AI, Digitalisation and the Transformation of Jobs: Ensuring Decent Work

The ILO's AI exposure index shows that while a significant proportion of the global workforce faces some degree of AI exposure, the share truly vulnerable to automation is much smaller. This is because most exposed jobs contain a diverse enough mix of tasks that not all can be automated. The distinction between exposure and actual automation risk is therefore critical.

Mr. Verick cited the ILO's own AI exposure index, which indicates that around one in four workers globally are exposed to AI, but only 3.3% of global employment are most exposed to automation because those jobs have less diversity in their tasks, leaving them more vulnerable . He emphasised that exposure does not equate to job loss .

AI-driven technological change is generating new employment opportunities beyond the high-profile roles in Silicon Valley, extending into supply chain work such as data annotation in developing countries. However, these new jobs come with significant decent work challenges. The picture is therefore more nuanced than simple narratives of job destruction suggest.

Mr. Verick noted that AI is creating jobs not only for AI engineers and machine learning engineers but also for those in the supply chain, including data annotators often based in the Global South . He acknowledged that while these roles create new opportunities, they also raise challenges in terms of decent work .

The dominant public debate focuses on whether AI will destroy jobs, but the more fundamental concern from a decent work perspective is how AI changes the nature and quality of existing jobs. Automated systems used for recruitment, performance monitoring, and work assignment are reshaping working conditions in ways that may be harmful. This dimension of AI's impact is often overlooked in headline analyses.

Mr. Verick highlighted that the impact of AI in the workplace is visible through the use of AI in automated systems for recruitment, monitoring tasks, and assigning work, which has implications for job quality, working conditions, and opportunities . He stressed that job quality is more fundamental than job quantity from a decent work perspective .

AI's ultimate impact on employment depends on whether it translates into higher productivity, which in turn lowers prices and stimulates new demand, potentially creating new jobs across the economy. Historical precedent from computerisation suggests these productivity effects can take a long time to appear. Whether AI will accelerate this process remains an open question.

Mr. Verick drew a parallel with computerisation, noting that it took a long time before productivity effects became visible, and questioned whether AI would produce these effects faster . He outlined the chain of effects: higher productivity leading to lower prices, higher demand, and ultimately new job creation .

The ILO's approach to skills development is grounded in evidence about actual labour market demands rather than prescribing specific roles such as AI ethics officers. It works within existing skills policy frameworks and governance systems to ensure they are adapted and modernised. This longer-term approach aims to ensure that training leads to real employment outcomes.

Mr. Verick explained that the ILO works with governments, trade unions, and employer organisations, focusing on what the evidence says about labour market demands rather than prescribing specific job titles . He noted that the ILO works within existing skills policy frameworks to ensure they are adapted and made more flexible to respond to emerging needs .

The ILO has established a dedicated Observatory on AI and Work in the Digital Economy to shed light on the broader implications of AI for the world of work, going beyond unemployment statistics. This research covers job quality, workers' rights, and the situation of small businesses. It serves as a key resource for policymakers and practitioners.

Mr. Verick encouraged participants to consult the ILO's Observatory on AI and Work in the Digital Economy for the latest research, including the analysis behind the exposure index figures he cited .

The ILO engages in capacity building with member states and constituents through a range of partnerships, including the DISCO initiative on digital skills and the International Training Centre based in Turin. These partnerships form a key pillar of the ILO's response to the challenges posed by AI and digitalisation. They complement the ILO's research and policy work.

Mr. Verick mentioned the ILO's capacity-building work with member states through the International Training Centre in Turin , and specifically flagged the DISCO initiative on digital skills as a notable partnership .

Rather than discarding existing policy frameworks, the ILO's approach is to first assess how current international labour standards and national regulations apply to AI-related challenges. The second step is to identify where genuine gaps exist and address them through new or updated frameworks. Social dialogue involving workers and employers is central to this process.

Mr. Verick stressed that existing policies on skills, employment, social protection, and occupational safety and health remain relevant and should not be discarded . He noted that the real question is identifying what is missing and addressing gaps through social dialogue and the involvement of workers, employers, and different ministries .

While existing standards provide a foundation, AI introduces novel challenges around algorithmic transparency, accountability, bias, and data privacy that current frameworks do not fully address. New policy frameworks are needed to fill these gaps. Developing these frameworks must involve workers and employers to ensure they are fair and effective.

Mr. Verick identified specific policy gaps including data privacy protection, algorithmic transparency, accountability, and bias and discrimination as new challenges for policymakers . He emphasised the need for social dialogue and the involvement of workers, employers, and ministries of both labour and technology in developing responses .

Young workers typically enter the labour market through low-level or entry-level tasks that provide on-the-job training and a pathway to more senior roles. If AI automates these tasks, young workers lose not just jobs but the developmental opportunities that enable career progression. Policymakers and technology developers need to consider how to preserve these entry points.

Ms. Uma Rani Amara argued that young workers entering the labour market do low-level tasks that provide training and help them build expertise to become senior professionals, and that automating these tasks removes the entry-level directions for many workers .

Women are disproportionately represented in sectors vulnerable to AI-driven job displacement, creating a risk of significant employment losses. Compounding this, AI hiring tools used by companies and institutions, including UN bodies, are known to be biased, meaning women also face discrimination in accessing new opportunities. This double disadvantage requires targeted policy attention.

Ms. Uma Rani Amara noted that women workers in certain sectors will be affected by job losses, and that many companies and institutions, including the UN, use AI hiring tools that are known to be biased, meaning women not only risk losing jobs but also face discrimination in hiring .

The impact of AI on informal economy workers is not primarily about job destruction but about the erosion of professional autonomy and independent judgement. Workers who previously exercised discretion in how they performed their tasks may find themselves increasingly directed by software systems. This raises broader questions about cognitive capacity and the nature of skilled work.

Ms. Uma Rani Amara used the example of a plumber who may not lose their job to AI but may be guided by software on how to perform tasks, thereby losing autonomy and discretion and becoming increasingly attached to a software product . She raised the broader question of where workers' thinking processes and cognitive abilities go in this context .

In the automobile sector, technologies such as ID badges and CCTV systems are used systematically to monitor workers and collect data on work processes in real time. While industry justifies this as a means of reducing errors and improving efficiency, it also enables the identification of individual workers responsible for faults and can lead to punitive consequences. The level of workplace surveillance has increased significantly as a result.

Ms. Uma Rani Amara described how ID badges and CCTVs are used in the automobile industry to monitor workers and collect data, with the stated aim of reducing errors and improving efficiency . She noted that this enables identification of which worker caused a fault and can lead to repercussions such as warnings or dismissal, and that data visualisation tools displaying performance metrics create stress .

Digital tools in healthcare, such as electronic health records and bed management systems, offer genuine benefits for hospital administration and patient experience. However, the same tools intensify the workload of healthcare professionals by enabling constant on-call availability and increasing the pace of work. Both nurses and doctors are affected, not just lower-skilled workers.

Ms. Uma Rani Amara described how electronic health records and workflow management systems allow hospitals to optimise patient flow and bed allocation, benefiting hospitals and patients . However, she noted that doctors and nurses face increased intensity of work and stress as a result, being on call on a regular basis , and that nurses clearly experience increased workload while doctors are also affected .

General-purpose communication tools are increasingly used in healthcare settings to share information among medical teams, improving transparency and communication channels. However, these same tools create significant data privacy risks because sensitive patient and clinical information is being shared through platforms not designed for confidential medical data. Performance dashboards built on such data also raise concerns about how doctors are assessed.

Ms. Uma Rani Amara noted that WhatsApp groups and Microsoft Notes are used in hospitals to record and transmit discussions among doctors, which improves communication but leads to data privacy issues because sensitive information is being shared with multiple people through inappropriate channels . She also highlighted that performance dashboards assessing doctors' decisions across multiple patients are being developed using such data .

Across industries, there is a fine and increasingly indistinct line between what constitutes automation and what constitutes AI, as data collected by various tools is automatically processed by algorithms. This blurring makes it difficult to categorise and regulate these technologies distinctly. Understanding this continuum is essential for developing appropriate policy responses.

Ms. Uma Rani Amara explained that there is a very fine line between automation and AI today, because data collection tools feed directly into algorithms that automatically process the data to ensure quality, predict outcomes, or perform other functions . She illustrated this with examples from the logistics sector, where warehouse management systems can be automated and algorithmically managed depending on the level of technology .

Behind the impressive capabilities of AI systems lies a vast and largely invisible workforce performing essential tasks such as labelling data, annotating images for autonomous vehicles, and moderating content for large language models. These workers are predominantly based in developing countries and their labour is fundamental to making AI systems function. Their existence and working conditions are rarely acknowledged in mainstream AI discourse.

Ms. Uma Rani Amara stated that millions of humans are invisibly performing tasks of labelling, categorising, and annotating data for autonomous cars and large language models . She noted that her colleague Mr. Verick had briefly mentioned this, and she elaborated that these workers are engaged at every step of the digital supply chain of building an AI .

The term 'artificial intelligence' obscures the reality that human labour is embedded at every stage of AI development, from data cleaning and annotation to content moderation. Ms. Uma Rani Amara proposed the alternative framing of 'human-in-the-loop intelligence' to make this reality visible. This reframing has implications for how AI is regulated and how the workers involved are protected.

Ms. Uma Rani Amara explicitly proposed that AI should be called 'human-in-the-loop intelligence' because there are humans involved in every step of the digital supply chain of building an AI . She also noted that even today, humans continue to moderate content for systems like ChatGPT to correct hallucinations and ensure quality outputs .

Existing policy models from other sectors, such as fair trade standards in agriculture and conflict mineral disclosure requirements in electronics, could be adapted to address the labour conditions of workers in AI supply chains. Such frameworks would require companies to disclose where and by whom their AI systems were trained and under what conditions. This would bring transparency to an otherwise opaque part of the AI industry.

Ms. Uma Rani Amara proposed considering fair trade labour standards similar to those in agriculture, and conflict mineral disclosure requirements similar to those in the electronics sector, applied to AI supply chains, requiring disclosure of which workers in which countries trained specific parts of an AI system and under what working conditions . She also referenced a California legislative bill that moves in this direction .

Drawing on the model of regulatory bodies such as food and drug administrations and transport regulators, a data provenance certification system could be developed for AI. Such a system would ensure that workers training AI systems have appropriate qualifications and that the data used meets quality standards. This would simultaneously protect workers' conditions and reduce liability risks for end-users of AI systems.

Ms. Uma Rani Amara suggested that, analogous to transport regulators and food and drug administrators, a data provenance certification could be developed to ensure proper training for workers who train AI systems and to verify their qualifications, thereby reducing liability for users while ensuring decent working conditions .

The ILO's Convention C193 on decent work in the platform economy addresses the working conditions of workers engaged on micro-task platforms, which includes many of those performing the invisible labour underpinning AI systems. This is regarded as a meaningful step forward in extending decent work protections to this previously unregulated group. However, it addresses only one part of the broader set of issues facing platform and AI supply chain workers.

Ms. Uma Rani Amara noted that the recent ILO Convention on decent work in the platform economy handles the decent work issues of workers engaged on micro-task platforms, describing it as a really good step forward .

A review of 75 employment policy documents from around the world found that many governments are beginning to use digital tools and AI in their employment policies, though this is still at an early stage. This digitalisation touches multiple areas including public employment services, skills delivery, labour market information, and e-formalization. The transition has already begun in many countries, even those with low internet penetration.

Mr. Chacaltana referenced the ILO's publication on the Digital Transformation of Employment Policies, which reviewed 75 policy documents globally and found that many governments are increasingly using digital tools for employment policies, covering areas such as modernising public employment services, improving labour market information, and supporting e-formalization . He noted that some governments began this integration in the early 2000s when internet penetration was below 10% .

The conventional response to technological change places the burden of adaptation on individuals, particularly young people, who are told to prepare themselves for an inevitable future. Mr. Chacaltana argued that this framing is inadequate and that the future of work must instead be actively shaped through policy. This requires a fundamental shift in how governments and institutions approach the challenge.

Mr. Chacaltana argued that the typical approach of telling young people to prepare themselves for the future places the burden on their shoulders and treats the future as fixed and untouchable . He stressed that the future of work should not be something we simply adapt to, but something we shape through policies .

The digital transformation of employment touches on areas spanning labour, education, and technology policy, making coordination across ministries essential. A siloed approach within any single ministry is insufficient to address the breadth of the challenge. Mr. Chacaltana described this as a whole-of-government approach that will unfold over coming years and decades.

Mr. Chacaltana described the need for coordination across labour, education, and technology ministries as part of a whole-of-government approach to digital transformation of employment policy . He also noted that bringing different parts of government together to discuss these issues is a key goal of the ILO, and that this is not always happening in a way that leads to comprehensive approaches .

On the positive side, digital tools enable governments to simplify administrative processes such as business registration and to strengthen labour inspection. However, there is a risk that governments use these tools primarily as instruments of enforcement and surveillance rather than to empower workers and businesses. The example of Estonia's digital identity system illustrated how extensive government data collection can feel intrusive.

Mr. Chacaltana described the positive use of digital tools to simplify business registration and strengthen labour inspection, noting that many countries accelerated this transition during the pandemic . He then described a visit from an Estonian delegation where a colleague demonstrated how much personal data the government held on him through a digital ID card, which Mr. Chacaltana found alarming, illustrating the enforcement risk .

As governments introduce digital identity systems, AI tools, and data-sharing mechanisms, workers and employers must be part of the conversation to ensure these tools are introduced fairly. Social dialogue is the ILO's preferred mechanism for achieving this inclusive approach. Without it, there is a risk that these powerful tools are used in ways that disadvantage workers.

Mr. Chacaltana stated that when governments introduce AI, digital identity, and data-sharing tools, workers and employers need to be part of the conversation, and that social dialogue is the way out of the risks associated with enforcement-focused use of these tools .

While digital and AI tools can support the process of formalising informal economy workers and businesses, they cannot replace the traditional drivers of formalization such as productivity growth, appropriate incentives, and strong institutions. Mr. Chacaltana used the analogy of AI improving a doctor's work but not replacing the need to be a doctor in the first place. Digital tools must be integrated into a broader, multi-faceted formalization strategy.

Mr. Chacaltana used the analogy that AI can improve the work of a doctor, but you still have to be a doctor, to illustrate that digital tools enhance but do not replace the expertise and institutional foundations needed for formalization . He stressed that traditional drivers of formalization - productivity, incentives, and strong institutions - cannot be forgotten and that digital tools alone will not reduce informality .

The ITU Academy is a significant platform for ICT professional development, with a large user base predominantly drawn from developing countries. Over the past five years, interest in AI among its users has grown threefold, reflecting the increasing centrality of AI to the work of ICT professionals globally. This data provides concrete evidence of the growing demand for AI-related skills.

Ms. Kumar stated that the ITU Academy caters to over 115,000 users, of whom 90% are from developing countries . She noted that in the last five years, interest in AI among users has grown threefold , and that AI has bled into every topic within the Academy's course offerings, from cybersecurity to blockchain .

For lifelong learning to be genuinely valuable, it must be designed around the needs of the learner rather than the convenience of the provider. This requires grounding in adult learning principles and solid instructional design, as well as interactive elements such as peer learning and case-based approaches. The ITU Academy uses data on user demographics and interests to anticipate and respond to learner needs.

Ms. Kumar argued that lifelong learning is most valuable when it aligns to the needs of the user and must be human-centred, grounded in adult learning principles and solid instructional design . She described how the ITU Academy collects data on user demographics and topics of interest to anticipate user needs and ensure courses are engaging, incorporating peer learning and case-based learning .

Data from the ITU Academy reveals that among new users signing up, women's interest in AI has surpassed that of men, which runs counter to common assumptions about gender and technology. This finding challenges narratives that position women as less engaged with or interested in advanced technologies. It suggests that given access and opportunity, women are actively seeking AI-related skills.

Ms. Kumar shared that for new users signing up to the ITU Academy, interest in AI among women has surpassed that of men, noting this is not something that is heard about very often given common assumptions about gender and technology .

Alongside the ITU Academy's focus on ICT professionals, the ITU's Digital Transformation Centres project targets underserved and hard-to-reach communities with basic and intermediate digital skills. This two-pronged approach ensures that capacity development reaches both professional and marginalised populations. The project recently reached the milestone of 700,000 people served.

Ms. Kumar described the Digital Transformation Centres project as targeting underserved communities and those who are hard to reach, covering basic and intermediate digital skills including communication skills and working with e-services . She announced that as of the previous week, the programme had served 700,000 people .

In response to the growing need for AI governance expertise, particularly in the Global South, the ITU has launched a free, self-paced AI governance course available in multiple languages. This is complemented by face-to-face funded courses delivered in different countries. However, Ms. Kumar acknowledged that access, motivation, and funding remain barriers to uptake.

Ms. Kumar described the launch of a self-paced AI governance course translated into multiple languages and available free of cost . She also noted that face-to-face funded AI governance courses have been held in locations including Nairobi and Bosnia , and acknowledged that the challenge is not just availability but who takes the courses and who has the capacity, funding, and motivation to do so .

The ITU is actively developing partnerships, including with UNESCO, to expand the reach and quality of AI governance learning opportunities around the world. These partnerships are necessary because the barriers to accessing and completing such courses go beyond mere availability and include motivation, funding, and institutional support. The ITU's approach combines e-learning, face-to-face courses, and partnerships to address these barriers.

Ms. Kumar noted that ITU colleagues were inaugurating a partnership with UNESCO at the AI for Good event to foster AI governance learning in the future . She described the ITU's broader approach of combining self-paced e-learning, face-to-face courses, and partnerships to expand access, while acknowledging that barriers of access, motivation, and funding remain .

Summarising the session's key messages, Ms. Prieto Berhouet emphasised that a human-centred approach was the consistent theme across all contributions, spanning policy, skills, and the nature of work itself. She stressed that the goal must be to shape a future of work that is supported by new technologies rather than simply following wherever technological change leads. The concept of 'humans in the loop', raised by Ms. Uma Rani Amara, was highlighted as a key principle.

Ms. Prieto Berhouet summarised the session's takeaways, noting that the human-centred approach was repeated over and over across discussions of policy and skills . She highlighted the need to shape a future of work supported by new technologies rather than passively following technological change , and referenced Ms. Uma Rani Amara's concept of 'humans in the loop' as something that must be considered .

An audience member raised the concern that while AI is increasingly being discussed and invested in across Africa, there is no corresponding infrastructure or academic preparation for the roles needed to oversee and govern AI systems. This creates a significant gap between the pace of AI adoption and the availability of qualified professionals to ensure human oversight and ethical governance. The audience member questioned whether the ITU Academy or ILO, or partnerships with major AI companies, are addressing this gap.

The audience member noted that they had just come from a session where it was stated that in Africa, none of the infrastructure for AI ethics officers or AI governance roles exists, despite a lot of innovation, AI buzz, and investment taking place . They observed that on the academic side, there is no infrastructure or investment to get people ready for the necessary jobs to oversee AI and ensure human-in-the-loop governance . They asked what the ITU Academy and ILO are doing about this, and whether there are partnerships with big AI companies to roll this out .

A second audience member raised the question of whether institutions such as the ILO would be open to considering digital or alternative pathways for job creation beyond conventional methods. The question implied that traditional approaches to recruitment and employment generation may be insufficient in the context of digital transformation. The audience member suggested that digital tools could offer complementary mechanisms for creating employment opportunities.

The audience member asked whether the ILO and similar institutions would be open to considering, alongside traditional ways of recruiting and creating jobs, a digital way to provide more job creation, suggesting there could be other ways of creating employment beyond existing approaches .