analysis with constructed panel data from the European Labour Force Survey and the U.S. Current Population Survey, Haapanala, Marx and Parolin (2022[32]) find that union density moderates employment increase in automation-exposed industries for younger workers but enhances employment increase for high-educated workers.

7.2.2. New OECD evidence on workers’ voice, AI adoption and its effect

The absence of empirical evidence on the link between social dialogue, AI adoption and its effects is largely due to data limitations. Most existing individual- and firm-level panel data do not simultaneously include indicators on these three aspects and require matching information from different sources or limiting the analysis to cross-sectional or constructed panel data. As reviewed above, the literature focuses on automation, rather than AI adoption, and workers’ representation mitigating effects on employment and wages.

Against this background, this section attempts to bring some insights on AI’s impact and the role of workers’ voice in the workplace, considering different workers’ voice arrangements.5 It examines first how representative workers’ voice might mitigate AI’s impact on several risks relating to working conditions in Europe (for a detailed overview, see Box 7.2). This evidence is based on the 3^rdEuropean Survey of Enterprises on New and Emerging Risks (ESENER-3) data that allows for distinguishing proxies for different types of AI components used in the establishment as part of technologies adopted, as well as different types of representative workers’ voices.6 As for working conditions, only data on non-monetary aspects, such as hard physical work, work intensity or social support (i.e. help and support from colleagues) at work are available in the dataset.

Box 7.2. How does representative workers’ voice mitigate the impact of AI on risks relating to working conditions?

The empirical strategy follows the model in Haapanala, Marx and Parolin (2022[32]) by estimating the moderating (e.g. interaction) effect of representative workers’ voice on AI’s impact on labour market outcomes. However, instead of looking at employment and wage effects like Haapanala, Marx and Parolin, the analysis is conducted at the workplace level and considers AI’s impact on several aspects of the quality of the working environment, such as heavy loads or long working hours. A probit regression is adopted instead of fixed effects regression due to the cross-sectional nature of the data.

The Third European Survey of Enterprises on New and Emerging Risks (ESENER-3) (EU-OSHA, 2019[33]) asked 45 420 establishments from 33 countries and different sectors how health and safety risks were managed at the workplace level, with a particular focus on digitalisation and psychosocial risks. The survey covers all EU Member States, Iceland, North Macedonia, Norway, Serbia, Switzerland and the United Kingdom. The ESENER-3 survey includes detailed information on various forms of representative workers’ voice that co-exist in the workplace, different types of technologies used and some aspects of non-monetary working conditions. AI adoption is defined as firms adopting at least one of the following technologies: i) robots that interact with workers; ii) wearable devices, such as smart watches, data glasses or other (embedded) sensor; iii) machines, systems or computer determining the content or pace of work; iv) machines, systems or computers monitoring workers’ performances.

The probit regression model is specified as follows:

$Y_{i j c} = β_{0} + β_{1} {A I}_{i j c} + β_{2} {W R}_{i j c} + β_{3} {W R}_{i j c} \times {A I}_{i j c} + β_{4} X_{i j c} + ε_{i j c}$

where subscript i denotes the establishment; $Y_{i}$ is the dependent variable (e.g. different working conditions outcomes as dummy variables); ${A I}_{i}$ is a dummy variable equal to 1 if the establishment i in industry j and located in country c uses either robots, smart devices or software to monitor workers or determine the content and pace of their work; ${W R}_{i}$ is a dummy variable for the presence of worker representation at the establishment level (tested separately for three types of representation, e.g. the presence of a trade union representative, a works council or a health and safety committee/representative elected by workers); $X_{i}$ is the vector of controls at the establishment level (e.g. country, industry, size, age, economic situation, share of old workers, if the person interviewed is the owner/manager, and if teleworking is possible) similarly to Belloc, Burdin and Landini (2022[29]), who use a similar European-level dataset; and $ε_{i}$ are the residuals.

The variable of interest here is the coefficient $β_{3}$ that captures the AI and worker representation interaction. If $β_{3}$ is negative, the effect of AI on the selected outcome will be mitigated in establishments with worker representation compared to establishments with no worker representation. The (unmitigated effect) of AI is denoted by $β_{1}$ and the actual mitigated effect of AI is obtained by adding $β_{1} + β_{3}$ (see Annex 7.B).

Source: EU-OSHA (2019), The Third European Survey of Enterprises on New and Emerging Risks, https://visualisation.osha.europa.eu/esener/en/survey/overview/2019; Haapanala, H., I. Marx and Z. Parolin (2022), “Robots and Unions: The Moderating Effect of Organised Labour on Technological Unemployment”, IZA Discussion Paper Series No. 15 080, https://www.iza.org/publications/dp/15080.

Results from probit regressions suggest that workers’ representation may mitigate the impact of technology with AI component on some risks relating to working conditions. Figure 7.2 reports marginal effects (interaction effect of AI adoption and representative voice):7 it shows for instance that in establishments using AI and having a works council, or a health and safety representative/committee, AI adoption is associated with a significantly larger reduction in worker exposure to heavy loads (by 3 percentage points and 4 percentage points respectively for works council and health and safety representation) than establishments using AI but without representative workers voice.8 Moreover, in AI-using establishments that have a trade union representative or health and safety representative/committee, AI adoption is associated with a significantly larger reduction in exposure to high noise than establishments using AI but do not have representative workers’ voice. Finally, the presence of a works council appears to reduce risks to be exposed to long working hours, while the presence of health and safety representative/committee tend to increase social support at work.

Figure 7.2. Representative workers’ voice is associated with mitigating AI’s impacts on risks relating to working conditions, but causality remains unclear

These results are robust to different sets of controls and checks. They confirm both the supporting and mitigating effects of representative workers’ voice when testing for the different types of AI components adopted by the establishments, although impacts may differ between AI-related software (e.g. workers more likely to be provided help and social support, less likely to be exposed to painful positions, heavy loads, repetitive arm movements, high noise, fumes or vapours or chemical products and long working hours) and AI-related hardware (e.g. workers less likely to be exposed to painful positions, repetitive arm movements and work at very high speed); moreover, in the case of AI-related software used for monitoring performance, the estimated mitigation effect for repetitive arm movements is positive, suggesting the possibility of reverse causality.

In terms of potential mechanisms driving the mitigating effect of representative workers’ voice on several risks related to working conditions, a recent paper suggests that representation indirectly affects the type of AI systems employers invest in by shaping job designs (Belloc et al., 2022[34]). Specifically, the authors find that in establishments with representative workers’ voice, job designs are richer, i.e. more complex and with tasks less routinised – and thus more difficult to monitor, potentially helping to orient AI-related investments towards those AI systems that improve working conditions.

Although the analysis controls for an extensive set of variables, it remains descriptive and mostly serves as a motivation to investigate further any causal relation between workers’ voice and the AI transition. Moreover, as illustrated by some estimates of the mitigation effect, results might also suffer from reverse causality. Finally, one cannot exclude the fact that the effects of AI use are properly identified and not driven by the general degree of technological advancement of the establishment. Unfortunately, data do not permit to control for this effect. Future research should further investigate the role of different measures of social dialogue and collective bargaining indicators, as well as workers’ voice arrangements.

On this latter aspect, the OECD carried out a cross-sectional survey with 5 334 workers and 2053 firms in the manufacturing and financial sectors in Austria, Canada, France, Germany, Ireland, the United Kingdom and the United States (Lane, Williams and Broecke, 2023[35]) considering different types of outcomes and of workers voice (representative and direct voices). As shown in Panel A of Figure 7.3, the survey revealed significant heterogeneity across OECD countries in the incidence of consultations as regards AI adoption, being typically twice as large in Germany or the United Kingdom than in the United States. Furthermore, averaging countries and sectors, results indicate that workers using AI are more likely to report that AI improved their performance and working conditions if their companies consulted workers or worker representatives regarding the adoption of new technologies in the workplace (Panel B of Figure 7.3). For example, workers in companies that consulted workers or worker representatives are 9 percentage points more likely to say that AI had improved their health and safety, compared to workers in companies that did not consult workers or workers representatives.

This is consistent with previous OECD research that found that direct voice between workers and managers (either alone or mixed forms, i.e. combined with representative workers’ voice) was associated with a higher quality working environment (OECD, 2019[2]). However, since the analysis cannot establish causality, it is not possible to say definitively that consultation encourages employers to deploy AI in a more productive, fulfilling and safe manner. It could also be that the act of being consulted generates positive perceptions of the AI, even if little has changed.

Figure 7.3. Employers that consult workers or worker representatives are more likely to report positive impacts of AI on worker productivity and working conditions

7.2.3. Fostering inclusive labour markets and easing technological transitions

Previous OECD research (OECD, 2019[2]) has highlighted the granularities of collective bargaining systems and workers’ voice arrangements, and the importance of understanding their actual organisation and functioning to properly assess how social dialogue may shape labour market outcomes and job quality.9 The main findings are reported in Figure 7.4.They show that collective bargaining, provided it has high coverage while leaving some margins of flexibility, can foster inclusive and dynamic labour markets when bargaining systems are co-ordinated 10 and the quality of labour relations between the social partners is high. Social partners can also contribute to determine what technologies, including AI, are adopted, facilitate their introduction, and anticipate skills needs: through their representation in skills council and training provisions in collective agreements, as well as their involvement in the process of developing, funding and managing adult educational and training programmes, social partners has also been found to be beneficial both in terms of the quality of training and accessibility for all workers. Finally, through workers’ voice, they can ease AI introduction in defining pragmatic responses to technological and organisational changes in the workplace and contribute to enhance the quality of the working environment 11 (OECD, 2018[1]; 2019[2]).

Figure 7.4. Workers’ voice and collective bargaining can improve labour market performance and the quality of the working environment

In addition, collective bargaining systems and workers’ voice arrangements also matter for job quality. The quality of the working environment is higher on average in countries with well-organised social partners and a large coverage of collective agreements. At the firm level, both direct and mixed forms of voice (where workers’ representatives coexist with direct dialogue between workers and managers) are also associated with a higher quality of the working environment compared to the absence of voice. By contrast, the presence of representative workers’ voice in firms where there are no parallel means of direct exchange between workers and managers is not associated with a better quality of the working environment. However, the presence of solely representative arrangements of voice could be characteristic of poor social dialogue contexts, where employers are unwilling to engage in direct exchanges with workers, so that workers react by mobilising formal worker representation bodies, thus blurring the empirical relationship between worker representation and quality jobs. While these results are not evidence of causal relationships, they highlight the importance of good labour relations and social dialogue context at the firm level (OECD, 2019[2]).

7.3. Social partners are engaging in outreach, awareness raising and advocacy

Depending on the national and regulatory settings as well as practices and traditions across OECD countries, social partners can engage in various initiatives and at different levels (e.g. workplace, firm, sectoral and national). They can raise voice and inform, advise policy makers, participate in decision-making for example when it comes to determining what technology is adopted, manage and fund programmes like training, negotiate agreements and monitor compliance of terms set out in agreements. Beyond these main activities, social partners are also increasing their efforts to broaden their outreach through the use of digital technologies – for example to attract, recruit and inform members through social platforms (Houghton and Hodder, 2021[36]) and to gain insights that strengthen their position in negotiations (Voss and Riede, 2018[5]). In this respect, AI technologies may also provide innovative solutions and new opportunities for social partners (for some examples, see Section 7.3.4).

Social partners have already taken several initiatives to shape the AI transition. Overall, the detailed review conducted for this chapter through the OECD questionnaires sent to social partners (Figure 7.5) suggests that social partners are mainly engaging in outreach and information activities, but very few have engaged in negotiating agreements.

Figure 7.5. Unions’ responses to the AI transition mainly concentrate on raising voice and advising policy

7.3.1. Raising voice and informing

Both unions and employers’ organisations across the OECD have engaged, at international and national levels, in outreach and awareness campaigns highlighting the need for new competences that will be required to work with digital tools, robotics and data and the need to become “AI literate”. (ETUI, 2021[37]) (ILO/IOE, 2019[38]; BusinessEurope, 2019[39]; UNI Europa ICTS, 2019[40]; ETUC, 2020[41]). Social partners have also expressed concerns about a number of issues, such as the trustworthy use of AI, ethical concerns, workers’ data privacy and protection, as well as training needs, mainly through position papers, guidelines about the application of AI and advice directed towards workers,12 employers but also policy makers (Cazes, 2021[42]).

Unions, for instance, are calling for greater involvement of workers and their representatives in AI-related decisions making. According to a survey from the German ver.di union, almost two-thirds of co-determination bodies at workplace and firm levels are not involved in the planning and implementation of AI projects, and one-third is not even aware of whether AI is being used (ver.di, 2019[6]). Against this backdrop, the European Trade Union Institute (ETUI) emphasises the need for a preventive engagement of workers and trade unions in the way algorithms are designed and deployed, calling for collective bargaining to ensure the interest of workers and fundamental rights are protected (ETUI, 2021[37]). This is echoed by national unions such as the Teamsters Union in the United States and the German Trade Union Confederation (DGB), which call for social dialogue and collective bargaining specifically over the parameters of AI-induced or exacerbated workplace surveillance (Teamster, 2018[43]; DGB, 2020[44]). The German union DGB also proposes a guiding framework for the introduction of AI and its deployment in a participative way (DGB, 2020[44]; Stowasser and Suchy, 2020[45]).

In addition, unions are calling for greater participation of workers and their representatives in the governance of AI adoption. For example, European social partners have proposed the adoption of data governance models for data stewardship in the form of data trusts, data collectives and co-operatives (Allen and Masters, 2021[46]; ETUC, 2020[41]; Colclough, 2020[47]; Ada Lovelace Institute and UK AI Council, 2021[48]; British Academy for the Humanites and Social Sciences/The Royal Society, 2017[49]). When used in the workplace, these governance mechanisms could provide workers with access and rights over the collection, analysis and storage of data that concerns them (Colclough, 2020[47]) – ultimately to promote a trustworthy and beneficial use of data that is collected or used by AI applications in the workplace (Salvi del Pero, Wyckoff and Vourc’h, 2022[23]).

Finally, unions have been very active in promoting a trustworthy use of AI and outlining training needs. For instance, the Association of Nordic Engineers proposes principles to strengthen transparency and develop technical standards and certifications to increase accountability (ANE/IT University of Copenhagen, 2018[50]; ANE et al., 2021[51]). Similarly, UNI Global Union proposes a list of principles relating to workers’ surveillance privacy and human dignity, which unions can use as guidance in negotiating agreements (UNI Global Union, 2019[52]; 2019[53]), while ETUI offers a capacity-building questionnaire for unions to go through when assessing the risks of algorithmic management in particular and forming initiatives in response (ETUI, 2021[37]). As for unions’ efforts to inform about the provision of training for workers affected by AI adoption, UNI Europa ICTS (2019[40]) has produced a position paper on AI adoption recommending social partners’ co-operation to identify training needs, design new education pathways, and find funding opportunities. This is also echoed by ETUC (2020[41]), which proposes AI and digital literacy schemes for workers to understand and be part of AI adoption at their workplace.

Employers and their representatives have also published a number of AI-related information and strategy papers, focusing on issues such as ensuring competitive advantage and growth (Ilsøe, 2017[54]; BusinessEurope, 2018[55]). These papers notably look at challenges for AI adoption, training needs, data sharing practices and cybersecurity, as well as funding issues. In its AI strategy, BusinessEurope (2020[56]) for instance proposes the creation of common European data spaces for business-to-business data access and sharing. In another paper, BusinessEurope (2019[39]) highlights the need to help workers establish a data culture and awareness of AI through re-skilling in job programmes, proposing to organise them through a cost-sharing approach – sponsored by the EU and co-ordinated by the European social partners. Along similar lines, the Confederation of British Industry (CBI) proposes the enhancement of social dialogue through the creation of joint commissions, comprising employers, academics, worker representatives and government officials in order to examine the impact of AI on jobs and jointly propose courses of actions (CBI, 2017[57]).

At the same time, a few employers’ organisations have started voicing concerns relating to a trustworthy use of AI (Salvi del Pero, Wyckoff and Vourc’h, 2022[23]). In its AI Utilisation Strategy, the Japan Business Federation Keidanren, for example, emphasises the need for ethical standards such as fairness, accountability and transparency, as well as rules that ensure a balance between the use and protection of personal data and guarantees for the safety and dependability of AI systems as a whole (Japan Business Federation-Keidanren, 2019[58]). Similarly, the World Employment Confederation (WEC) adopted a Code of Ethical Principles in the use of Artificial Intelligence (https://wecglobal.org/uploads/2023/04/AI-principles-WEC-AI-Code-of-Conduct-March-2023.pdf), while the US Chamber of Commerce’s Technology Engagement Center published a report with Deloitte, recommending the development of standards for AI trustworthiness, the rapid implementation of an AI risk management framework, and the development of international partnerships and standards including by the OECD (Deloitte/U.S. Chamber of Commerce Technology Engagement Center, 2021[59]).

7.3.2. Advising policy

Raising voice, informing and alerting can be ways to inform workers and employers, but also to shape policy debates. Additionally, some social partners have started explicitly calling for policy responses, which revolve around reviewing and further developing existing regulations in areas related to AI adoption, as well as closing regulatory gaps of AI-induced or exacerbated risks with new legislation.

Regarding the first aspect, social partners’ discussion has focuses to a large extent on data protection – and in Europe, the GDPR 13 is the most advanced legal instrument in Europe in this respect − but also on occupational health and safety issues, labour law and co-determination rights.14 Social partners have also started developing proposals for closing regulatory gaps. At the European level, ETUI for instance calls for European regulation that will ensure that AI algorithms will be required to have transparent purposes in the workplace 15 (Ponce del Castillo, 2020[60]). In its resolution, ETUC calls for the reinforcement of workers’ protections from undue surveillance, as well as from biased discrimination in the workplace (ETUC, 2020[41]). On the employers’ side, BusinessEurope (2020[56]) published a position paper on AI, which calls for legal certainty, specific responsibilities for all actors involved and a clear framework for firm compliance so that AI-based products are covered by a single set of clearly assigned product safety rules.

Finally, national unions across OECD countries are making proposals for new legislation in their countries. The British TUC, for example, proposes the introduction of a universal right to human review of high-risk decisions and the right of human contact when important decisions are made about people at work in addition to the right to data reciprocity giving workers the right to collect and combine workplace data (TUC, 2021[61]). The Association of Nordic Engineers also provides AI-related policy recommendations, including the need for defining responsibility (notably beyond the engineering profession) and the need for frameworks about explainability of AI-influenced decisions (ANE/IT University of Copenhagen, 2018[50]; ANE et al., 2021[51]). In the United States, the union AFL-CIO (2019[62]) highlights that, in the absence of data protection regulation similar to the European GDPR, platforms are already using algorithms and AI tools to make decisions about hiring and firing, promotions and work organisation that are often implemented without the consent of workers.

7.3.3. Negotiating new framework and collective agreements

Social partners have also started to provide guidance through framework agreements and, to a lesser extent, negotiate collective agreements. This modest engagement in bargaining activities reflects the scarcity of collective agreements on digital technologies more generally, especially with respect to non-monetary aspects of work (Kreinin, Artale and Kossow, 2022[63]). Moreover, the language of collective agreements that relates to new technologies may need to be updated to stay relevant, as highlighted for example in the AI OECD surveys in the United States and Canada (Milanez, 2023[64]).

In Europe, the European Social Partners Framework Agreement on Digitalisation (2020[65]) provides guidance on issues related to data, consent, privacy protection and surveillance, and the need to systematically link the collection and storage of data to ensure transparency – using the EU GDPR as a reference.16 The framework also calls for a fair deployment of AI systems, i.e. ensuring that workers and groups are free from unfair bias and discrimination. At sectoral level in the insurance and telecommunication sector, European social partners have also signed two framework agreements on AI that addressed similar elements (UNI Europa Finance; Insurance Europe; Amice; Bipar, 2021[66]; UNI Europa ICTS and ETNO, 2021[67]).

More recently, social partners have started engaging in “algorithm negotiations”, i.e. they are including as a subject of bargaining the use of artificial intelligence, big data and electronic performance monitoring (“people analytics”) in the workplace, as well as their implications for occupational health and safety, privacy, evaluation of work performance and hiring and firing decisions (De Stefano, 2018[20]).

To this date, a few AI-related collective agreements have already been signed in OECD countries. Although these agreements are rarely exclusively on AI, they include aspects of AI use and resulting implications for occupational health and safety, privacy, evaluation of work performance and hiring and firing decisions in other bargaining processes – see Box 7.3 and De Stefano (2018[20]). Moreover, several collective agreements have started regulating the use of AI not only in monitoring workers but also in directing their work (Moore, Upchurch and Whittaker, 2017[68]; OECD, 2019[2]).

At the same time, a lack of collective agreements specifically pertaining to AI-related issues in some countries may also reflect the strength of existing regulations and social dialogue structures. For example in Sweden, a report by the largest trade union in the private sector finds that the combination of existing collective agreements, ensured co-determination through the Workplace Act and other regulations including the Work Environment Act and the GDPR already provides a good basis for dealing with AI challenges relating to digital surveillance at the firm level – while legislation protecting personal integrity for instance could be strengthened (Unionen, 2022[69]).

Box 7.3. Algorithm negotiation and collective agreements

A few AI-related collective agreements have recently been signed in OECD countries and at various levels. In Spain, social partners in the banking sector signed an agreement in 2021 guaranteeing workers the right to not be subject to decisions taken solely by algorithmic systems, nor discriminated against based on algorithmic decisions. In addition, the agreement requires banks to inform workers’ representatives about the data and algorithmic systems used by “digital models with no human intervention” (Boletín Oficial del Estado, 2022[70]). Similarly, a tripartite national agreement for platform workers in Spain, has been signed that guarantees platform workers’ rights to algorithmic transparency. It required a modification of workers right to information, making it mandatory for platforms to inform workers’ legal representatives about the mathematical or algorithmic formula determining their working conditions (Aranguiz, 2021[71]).

At the local level, the General Staff Council of Stuttgart in Germany and the city as a public employer recently agreed on transparent design processes for new technologies, which in the case of AI must take place before its adoption (Forum Soziale Technikgestaltung, 2022[72]).

In Switzerland, the trade unions Syndicom and Transfair agreed at the firm level with Swisscom about a “Smart Data” policy in 2018, which includes principles when processing workers’ data (Syndicom/Swisscom/Transfair, 2018[73]). Similarly in the United States, the Teamsters union agreed with UPS in 2018 to establish a national and joint technological change committee to review any planned technological changes and to ensure affected workers are retrained instead of dismissed (Teamster/UPS, 2018[74]).

7.3.4. Making use of digital and AI tools

A few social partners have started reviewing how digital and AI technologies can facilitate their work and help address AI-specific concerns, such as reducing data asymmetries between workers, employers and their representatives. While they face difficulties in incorporating many of the technological features brought by digitalisation (Rotila, 2019[75]), social partners tend to agree that they need to make better use of digital tools to help them organise (Voss and Riede, 2018[5]).

The use of digital technologies for example provides an opportunity to social partners to increase their representation, self-organise and improve communication with their members (Adler-Bell and Miller, 2018[21]). In this respect, AI technologies may further support unions through increased outreach, especially to younger members and facilitate the renewal and management of memberships (Vandaele, 2018[76]). Along these lines, Flanagan and Walker (2020[77]) provide an illustration of an AI-enabled chatbot originally created by IBM, which was adapted for use by the alt-labour network Organization the United for Respect (OUR) to inform workers about their rights and then reconfigured for use by a more traditional union, the United Workers Union (UNU) in Australia. Similarly, the National Domestic Workers Alliance (NDWA) in the United States, which is not formally a union with legal bargaining rights but a non-profit organisation that campaigns for domestic workers’ rights, also developed a chatbot for Spanish speaking domestic workers. To increase the visibility of these workers’ experience during the COVID-19 pandemic, NDWA then adapted the chatbot into a survey tool and published related insights in a report (NDWA Labs, 2020[78]). The NDWA initiative highlights that beyond outreach, digital and AI tools could also inform social partners in their work 17 (Vandaele, 2018[76]) – a proposal that has recently been echoed by the British TUC, suggesting that unions could investigate ways to collect and make use of worker data, for example through engaging data scientists and developing AI-powered tools (TUC, 2021[61]).

One prominent tool along these lines is WeClock,18 an open-source self-tracking app for workers to gather key data about the time spent at work, commutes or mistreatments among many other aspects. Workers can then use this information for wage-related or similar negotiations with their own manager, but also hand it over to their unions ultimately to inform broader advocacy campaigns or bargaining processes (UNI Global Union, 2019[79]).

7.4. Promoting social partners’ expertise to shape the AI transition

Despite their involvement in developing initiatives to accompany AI adoption, social partners activities may be limited by their lack of AI-related knowledge (as reported in the OECD questionnaire in Box 7.1), as well as the lack of capacities and resources to attain it. Along these lines, it would be important not only to offer social partners training opportunities or secure expertise on AI at the workplace or firm levels – see Krämer and Cazes (2022[80]) for various examples across OECD countries, but also to consider allowing them access to some possibly sensitive information and data. In Spain, for instance, the 2021 Labour Law Reform, incorporated the right of the Workers Committee to be informed by the company of the parameters, rules and instructions on which algorithms or AI systems that may affect decisions relating to working conditions, profiling, etc.

One proposal to secure the necessary knowledge on AI at the workplace and firm levels beyond the training of social partners themselves is the recruitment or consultation of technical experts. This could not only ensure more technological understanding within unions and employers’ organisations, but also that worker interests are recognised in the workplaces where technology is being developed – which could in turn also contribute towards more trustworthy technology (TUC, 2021[17]).

Yet, while consulting technical experts could be promising ways to foster knowledge of AI in the workplace, social partners need government support to ensure that a broad access to such expertise can be provided. One recent example in this respect is the German Works Council Modernization Act passed in 2021, which grants works councils the right of consulting an external expert if the introduction or application of AI is in discussion 19 – for a discussion, see for example Maily (2021[81]) or Polkowski and Deja (2021[82]).

7.5. Concluding remarks

Social dialogue can play an important role in addressing some of the key challenges created by AI technologies. Previous OECD evidence has shown that when social partners work co-operatively, social dialogue can support and usefully complement public policies in easing technological transitions, for instance by identifying pragmatic solutions to labour market challenges at the firm level and anticipating skill needs (OECD, 2018[1]; 2019[2]). Moreover, collective bargaining systems, when co-ordinated, can reduce inequalities and foster inclusive labour markets. However, social partners are facing the challenges presented by AI technologies at a time when they are already dealing with ongoing pressures due, among other things, to declining representation.

This chapter presents new evidence on the role of social dialogue in shaping the AI transition and initiatives by social partners in this area. In doing so, it contributes to the implementation of the OECD AI Principles’ recommendation on “Building human capacity and preparing for labour market transformation” which states that governments should take steps, including through social dialogue, to ensure a fair transition for workers as AI is adopted. Governments should also work closely with stakeholders to promote the responsible use of AI at work, enhance the safety of workers and the quality of jobs, foster entrepreneurship and productivity, and ensure that the benefits AI are broadly shared.

In line with the existing empirical literature on the role of workers’ voice in mitigating the impact of automation on wages and employment, new findings concerning AI adoption suggest that the presence of workers’ representative bodies mitigates AI’s negative impact on working conditions, although these relationships are not necessarily causal. Mapping social partners’ ongoing responses to AI adoption, the chapter provides examples of social partners’ information campaigns, advocacy and the first AI-related agreements. However, many social partners are still at the very beginning of this process and face considerable challenges. In particular, they often lack AI-related expertise and do not have the capacity and resources to acquire this expertise.

Nevertheless, social partners could gain some AI-related expertise by joining forces and co-operating in the use of existing resources, such as capacity-building questionnaires, guidelines and similar information published by other social partners and governments. It is also important that social partners continue adapting to the changing world of work, particularly by reaching out to under-represented workers and businesses in AI-exposed sectors and occupations. Moreover, social partners can themselves seek to use AI and digital tools more broadly as these offer opportunities for outreach, organisation and bargaining activities, as well as for tackling issues caused or exacerbated by AI, such as information asymmetries. However, they have made little use of AI technologies for such purposes thus far.

Finally, some avenues exist for public policies to accompany social partners’ efforts to shape the AI transition. While each country’s situation and labour relations systems differ, policy makers could promote national consultations and discussions on the AI transition with social partners and other stakeholders, to discuss challenges such as training, data use, implementation in the workplace, as well as share practices on new initiatives through common knowledge platforms. They could also support the development of AI-related expertise in the workplace for management, workers and their representatives (through educational programmes, for example) and make it easier to bring external experts into the workplace.

Ultimately, the impact of AI on labour markets and workplaces will depend on how it is implemented – which includes both the role of regulation in AI adoption and the extent to which workers and employers are involved through social dialogue at workplace, firm, sectoral, national and international levels. In this respect, regulations and social dialogue can complement each other, for example when AI-related regulations set minimum standards and specify terms that require further dialogue and bargaining. To better understand the relationship between social dialogue, regulations and a beneficial AI transition for both workers and employers in the future, more data and analysis at the individual and firm levels will be necessary. In particular, this will require firm-level surveys that bring together information on AI adoption, social dialogue and labour market outcomes.

References

[31] Acemoglu, D. and P. Restrepo (2018), “The Race between Man and Machine: Implications of Technology for Growth, Factor Shares, and Employment”, American Economic Review, Vol. 108/6, https://doi.org/10.1257/aer.20160696.

[48] Ada Lovelace Institute and UK AI Council (2021), Exploring legal mechanisms for data stewardship, https://understandingpatientdata.org.uk/news/accountability-transparency-.

[11] Adams-Prassl, J. (2019), “What if Your Boss Was an Algorithm? The Rise of Artificial Intelligence at Work”, SSRN Scholarly Paper No. ID 3661151, Social Science Research Network, Rochester, NY, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3661151.

[21] Adler-Bell, S. and M. Miller (2018), The Datafication of Employment, How Surveillance and Capitalism Are Shaping Workers’ Futures without Their Knowledge, The Century Foundation.

[62] AFL-CIO (2019), Report of the AFL-CIO Commission on the Future of Work and Unions (Report to the AFL-CIO General Board).

[15] Albrecht, T. and C. Kellermann (2020), Künstliche Intelligenz und die Zukunft der digitalen Arbeitsgesellschaft: Konturen einer ganzheitlichen Technikfolgenabschätzung, Hans-Böckler-Stiftung, Düsseldorf, http://hdl.handle.net/10419/228963.

[46] Allen, R. and D. Masters (2021), Technology Managing People - The legal implications, Trade Union Congress, https://www.tuc.org.uk/research-analysis/reports/technology-managing-people-legal-implications.

[16] Aloisi, A. (2021), Bringing the algorithm to court, ETUI Workshop, https://www.slideshare.net/AntonioAloisi1/bringing-the-algorithm-to-court-etui-workshop (accessed on 7 January 2022).

[51] ANE et al. (2021), Addressing Ethical Dilemmas in AI: Listening to Engineers, Association of Nordic Engineers, https://nordicengineers.org/wp-content/uploads/2021/01/addressing-ethical-dilemmas-in-ai-listening-to-the-engineers.pdf.

[50] ANE/IT University of Copenhagen (2018), Nordic engineers’ stand on Artificial Intelligence and Ethics: Policy Recommendations and Guidelines, Association of Nordic Engineers, https://www.vfi.is/media/utgafa/32102-ANE-Anbefalinger-Rapport-A4-CMYK.pdf (accessed on 26 November 2021).

[71] Aranguiz, A. (2021), “Spain’s platform workers win algorithm transparency”, Social Europe, https://socialeurope.eu/spains-platform-workers-win-algorithm-transparency (accessed on 25 May 2022).

[34] Belloc, F. et al. (2022), “Coevolution of job automation risk and workplace governance”, Research Policy, Vol. 51/3, p. 104441, https://doi.org/10.1016/j.respol.2021.104441.

[29] Belloc, F., G. Burdin and F. Landini (2022), “Robots, Digitalization, and Worker Voice”, SSRN Electronic Journal, https://doi.org/10.2139/ssrn.4009136.

[70] Boletín Oficial del Estado (2022), Convenio colectivo del sector de la banca [Banking Collective Bargaining Agreement], https://www.boe.es/diario_boe/txt.php?id=BOE-A-2021-5003 (accessed on 24 May 2022).

[49] British Academy for the Humanites and Social Sciences/The Royal Society (2017), Data management and use: Governance in the 21st century A joint report by the British Academy and the Royal Society, https://www.thebritishacademy.ac.uk/documents/105/Data_management_and_use_-_Governance_in_the_21st_century.pdf.

[3] Brynjolfsson, E., D. Rock and C. Syverson (2017), “Artificial intelligence and the Modern Productivity Paradox: A Clash of Expectations and Statistics”, No. 24001, NBER Working Paper Series, https://doi.org/10.3386/w24001.

[56] BusinessEurope (2020), AI: A European approach to excellence and trust, BusinessEurope strategy paper, https://www.businesseurope.eu/publications/ai-european-approach-excellence-and-trust (accessed on 1 April 2022).

[39] BusinessEurope (2019), Artificial Intelligence recommendations, BusinessEurope strategy paper, https://www.businesseurope.eu/publications/artificial-intelligence-recommendations (accessed on 1 April 2022).

[55] BusinessEurope (2018), Robotics and automation, BusinessEurope strategy paper, https://www.businesseurope.eu/publications/robotics-and-automation-businesseurope-strategy-paper (accessed on 1 April 2022).

[65] BusinessEurope et al. (2020), European Social Partners Framework Agreement on Digitalisation, https://www.etuc.org/system/files/document/file2020-06/Final%2022%2006%2020_Agreement%20on%20Digitalisation%202020.pdf.

[42] Cazes, S. (2021), “The impact of Artificial Intelligence on the labour market and the workplace: What role for social dialogue?”, Global Deal Thematic Brief, Global Deal, Paris, https://www.theglobaldeal.com/news/The-impact-of-artificial-intelligence-on-the-labour-market-and-the-workplace.pdf.

[83] Cazes, S., A. Hijzen and A. Saint-Martin (2015), “Measuring and Assessing Job Quality: The OECD Job Quality Framework”, OECD Social, Employment and Migration Working Papers, No. 174, OECD Publishing, Paris, https://doi.org/10.1787/5jrp02kjw1mr-en.

[57] CBI (2017), Disrupting the future: embracing new technologies, The Business View, https://www.cbi.org.uk/articles/disrupting-the-future-embracing-new-technologies/ (accessed on 1 April 2022).

[18] Colclough, C. (2022), “Reshaping the Digitization of Public Services”, New England Journal of Public Policy, Vol. 34/1, Article 9, https://scholarworks.umb.edu/nejpp/vol34/iss1/9 (accessed on 26 July 2022).

[47] Colclough, C. (2020), Towards Workers’ Data Collectives, https://itforchange.net/digital-new-deal/2020/10/22/towards-workers-data-collectives/ (accessed on 21 October 2021).

[24] Data Protection Working Party (2017), Guidelines on Consent under Regulation 2016/679.

[9] Dauth, W. et al. (2021), “The Adjustment of Labor Markets to Robots”, Journal of the European Economic Association, Vol. 19/6, pp. 3104–3153, https://doi.org/10.1093/jeea/jvab012.

[4] De Stefano, V. (2019), Introduction: Automation, Artificial Intelligence, and Labour Protection, Social Science Research Network, Rochester, NY.

[20] De Stefano, V. (2018), “Negotiating the Algorithmm: Automation, Artificial Intelligence and Labour Protection”, SSRN Electronic Journal, https://doi.org/10.2139/ssrn.3178233.

[26] De Stefano, V. (2016), “The rise of the ‘just-in-time workforce’: On-demand work, crowdwork and labour protection in the ‘gig-economy’”, Conditions of Work and Employment, No. 71, International Labour Office.

[59] Deloitte/U.S. Chamber of Commerce Technology Engagement Center (2021), Investing in trustworthy AI.

[44] DGB (2020), Artificial Intelligence (AI) for Good Work, Deutscher Gewerkschaftsbund, https://www.dgb.de/downloadcenter/++co++b794879a-9f2e-11ea-a8e8-52540088cada.

[22] EESC/CFDT Cadres (2022), A guide to Artificial Intelligence at the workplace, European Economic and Social Committee in partnership with CFDT Cadres, https://www.eesc.europa.eu/en/our-work/publications-other-work/publications/guide-artificial-intelligence-workplace (accessed on 17 August 2022).

[41] ETUC (2020), Resolution on the European strategies on artificial intelligence and data, https://www.etuc.org/en/document/resolution-european-strategies-artificial-intelligence-and-data.

[37] ETUI (2021), Algorithmic management and collective bargaining, European Trade Union Institute (ETUI), https://www.etui.org/publications/algorithmic-management-and-collective-bargaining.

[33] EU-OSHA (2019), The Third European Survey of Enterprises on New and Emerging Risks, https://visualisation.osha.europa.eu/esener/en/survey/overview/2019 (accessed on 20 April 2022).

[77] Flanagan, F. and M. Walker (2020), How can unions use Artificial Intelligence to build power? The use of AI chatbots for labour organising in the US and Australia, New Technol. Work Employ, https://doi.org/10.1111/ntwe.12178.

[72] Forum Soziale Technikgestaltung (2022), Gesamtpersonalrat erreicht epochale Vereinbarung zur Gestaltung der Digitalisierung und der sogenannten „KI“, Blog: Zukunft der Arbeit, http://www.blog-zukunft-der-arbeit.de/gesamtpersonalrat-erreicht-epochale-vereinbarung-zur-gestaltung-der-digitalisierung-und-der-sogenannten-ki/ (accessed on 22 August 2022).

[28] Genz, S., L. Bellmann and B. Matthes (2018), Do German Works Councils Counter or Foster the Implementation of Digital Technologies?, http://hdl.handle.net/10419/185076.

[10] Georgieff, A. and R. Hyee (2021), “Artificial intelligence and employment : New cross-country evidence”, OECD Social, Employment and Migration Working Papers, No. 265, OECD Publishing, Paris, https://doi.org/10.1787/c2c1d276-en.

[8] Georgieff, A. and A. Milanez (2021), “What happened to jobs at high risk of automation?”, OECD Social, Employment and Migration Working Papers, No. 255, OECD Publishing, Paris, https://doi.org/10.1787/10bc97f4-en.

[32] Haapanala, H., I. Marx and Z. Parolin (2022), “Robots and Unions: The Moderating Effect of Organised Labour on Technological Unemployment”, IZA Discussion Paper Series No. 15080, https://www.iza.org/publications/dp/15080.

[36] Houghton, D. and A. Hodder (2021), “Understanding trade union usage of social media: A case study of the Public and Commercial Services union on Facebook and Twitter”, New Technology, Work and Employment, Vol. 36/2, pp. 219-239, https://doi.org/10.1111/ntwe.12209.

[38] ILO/IOE (2019), Changing Business and Opportunities for Employer and Business Organizations, International Labour Organization and International Organisation of Employers, https://www.ilo.org/actemp/areas-of-work/WCMS_679582/lang--en/index.htm.

[54] Ilsøe, A. (2017), “The digitalisation of service work – social partner responses in Denmark, Sweden and Germany”, Transfer: European Review of Labour and Research, Vol. 23/3, pp. 333-348, https://doi.org/10.1177/1024258917702274.

[7] INPUT Consulting/humAIn work lab (2021), Online Befragung “Künstliche Intelligenz im Betrieb”: Handlunsgfelder und Gestaltungsfordernisse für Interessenvertreter*innen, https://www.input-consulting.de/aktuelles/artikel/kuenstliche-intelligenz-im-betrieb-handlungsfelder-und-gestaltungserfordernisse-fuer-betriebliche-interessenvertreter-innen.html (accessed on 17 August 2022).

[58] Japan Business Federation-Keidanren (2019), AI Utilization Strategy: For an AI-Ready Society, https://www.keidanren.or.jp/en/policy/2019/013_outline.pdf (accessed on 26 November 2021).

[84] Klengel, E. and J. Wenckebach (2021), “Artificial intelligence, work, power imbalance and democracy – why co-determination is essential”, Italian Labour Law e-Journal, Vol. 14/2, https://doi.org/10.6092/issn.1561-8048/14099.

[80] Krämer, C. and S. Cazes (2022), “Shaping the transition: Artificial intelligence and social dialogue”, OECD Social, Employment and Migration Working Papers, No. 279, OECD Publishing, Paris, https://doi.org/10.1787/f097c48a-en.

[63] Kreinin, H., W. Artale and N. Kossow (2022), Trade unions and the multiple crisis of environment, society, economy and work, The Foundation for European Progressive Studies (FEPS) and the Karl Renner Institute, https://www.feps-europe.eu/attachments/publications/ps%20-%20trade%20unions_3.pdf (accessed on 17 March 2022).

[14] Lane, M. and A. Saint-Martin (2021), “The impact of Artificial Intelligence on the labour market : What do we know so far?”, OECD Social, Employment and Migration Working Papers, No. 256, OECD Publishing, Paris, https://doi.org/10.1787/7c895724-en.

[35] Lane, M., M. Williams and S. Broecke (2023), “The impact of AI on the workplace: Main findings from the OECD AI surveys of employers and workers”, OECD Social, Employment and Migration Working Papers, No. 288, OECD Publishing, Paris, https://doi.org/10.1787/ea0a0fe1-en.

[12] Liu, Y. et al. (2021), “The Coin of AI Has Two Sides: Matching Enhancement and Information Revelation Effects of AI on Gig-Economy Platforms”, SSRN Electronic Journal, https://doi.org/10.2139/ssrn.3877868.

[81] Maily, K. (2021), Works Council Modernization Act – Little modernization and a lot of regulation?, Hogan Lovells, Legal Insights and Analysis, https://www.engage.hoganlovells.com/knowledgeservices/news/works-council-modernization-act-little-modernization-and-a-lot-of-regulation/ (accessed on 25 May 2022).

[13] Malik, A., P. Budhwar and N. Srikanth (2020), “Gig Economy, 4IR and Artificial Intelligence: Rethinking Strategic HRM”, in Human & Technological Resource Management (HTRM): New Insights into Revolution 4.0, Emerald Publishing Limited, https://doi.org/10.1108/978-1-83867-223-220201005.

[64] Milanez, A. (2023), “The impact of AI on the workplace: Evidence from OECD case studies of AI implementation”, OECD Social, Employment and Migration Working Papers, No. 289, OECD Publishing, Paris, https://doi.org/10.1787/2247ce58-en.

[25] Moore, P. (2020), Surveillance & monitoring: The future of work in the digital era, STOA.

[68] Moore, P., M. Upchurch and X. Whittaker (2017), “Humans and Machines at Work: Monitoring, Surveillance and Automation in Contemporary Capitalism”, in Humans and Machines at Work, Springer International Publishing, Cham, https://doi.org/10.1007/978-3-319-58232-0_1.

[78] NDWA Labs (2020), 6 Months in Crisis: The Impact of COVID-19 on Domestic Workers, National Domestic Workers Alliance, https://static1.squarespace.com/static/5fa48611e8c1e35b60fa89a7/t/60da342ba6204756da958934/1624912940692/6+Months+in+Crisis+-+the+Impact+of+COVID-19+on+Domestic+Workers_NDWA+Labs.pdf.

[2] OECD (2019), Negotiating Our Way Up: Collective Bargaining in a Changing World of Work, OECD Publishing, Paris, https://doi.org/10.1787/1fd2da34-en.

[1] OECD (2018), OECD Employment Outlook 2018, OECD Publishing, Paris, https://doi.org/10.1787/empl_outlook-2018-en.

[27] Onorato, D. (2018), “Robots, Unions, and Aging: Determinants of Robot Adoption Evidence from OECD Countries”, Atlantic Economic Journal, Vol. 46/4, pp. 473-474, https://doi.org/10.1007/s11293-018-9599-1.

[30] Parolin, Z. (2019), “Automation and occupational wage trends: What role for unions and collective bargaining?”, LIS Working Paper Series, No. 767, http://hdl.handle.net/10419/203053.

[82] Polkowski, N. and A. Deja (2021), Reform of the works constitution law through the Betriebsrätemodernisierungsgesetz (Works Council Modernisation Act), Deloitte, https://www2.deloitte.com/dl/en/pages/legal/articles/reform-betriebsrat.html (accessed on 25 May 2022).

[60] Ponce del Castillo, A. (2020), “Labour in the Age of AI: Why Regulation is Needed to Protect Workers”, SSRN Electronic Journal, https://doi.org/10.2139/ssrn.3541002.

[19] Rani, U. and P. Singh (2019), “Digital Platforms, data, and development: Implications for Workers in Developing Economies”, in Comparative Labour Law and Policy Journal, Automation, Artificial Intelligence, & Labor Law.

[75] Rotila, V. (2019), “The Future of Social Dialogue in the Age of Artificial Intelligence”, Postmodern Openings, Vol. 10/3, https://doi.org/10.18662/po/87.

[23] Salvi del Pero, A., P. Wyckoff and A. Vourc’h (2022), “Using Artificial Intelligence in the workplace: What are the main ethical risks?”, OECD Social, Employment and Migration Working Papers, No. 273, OECD Publishing, Paris, https://doi.org/10.1787/840a2d9f-en.

[45] Stowasser, S. and O. Suchy (2020), “Good change management: How AI becomes successful in companies”, Lernende Systeme – Germany’s Platform for Artificial Intelligence, Munich, https://www.plattform-lernende-systeme.de/aktuelles-newsreader-24/good-change-management-how-ai-becomes-successful-in-companies.html (accessed on 22 August 2022).

[73] Syndicom/Swisscom/Transfair (2018), Collective Employment Agreement, https://syndicom.ch/fileadmin/user_upload/Web/Website/GAV/18_Swisscom/EN/GAV_Swisscom_2018_en.pdf.

[85] TCO (2021), Europeiska kommissionens förslag till förordning om harmoniserade regler för artificiell intelligens [The European Commission’s proposal for a regulation on harmonized rules for artificial intelligence: Referral response: 25 June 2021], Swedish Confederation of Professional Employees, https://tco.se/fakta-och-politik/arbetsmarknad/europeiska-kommissionens-forslag-till-forordning-om-harmoniserade-regler-for-artificiell-intelligens (accessed on 16 June 2023).

[43] Teamster (2018), The Future of Work is Now bargaining strategies: From Machine Breaking to Artificial Intelligence, https://teamster.org/wp-content/uploads/2018/12/tlc3092518topic7presenterdrosenfeld.pdf.

[74] Teamster/UPS (2018), National Master United Parcel Service Agreement, https://teamster.org/wp-content/uploads/2018/12/ups18nationalmaster.pdf.

[61] TUC (2021), Dignity at Work and the AI Revolution: A TUC Manifesto, Trade Union Congress, London, https://www.tuc.org.uk/research-analysis/reports/dignity-work-and-ai-revolution.

[17] TUC (2021), When AI is the boss, Trade Union Congress, London, https://www.tuc.org.uk/sites/default/files/When_AI_Is_The_Boss_2021_Reps_Guide_AW_Accessible.pdf.

[66] UNI Europa Finance; Insurance Europe; Amice; Bipar (2021), Joint declaration on Artificial Intelligence by the European social partners in the insurance sector, https://www.uni-europa.org/wp-content/uploads/2021/03/20210316_ISSDC_Joint-declaration-on-Artificial-Intelligence-1.pdf (accessed on 13 April 2021).

[40] UNI Europa ICTS (2019), Position on Artificial Intelligence, UNI Europa.

[67] UNI Europa ICTS and ETNO (2021), Joint declaration on Artificial Intelligence by the European social partners in the telecom sector.

[79] UNI Global Union (2019), Connective Action: Digital Tools for Trade Unions, UNI Global Union, Nyon, Switzerland, http://www.thefutureworldofwork.org/the-lab/report-connective-action/ (accessed on 20 May 2022).

[53] UNI Global Union (2019), Top 10 Principles for Ethical Artificial Intelligence, UNI Global Union, http://www.thefutureworldofwork.org/opinions/10-principles-for-ethical-ai/.

[52] UNI Global Union (2019), Top 10 Principles for Workers’ Data Privacy and Protection, UNI Global Union, http://www.thefutureworldofwork.org/docs/10-principles-for-workers-data-rights-and-privacy/.

[69] Unionen (2022), Total Data Control at Work? Surveillance and Data-driven Decisions in the Workplace, Unionen Sweden, https://www.unionen.se/en/filer/rapport/total-data-control-work.

[76] Vandaele, K. (2018), “Will trade unions survive in the platform economy? Emerging patterns of platform workers’collective voice and representation in Europe”, The European Trade Union Institute, https://www.etui.org/publications/working-papers/will-trade-unions-survive-in-the-platform-economy-emerging-patterns-of-platform-workers-collective-voice-and-representation-in-europe.

[6] ver.di (2019), Innovationsbarometer 2019: Künstliche Intelligenz, https://innovation-gute-arbeit.verdi.de/innovation/innovationsbarometer/++co++10b323c8-0ad2-11ea-a165-525400f67940 (accessed on 24 May 2022).

[5] Voss, E. and H. Riede (2018), Digitalisation and workers participation: What trade unions, company level workers and online platform workers in Europe think, European Trade Union Confederation (ETUC), https://www.etuc.org/sites/default/files/publication/file/2018-09/Voss%20Report%20EN2.pdf. (accessed on 3 June 2021).

Annex 7.A. OECD Questionnaire on AI impact and social dialogue

Annex Table 7.A.1. List of social partners contributing to the OECD questionnaire
Name of the organisation	Type of organisation	Country	Domain	Representativeness
Association of Nordic Engineers (ANE)	Trade union association	Denmark, Finland, Iceland, Norway, and Sweden	Association of engineer’s trade unions co-ordinating positions across Nordic countries.	More than 500 000 engineers across Nordic countries.
Australian Chamber of Commerce and Industry (ACCI)	Employer association	Australia	Private sector companies of all sizes.	9 000 companies
Canadian Chamber of Commerce (CCC)	Employer association	Canada	Private sector companies of all sizes.	200 000 businesses.
Central Organisation of Finnish Trade Unions (SAK)	Trade union confederation	Finland	All workers in manufacturing, public sector, transport, private services, and culture.	861 371 employees in 2021 (around 45% of total union membership).
Confederation of Danish Employers (DA)	Employer organisation	Denmark	Private sector companies of all sizes.	25 000 companies representing 1/3 of all employees in Denmark
Confederation of German Employers’ Associations (BDA)	Employer organisation	Germany	Private sector employers in manufacturing industry, commerce, banking, insurance, small-scale crafts and trades, agriculture, transport and the newspaper industry	48 sectoral employer organisations and 14 regional employer organisations
Confederation of Professional Associations in Denmark (Akademikerne)	Trade union confederation	Denmark	Professional and managerial staff graduated from universities and other higher educational institutions	302 626 employees in 2021 (16% of total union membership).
Confederation of Swedish Enterprise (SN)	Employer organisation	Sweden	Private sector companies of all sizes.	60 000 firms and 50 business and employer organisations
Financial Services Union (FSU)	Trade union	Ireland	Financial Services, Fintech and Tech sectors. Trade union affiliated to the Irish Congress of Trade Unions (ICTU).	9 800 employees (around 2% of total union membership)
French Confederation of Professional and Managerial Staff – General Confederation of Professional and Managerial Staff (CFE-CGC)	Trade union confederation	France	White collars in private and public sectors.	152 000 union members in 2019 (6.5% of total union membership).
General Confederation of Liberal Trade Unions of Belgium (CGSLB/ACLVB)	Trade union confederation	Belgium	Blue and white collars in private and public sectors.	Around 300 000 union members in 2019 (9% of total union membership in Belgium).
General Labour Federation of Belgium (FGTB/ABVV)	Trade union confederation	Belgium	Blue and white collars in private and public sectors.	Around 1 500 000 union members in 2019 (45% of total union membership in Belgium)
German Trade Union Confederation (DGB)	Trade union confederation	Germany	Blue and white collars in private and public sectors.	5 934 000 union members in 2019 (75% of total union membership).
Hellenic Federation of Enterprises (SEV)	Employer organisation	Greece	Private sector companies of all sizes.	Over 600 direct member companies
New Zealand Educational Institute (NZEI)	Trade union	New Zealand	Education (public sector). Trade union affiliated to the New Zealand Council of Trade Unions (NZCTU).	Around 50 000 employees in 2019 (13.5% of total union membership).
Prospect	Trade union	United Kingdom	Scientists, engineers, and tech experts of the private sector. Trade union affiliated to the Trades Union Congress (TUC)	Around 150 000 employees (2.7% of total union membership).
Public Service Association (PSA)	Trade union	New Zealand	Central government, state-owned enterprises, local councils, health boards and community groups. Trade union affiliated to the New Zealand Council of Trade Unions (NZCTU).	Around 80 000 employees in 2022 (31.5% of total union membership).
Trades Union Congress (TUC)	Trade union confederation	United Kingdom	Blue and white collars in private and public sectors.	5 522 739 employees in 2019.
Union of Italian Workers (UIL)	Trade union confederation	Italy	Blue and white collars in private and public sectors.	1 376 000 employees in 2021 (around 23% of total union membership).
Unionen	Trade union	Sweden	White-collar workers in the private sector. Trade union affiliated to the Swedish Confederation of Professional Employees (TCO).	592 400 union members in 2021 (19% of total union membership).
Note: An employer confederation is specialised in representing only interests related to the labour market and industrial relations; an employer association represents only the product market interests of their members.

Extracts of the OECD questionnaire on AI impact and social dialogue

Trade union questionnaire	Employer organisation questionnaire
Question 2.1. When thinking of AI technologies and their impact on workers, what are according to you the three most important opportunities/benefits? □ Increased productivity gains through more efficient work processes and greater economic competitiveness, with potentially positive effects on wages □ Creation of new tasks and jobs (including data analysts, computer engineers, scientists, network experts…) □ Higher job quality through effective collaboration between workers and AI, for instance resulting in a reduction of working time, a possible focus on more interesting tasks and increased work autonomy □ Higher job safety, for example because existing machines will be safer or certain dangerous, physically demanding tasks are supported or performed by AI □ Reductions in human bias in HR and management processes among others, increasing fairness for discriminated groups and minorities □ Other	Question 2.1. When thinking of AI technologies and their impact on businesses, what are according to you the three most important opportunities/benefits? □ Increased productivity gains through more efficient work processes and greater economic competitiveness □ Creation of new tasks and jobs (including data analysts, computer engineers, scientists, network experts…) □ Higher job quality through effective collaboration between workers and AI, for instance resulting in a reduction of working time, a possible focus on more interesting tasks and increased work autonomy □ Higher job safety, for example because existing machines will be safer or certain dangerous, physically demanding tasks are supported or performed by AI □ Reductions in human bias in HR and management processes among others, increasing fairness for discriminated groups and minorities □ Other
Question 2.2. When thinking of AI technologies and their impact on workers, what are according to you the three most important challenges/risks? □ Destruction of jobs and displacement of workers □ Widening inequalities between workers, as occupations and sectors may experience different wage growth or decline as a result of AI applications □ Rapidly changing/increasing skill requirements, associated with high re- and upskilling costs for both businesses and workers □ Lower job quality, for example through work intensification or less personal work relationships □ Health and psychological risks at the workplace, for example through excessive AI-based surveillance □ Concerns regarding data privacy, data leakages, possible violations of worker’s human rights and dignity, and discriminations through automated decisions based on biased data □ Liability risks for decisions and outcomes caused by AI applications □ Weakening collective action and social dialogue, for example through increasing information asymmetries and physical distance between workers and businesses □ Other	Question 2.2. When thinking of AI technologies and their impact on businesses, what are according to you the three most important challenges/risks? □ Impairment or displacement of business models □ Unfair competition between businesses possessing different amounts and kinds of data, with effects on revenues, consumer prices and wage □ Rapidly changing/increasing skill requirements, associated with increased re- and upskilling costs for both businesses and workers □ Lower job quality, for example through work intensification or less personal work relationships □ Health and psychological risks at the workplace, for example through excessive AI-based surveillance □ Concerns regarding data privacy, data leakages, possible violations of workers’ human rights and dignity, and discriminations through automated decisions based on biased data □ Liability risks for decisions and outcomes caused by AI applications □ Weakening collective action and social dialogue, for example through an increasing physical distance between workers and businesses □ Other
Question 3.1. What role does AI adoption in the workplace currently play in your union’s agenda and strategy? Please select all options that apply. □ We have not considered AI in our agenda yet □ We have not considered AI in our agenda yet, but are planning to do so □ We are discussing possible strategies and initiatives □ We are developing concrete strategies and initiatives □ We have already implemented concrete strategies and initiatives □ We are in contact/co-operation with other national or international partners on the role of AI in our agenda □ Other	Question 3.1. What role does AI adoption in the workplace currently play in your employer organisation’s agenda and strategy? Please select all options that apply. □ We have not considered AI in our agenda yet □ We have not considered AI in our agenda yet, but are planning to do so □ We are discussing possible strategies and initiatives □ We are developing concrete strategies and initiatives □ We have already implemented concrete strategies and initiatives □ We are in contact/co-operation with other national or international partners on the role of AI in our agenda □ Other
Question 3.3. Which of the following initiatives has your union already carried out to support workers in the transition to AI adoption? Please select all applicable options. □ Outreach and information activities, such as information sessions and awareness campaigns □ Guides and principle frameworks to facilitate the understanding and use of AI applications □ Publication of an AI strategy or similar documents □ Training activities, such as workshops, online courses or partnerships with educational institutions □ Structural adjustments in the union, such as new working groups, consultation meetings, contact points for AI-related questions or the engagement of new personnel □ Identification of issues for bargaining activities and collective agreements □ Engagement in bargaining activities and collective agreements □ Involvement in monitoring and compliance procedures relating to AI applications □ Targeted communication and contact with policy makers to initiate and shape AI regulations or support measures □ Other	Question 3.3. Which of the following initiatives has your employer organisation already carried out to support businesses in the transition to AI adoption? Please select all applicable options. □ Outreach and information activities, such as information sessions and awareness campaigns □ Guides and principle frameworks to facilitate the understanding and use of AI applications □ Publication of an AI strategy or similar documents □ Training activities, such as workshops, online courses or partnerships with educational institutions □ Structural adjustments in the employer organisation, such as new working groups, consultation meetings, contact points for AI-related questions or the engagement of new personnel □ Identification of issues for bargaining activities and collective agreements □ Engagement in bargaining activities and collective agreements □ Involvement in monitoring and compliance procedures relating to AI applications □ Targeted communication and contact with policy makers to initiate and shape AI regulations or support measures □ Other

Trade union questionnaire

Employer organisation questionnaire

Question 2.1. When thinking of AI technologies and their impact on workers, what are according to you the three most important opportunities/benefits?

□ Increased productivity gains through more efficient work processes and greater economic competitiveness, with potentially positive effects on wages

□ Creation of new tasks and jobs (including data analysts, computer engineers, scientists, network experts…)

□ Higher job quality through effective collaboration between workers and AI, for instance resulting in a reduction of working time, a possible focus on more interesting tasks and increased work autonomy

□ Higher job safety, for example because existing machines will be safer or certain dangerous, physically demanding tasks are supported or performed by AI

□ Reductions in human bias in HR and management processes among others, increasing fairness for discriminated groups and minorities

□ Other

Question 2.1. When thinking of AI technologies and their impact on businesses, what are according to you the three most important opportunities/benefits?

□ Increased productivity gains through more efficient work processes and greater economic competitiveness

□ Creation of new tasks and jobs (including data analysts, computer engineers, scientists, network experts…)

□ Higher job safety, for example because existing machines will be safer or certain dangerous, physically demanding tasks are supported or performed by AI

□ Reductions in human bias in HR and management processes among others, increasing fairness for discriminated groups and minorities

□ Other

Question 2.2. When thinking of AI technologies and their impact on workers, what are according to you the three most important challenges/risks?

□ Destruction of jobs and displacement of workers

□ Widening inequalities between workers, as occupations and sectors may experience different wage growth or decline as a result of AI applications

□ Rapidly changing/increasing skill requirements, associated with high re- and upskilling costs for both businesses and workers

□ Lower job quality, for example through work intensification or less personal work relationships

□ Health and psychological risks at the workplace, for example through excessive AI-based surveillance

□ Concerns regarding data privacy, data leakages, possible violations of worker’s human rights and dignity, and discriminations through automated decisions based on biased data

□ Liability risks for decisions and outcomes caused by AI applications

□ Weakening collective action and social dialogue, for example through increasing information asymmetries and physical distance between workers and businesses

□ Other

Question 2.2. When thinking of AI technologies and their impact on businesses, what are according to you the three most important challenges/risks?

□ Impairment or displacement of business models

□ Unfair competition between businesses possessing different amounts and kinds of data, with effects on revenues, consumer prices and wage

□ Rapidly changing/increasing skill requirements, associated with increased re- and upskilling costs for both businesses and workers

□ Lower job quality, for example through work intensification or less personal work relationships

□ Health and psychological risks at the workplace, for example through excessive AI-based surveillance

□ Concerns regarding data privacy, data leakages, possible violations of workers’ human rights and dignity, and discriminations through automated decisions based on biased data

□ Liability risks for decisions and outcomes caused by AI applications

□ Weakening collective action and social dialogue, for example through an increasing physical distance between workers and businesses

□ Other

Question 3.1. What role does AI adoption in the workplace currently play in your union’s agenda and strategy? Please select all options that apply.

□ We have not considered AI in our agenda yet

□ We have not considered AI in our agenda yet, but are planning to do so

□ We are discussing possible strategies and initiatives

□ We are developing concrete strategies and initiatives

□ We have already implemented concrete strategies and initiatives

□ We are in contact/co-operation with other national or international partners on the role of AI in our agenda

□ Other

Question 3.1. What role does AI adoption in the workplace currently play in your employer organisation’s agenda and strategy? Please select all options that apply.

□ We have not considered AI in our agenda yet

□ We have not considered AI in our agenda yet, but are planning to do so

□ We are discussing possible strategies and initiatives

□ We are developing concrete strategies and initiatives

□ We have already implemented concrete strategies and initiatives

□ We are in contact/co-operation with other national or international partners on the role of AI in our agenda

□ Other

Question 3.3. Which of the following initiatives has your union already carried out to support workers in the transition to AI adoption? Please select all applicable options.

□ Outreach and information activities, such as information sessions and awareness campaigns

□ Guides and principle frameworks to facilitate the understanding and use of AI applications

□ Publication of an AI strategy or similar documents

□ Training activities, such as workshops, online courses or partnerships with educational institutions

□ Structural adjustments in the union, such as new working groups, consultation meetings, contact points for AI-related questions or the engagement of new personnel

□ Identification of issues for bargaining activities and collective agreements

□ Engagement in bargaining activities and collective agreements

□ Involvement in monitoring and compliance procedures relating to AI applications

□ Targeted communication and contact with policy makers to initiate and shape AI regulations or support measures

□ Other

Question 3.3. Which of the following initiatives has your employer organisation already carried out to support businesses in the transition to AI adoption? Please select all applicable options.

□ Outreach and information activities, such as information sessions and awareness campaigns

□ Guides and principle frameworks to facilitate the understanding and use of AI applications

□ Publication of an AI strategy or similar documents

□ Training activities, such as workshops, online courses or partnerships with educational institutions

□ Structural adjustments in the employer organisation, such as new working groups, consultation meetings, contact points for AI-related questions or the engagement of new personnel

□ Identification of issues for bargaining activities and collective agreements

□ Engagement in bargaining activities and collective agreements

□ Involvement in monitoring and compliance procedures relating to AI applications

□ Targeted communication and contact with policy makers to initiate and shape AI regulations or support measures

□ Other

Annex 7.B. Supplementary materials

Annex Figure 7.B.1. Representative workers’ voice is associated with mitigating AI’s impacts on risks relating to working conditions, but causality remains unclear – OLS estimates

Annex Table 7.B.1. Representative workers’ voice and AI’s impacts on risks relating to working conditions
Marginal effects, i.e. percentage change in the probability of outcome variable following a discrete change in the relevant explanatory variable
	Hard physical conditions	Painful positions	Heavy loads	Repetitive arm movements	High / low temperature	High noise	Fumes, Vapours or Chemical products	Work intensity	Work at very high speed	Long working hours	Social support at work
Work Council (WC)
AI adoption (β₁)	5.56***	7.58***	8.92***	9.21***	7.06***	8.67***	9.41***	11.00***	10.5***	6.29***	-5.23***
AI adoption (β₁)	(0.007)	(0.008)	(0.010)	(0.009)	(0.009)	(0.008)	(0.009)	(0.010)	(0.010)	(0.007)	(0.007)
Representation by WC (β₂)	1.66**	3.95***	4.09***	3.03***	3.38***	3.44***	2.55**	3.81***	4.16***	1.84**	-3.48***
Representation by WC (β₂)	(0.007)	(0.009)	(0.011)	(0.010)	(0.010)	(0.009)	(0.010)	(0.011)	(0.011)	(0.008)	(0.007)
AI adoption and WC (β₃)	0.73	-0.74	-3.21*	2.46	0.31	-1.29	-1.90	-1.75	-1.49	-2.93**	2.14*
AI adoption and WC (β₃)	(0.014)	(0.016)	(0.019)	(0.018)	(0.017)	(0.015)	(0.017)	(0.019)	(0.018)	(0.014)	(0.011)
Trade union (TU)
AI adoption (β₁)	5.94***	8.11***	8.73***	9.49***	7.44***	8.56***	9.07***	10.8***	10.10***	6.14***	-4.36***
AI adoption (β₁)	(0.007)	(0.008)	(0.010)	(0.009)	(0.009)	(0.008)	(0.009)	(0.010)	(0.010)	(0.007)	(0.007)
Representation by TU (β₂)	3.25***	5.68***	7.26***	3.47***	5.63***	7.03***	6.33***	6.43***	7.05***	3.78***	-4.04***
Representation by TU (β₂)	(0.008)	(0.011)	(0.013)	(0.011)	(0.012)	(0.010)	(0.011)	(0.012)	(0.012)	(0.010)	(0.008)
AI adoption and TU (β₃)	2.01	-2.22	-2.01	0.61	-0.33	-2.87*	-2.88	0.35	-1.03	-1.01	1.38
AI adoption and TU (β₃)	(0.016)	(0.017)	(0.020)	(0.019)	(0.018)	(0.016)	(0.018)	(0.021)	(0.020)	(0.015)	(0.012)
Health and Safety Representative (HSR)
AI adoption (β₁)	5.80***	8.27***	9.94***	8.79***	7.24***	9.35***	9.41***	11.20***	10.70***	6.17***	-5.93***
AI adoption (β₁)	(0.007)	(0.010)	(0.011)	(0.010)	(0.010)	(0.010)	(0.011)	(0.011)	(0.011)	(0.009)	(0.008)
Representation by HSR (β₂)	2.62***	2.51***	5.40***	3.02***	2.83***	3.48***	3.99***	2.07**	2.85***	0.64	-2.06***
Representation by HSR (β₂)	(0.006)	(0.008)	(0.010)	(0.009)	(0.009)	(0.008)	(0.009)	(0.009)	(0.009)	(0.007)	(0.006)
AI adoption and HSR (β₃)	0.70	-1.96	-4.30***	2.24	0.05	-2.94**	-1.70	-0.13	-0.29	-0.88	3.19***
AI adoption and HSR (β₃)	(0.012)	(0.014)	(0.017)	(0.016)	(0.015)	(0.014)	(0.015)	(0.016)	(0.016)	(0.012)	(0.011)
Note: Results are based on probit regressions including establishment-level controls (country, industry, size, age, economic situation, share of old workers, if the person interviewed is the owner/manager and if teleworking is possible). , , ** denote the statistically significance at the 10%, 5%, and 1% level, respectively. Robust standard errors in parentheses.
Source: OECD estimates based on the Third European Survey of Enterprises on New and Emerging Risks (2019) (ESENER-3).
StatLink https://stat.link/mgp8fz

Annex Table 7.B.2. Definitions of the job quality outcomes
Job quality outcome	ESENER questions	ESENER variable	ESENER coding	Final coding
Painful positions (d1)	Is this risk factor present or not in your establishment? regardless of whether it is currently under control and regardless of the number of employees it affects: Tiring or painful positions	Q200_4	Yes/No	1 if Q200_4=Yes; 0 otherwise
Heavy loads (d2)	Is this risk factor present or not in your establishment? regardless of whether it is currently under control and regardless of the number of employees it affects: Lifting or moving people or heavy loads	Q200_1	Yes/No	1 if Q200_1=Yes; 0 otherwise
Repetitive arm movements (d3)	Is this risk factor present or not in your establishment? regardless of whether it is currently under control and regardless of the number of employees it affects: Repetitive hand or arm movements	Q200_2	Yes/No	1 if Q200_2=Yes; 0 otherwise
High temperatures or low temperatures (d4)	Is this risk factor present or not in your establishment? regardless of whether it is currently under control and regardless of the number of employees it affects: Heat, cold or draught	Q200_6	Yes/No	1 if Q200_6=Yes; 0 otherwise
High noise (d5)	Is this risk factor present or not in your establishment? regardless of whether it is currently under control and regardless of the number of employees it affects: Loud noise	Q200_5	Yes/No	1 if Q200_5=Yes; 0 otherwise
Hard physical conditions = 1 if (d1 + d2 + d3 + d4 +d5) ≥ 1; 0 otherwise
Work at very high speed (d17)	Is this risk factor present or not in your establishment? Time pressure	Q201_1	Yes/No	1 if Q201_1=Yes; 0 otherwise
Long working hours (d19)	Is this risk factor present or not in your establishment? Long or irregular working hours	Q201_5	Yes/No	1 if Q201_5=Yes; 0 otherwise
Work intensity = 1 if (d17 + d19) ≥ 1; 0 otherwise
Help and support from colleagues (r1)	Is this risk factor present or not in your establishment? Poor communication or co-operation within the organisation	Q201_2	Yes/No	1 if Q201_2=No; 0 otherwise
Social support at work = 1 if r1 = 1; 0 otherwise

Notes

← 1. The chapter also makes use of insights of two OECD expert meetings on AI adoption and consultations with researchers, social partners, employers and AI developers during the course of the OECD project on AI in Work, Innovation, Productivity and Skills (AI-WIPS).

← 2. In the OECD questionnaire, questions revolving around trustworthy issues relate to data privacy, data leakages, possible violation of workers’ rights and dignity, as well as discriminations based on biased data (see Annex 7.A).

← 3. In a survey on risks and benefits due to digitalisation carried out with unions representatives in Europe, job destruction (respectively job creation) due to automation were ranked as the most important risk (benefit) of AI in the future of work (with 52% and 45% respectively) (Voss and Riede, 2018[5]). Similarly, two-thirds of works councils, HR councils and supervisory boards surveyed by the ver.di union in Germany a year later feared AI-induced decreases in available jobs while only half of them expected increasing work intensity (ver.di, 2019[6]).

← 4. Possible mechanisms include that the use of AI may reduce stress, fatigue and safety risks through a better work organisation and task optimisation. For example, AI can support or substitute repetitive or physically and mentally strenuous tasks, thereby allowing workers to focus on more interesting and safe tasks. Moreover, AI can also offer opportunities to reduce discrimination in the workplace, or better monitor the well-being and security of workers (Cazes, 2021[42]).

← 5. Workers’ voice is made of the various institutionalised forms of communication between workers and managers that offer an alternative to exit (i.e. dissatisfied employees quitting) in addressing collective problems. Voice is often mediated through representative institutions – “representative workers” voice – such as local trade union representatives, work councils or workers representatives; in addition, voice also materialise in the workplace through the organisation of direct exchanges between workers and managers, via regular townhall meetings or direct consultations – “direct voice”. Finally, in “mixed voice” systems, both direct and representative arrangements of workers’ voice co-exist (see more details in (OECD, 2019[2]).

← 6. In the ESENER-3 survey, 26% of establishments report using AI, of which AI-related management software is more common than AI-related hardware devices. While only 5% establishments report using robots or wearable devices, 12% and 15% of establishments report using software to monitor workers or to determine the content and pace of their work respectively. Moreover, 63% of establishments report having at least one form of worker representation. Worker elected health and safety representatives and committees are the most common form of representative workers’ voice in surveyed establishments with almost 50%, while only 30% of establishments report having a trade union representative or a works council (forms of representations may co-exist in one establishment).

← 7. The ultimate effect of AI mitigated by workers’ voice can be derived from the cumulation of the AI adoption effect and the mitigation effect.

← 8. The dummy variable equals 1 if a worker elected health and safety committee/representative exists in an establishment and it equals 0 if there is no health and safety committee/representative or if it is appointed by the business.

← 9. The OECD Framework for Measuring and Assessing Job Quality takes a multidimensional approach and defines job quality in terms of earnings quality, labour market security and the quality of the working environment (Cazes, Hijzen and Saint-Martin, 2015[83]).

← 10. The presence and degree of co-ordination within and between social partners is important not only to produce independent negotiations, but to ensure inclusiveness across firms and sectors. Co-ordination mechanisms can exist between different levels, for example when sectoral or firm level agreements follow the guidelines fixed by peak-level organisations or by a social pact, or at the same level, for example when sectors or firms follow the standard set in another.

← 11. For instance, workers’ voice can help avoiding the distrust that may be generated by the introduction of new technologies that workers have to work with, through consultation and/or involving them. They can also provide management with better workplace related information including how workers perceive the introduction of new technology and the difficulties they identify.

← 12. The material and position papers developed by unions largely focus on promoting a trustworthy use of AI and training – see for instance the principles of Nordic Engineers to develop technical standards and certifications to increase accountability (ANE/IT University of Copenhagen, 2018[50]; ANE et al., 2021[51]) or UNI.

← 13. On 25 May 2018, the European Union replaced the Data Protection Directive (European Union, 1995), by the General Data Protection Regulation (GDPR) framework (European Union, 2016). The GDPR introduced new rules governing the collection, process, and free flow of personal data regarding data subjects in the European Union. When data originating in EU member states are transferred abroad, the GDPR ensures that personal data protections travel with them. – see Chapter 6 for more details.

← 14. As pointed out by the Hans Böckler Foundation in Germany, the GDPR contains important principles, such as privacy by default and other aspects, which also apply to AI technologies. Article 88 also opens up scope for more specific regulations on data protection by national legislators (Albrecht and Kellermann, 2020[15]) as well as more specific measures by collective agreements in the Member States – especially for those, which ensure the protection of workers’ rights (Klengel and Wenckebach, 2021[84]). Along these lines, the British TUC proposes in its AI Manifesto to enhance the existing British GDPR with a statutory guidance for employers on matters of automatic or AI-influence decision-making (TUC, 2021[61]).

← 15. In this respect, much of the discussion evolves around the proposed AI Act of the European Commission that aims to govern the development and use of AI systems in the EU on a risk-based categorisation approach, with specific safeguards for high-risk uses (see Chapter 6). The proposed EU AI Act, similarly to the GDPR, has raised concerns of some European unions about its articulation with existing collective bargaining regulation and its capacity to adequately address workplace issues in targeting more at consumers rights (TCO, 2021[85]) (Klengel and Wenckebach, 2021[84]).

← 16. Notably to the Article 88 of the EU GDPR which refers to the possibility to lay down in collective agreements more specific rules to ensure the protection of the rights and freedom with regards to the processing of employees’ personal data in the context of employment relationships.

← 17. According to social partners attending the OECD expert workshops, avenues for such investigation could for example include analysing large amounts of wage statements to ensure workers’ correct remuneration or to evidence safety concerns with data on occupational health and safety aspects across workplaces and sectors.

← 18. More information on this tool can be accessed through the website https://www.weclock.it.

← 19. Similarly, the recent agreement between the General Staff Council of the city Stuttgart and the city as a public employer stipulates that the works council may use external consulting services at the city’s expense (Forum Soziale Technikgestaltung, 2022[72]).

╳

Metadata, Legal and Rights

This document, as well as any data and map included herein, are without prejudice to the status of or sovereignty over any territory, to the delimitation of international frontiers and boundaries and to the name of any territory, city or area. Extracts from publications may be subject to additional disclaimers, which are set out in the complete version of the publication, available at the link provided.

https://doi.org/10.1787/08785bba-en

The use of this work, whether digital or print, is governed by the Terms and Conditions to be found at https://www.oecd.org/termsandconditions.