The study, published in Discover Artificial Intelligence, tackles the task of linking mentions in job advertisements to structured reference inventories, specifically the European Skills, Competences, Qualifications and Occupations taxonomy known as ESCO and the European Qualifications Framework, or EQF. Most existing systems treat this as a single retrieval problem: encode a piece of vacancy text, compare it against a reference inventory, and return the nearest node. The author of the new work, Yi He, demonstrates that this uniform treatment is fundamentally mismatched to the material, because the three constructs a posting contains differ in how their evidence is distributed across the text.

The core insight is deceptively simple. A skill is typically fixed by a short phrase, so piling on surrounding context dilutes the signal. An occupation is anchored by a title or a whole sentence, and tolerates context without much gain or loss. A qualification, meanwhile, is not identified by any particular string at all but by an ordered level on an eight-rung scale, where the correct answer is frequently just one step away from the prediction. The paper backs this up with a controlled experiment: widening the query from a detected span to the whole advertisement costs skills 0.343 in rank-one accuracy, roughly seven tenths of their best score, while occupations move by only 0.068 across the entire range. A single fixed query width, as every prior pipeline commits to, is therefore optimal for at most one construct and compromised for the rest.

From this diagnosis the study derives CARL, a Construct-Aware Retrieval and Linking layer whose single design principle is that every linking decision should be conditioned on the construct being linked. Crucially, CARL is a layer rather than a new model: it sits between an off-the-shelf recognizer and an off-the-shelf sentence encoder, changes no backbone, adds no training loop and introduces no trainable parameters into the encoder. Its only free quantities, a handful of fusion weights, a pooling depth, a decay factor and abstention thresholds, are fitted by grid search on a held-out validation split, which means the approach can be dropped into any pipeline of the standard shape.

Three modules realise the principle, one for each axis along which constructs differ. Granularity Fusion builds several queries from the same mention, spanning the mention alone, a clause window, the enclosing sentence, the whole posting and the job title, then combines their similarity scores with weights fitted per construct. Boundary Arbitration addresses a subtler failure: recognition and linking normally run strictly in sequence, so a mention boundary chosen by the tagger can never be revised even when the inventory makes clear it is wrong. The module enumerates a small set of boundary variants of the detected span and lets the retrieval score choose among them, the one point at which linking feeds information back into recognition. Because the original span is always among the candidates, the modification is provably safe: it either keeps the tagger’s proposal or replaces it with one the inventory supports better.

The third module, Ordinal Aggregation, is where the largest gain appears. Qualifications in the EQF are represented by 814 national credential strings, but the true label is one of eight ordered levels, and many strings share a level. Flat nearest-neighbour search over the strings lets a single noisy entry decide a level and makes an adjacent-level near miss compete with the correct answer instead of supporting it. The module instead pools evidence across all reference entries sharing a level and lets neighbouring levels reinforce one another through an ordinal kernel, converting an 814-way decision over noisy strings into an eight-way decision over aggregated evidence. It also decides whether to link at all, abstaining when the top score falls below a threshold fitted against the criterion the system is actually scored on, rather than a hand-set constant inherited from earlier work.

Evaluated against the strongest fixed pipeline the literature offers, CARL raises rank-one accuracy by 0.081 on qualifications and 0.040 on skills, both significant under paired testing with correction for multiple comparisons, and by 0.033 on occupations, which the author reports as directional rather than significant, exactly as the granularity analysis predicts. A leave-one-out ablation shows each module contributing precisely where its motivating measurement said it should, and the fitted fusion weights independently reproduce the measured granularity profile of each construct, evidence that the layer exploits genuine structure rather than fitting noise. Rank-tolerant metrics add a telling detail: under the baselines, qualification mapping placed the reference level in the top five for three quarters of mentions while reaching only 0.27 at rank one, meaning the retriever was finding roughly the right region of the label space and failing to order it, which is exactly what the ordinal module repairs.

The study is equally candid about what remains unsolved. An error budget separating recognition error, retrieval error and the ceiling imposed by taxonomy fit shows that CARL recovers only 7 to 14 percent of the addressable error. A human expert benchmark, in which two domain experts solved the same mapping task on identical samples, shows that people still lead the best system by 0.15 to 0.23, while also confirming that part of the residual error is intrinsic to ambiguous wording and imperfect fit between the vacancies and the inventories. Nearly three reference occupation labels in ten turned out to be nearest-available approximations rather than exact matches, a ceiling no matcher can remove.

Perhaps the most striking negative result concerns large language models. The author ran five decoder models three times each on fixed data and prompts and found that none approached the supervised recognizer, with the best trailing it by 0.21 in span-level F1. More damaging still, between 43 and 61 percent of directly generated occupation labels were not valid inventory entries at all. The lesson echoes a broader theme in modern AI evaluation: when the required output is a formal artifact checked against an external reference rather than fluent prose, the reliable remedy is to make the constraint structural, by committing to the node set as the only admissible answer, rather than policing free generation afterwards. That is why CARL keeps retrieval, not generation, as the mechanism that produces a label.

The work also carries honest caveats. The occupation evaluation rests on 542 English-language vacancies from an Ethiopian online platform, 83.6 percent of them white-collar, so the results describe formal, urban recruitment rather than a whole labour market. Qualification annotations show only moderate inter-annotator agreement, and the author reports sensitivity checks confirming the conclusions are robust to how disagreements were settled. The fitted quantities should be refitted before transfer to another market, inventory or language. Even so, the central message stands and is likely to shape how the field builds its next generation of pipelines: skills, occupations and qualifications are different kinds of evidence, and a system that treats them identically will be optimal for none of them. With job advertisements remaining the largest continuously refreshed record of what employers actually demand, getting this normalisation right matters far beyond the machine learning literature.

Subject of Research: Construct-aware retrieval and linking of skills, occupations and qualifications in job advertisements to the ESCO and EQF taxonomies

Article Title: Matching granularity matters for occupation skill and qualification linking in job advertisements

Article References: He, Y. (2026). Matching granularity matters for occupation skill and qualification linking in job advertisements. Discover Artificial Intelligence, 6(1), Article 1284. https://doi.org/10.1007/s44163-026-02232-w

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02232-w

Keywords: job advertisements, skill extraction, entity linking, ESCO, European Qualifications Framework, retrieval, taxonomy matching, natural language processing, machine learning, labour market, ordinal classification, large language models

The strength of the dissertation lies in its methodology. Rather than relying on speculation or surveys of opinion, Musazade grounded his analysis in two complementary types of evidence. The first is a longitudinal examination of job advertisements spanning several years, a technique that allows researchers to trace how the language of skill requirements evolves in real time as employers respond to technological change. The second is a set of experiments investigating how people actually use AI in practice, moving beyond what organisations say they want to observe what happens when human workers and machine systems interact on genuine tasks. Together, these approaches provide a rare dual perspective on the AI transition: one looking at the demand side of the labour market, the other at the behavioural reality of human-AI collaboration.

One of the dissertation’s central findings concerns the role that generative AI currently plays in working life. Despite the breathless rhetoric that often surrounds these systems, Musazade’s research shows that generative AI functions primarily as a support tool rather than a replacement for human workers. Even when AI is embedded in workflows, human oversight, analytical thinking and problem-solving skills remain essential. This finding aligns with a growing understanding in the information systems field that large language models and similar technologies are powerful pattern completers and content generators, but they lack the contextual judgement, accountability and domain reasoning that professional work demands. The machine can draft, summarise and suggest; the human must verify, interpret and decide.

Perhaps the most practically significant result comes from an exploratory experiment comparing different styles of AI use. Participants who treated AI as a collaborative tool, combining its input with their own judgement, performed better than those who relied more heavily on AI-generated responses. The implication is subtle but profound: the value of AI is not fixed by the quality of the model alone, but by the mode of engagement that the human user brings to the interaction. Passive acceptance of machine output can flatten performance, while active, critical integration of AI suggestions with independent reasoning amplifies it. In other words, AI alone does not guarantee better outcomes, and the difference between augmentation and substitution may be the difference between success and failure.

“The future of work is not about humans or AI. It is about how effectively people can collaborate with AI. Those who can combine the capabilities of technology with their own judgement will have a clear advantage,” says Musazade. This framing reframes the popular debate in an important way. The binary question of whether machines will replace people has dominated public discourse since the earliest automation scares, yet the empirical picture emerging from studies like this one is far more nuanced. The competitive frontier is not human versus machine but skilled collaborators versus unskilled ones, and the skills in question are deeply human: the capacity to evaluate evidence, to sense when a model’s confident answer is wrong, and to weave technological output into organisational and ethical context.

The dissertation also documents a rapid evolution in what employers actually ask for. Skill requirements in data- and AI-related fields have shifted noticeably in a short period, and the direction of that shift is telling. The focus has moved from developing AI systems to applying them in practice. A few years ago, the scarce and celebrated expertise centred on building models, training algorithms and engineering the underlying infrastructure. Today, employers are increasingly seeking professionals who can use AI tools effectively, integrate them with other systems and adapt them to organisational needs. The technology has matured from an experimental artefact into a general-purpose instrument, and the labour market now rewards those who can deploy it well rather than merely those who can construct it.

Notably, this shift has not rendered traditional technical skills obsolete. The research finds that more conventional competencies, such as programming in Python, remain highly valued by employers. This coexistence of old and new skill demands paints a more complicated picture than the simple narrative of disruption. Organisations still need people who understand how the machinery works beneath the interface, both to build custom solutions and to exercise informed oversight of what the tools produce. The ideal profile emerging from the job advertisement analysis is therefore hybrid: fluent in the practical use of AI applications, capable of integrating them into broader technical architectures, and grounded in the foundational programming and analytical skills that make such fluency trustworthy rather than superficial.

From a theoretical standpoint, the dissertation’s subtitle, A Human-Task-Technology Perspective, signals its intellectual anchoring. Information systems research has long recognised that technology does not act upon organisations in isolation; outcomes emerge from the interplay between people, the tasks they perform and the tools they use. By applying this lens to the AI transition, Musazade avoids the technological determinism that plagues much commentary on the subject. The same AI system can substitute for a worker on one task and augment another, depending on how the task is structured and how the human engages with it. Skill requirements, seen through this framework, are not dictated by the technology itself but negotiated at the intersection of what machines can do, what tasks demand and what organisations need.

The practical implications extend well beyond individual career advice. Musazade argues that the findings can help educational institutions and employers respond to changing skill requirements during a period of rapid technological change. For universities and professional training providers, the message is that curricula must balance foundational technical knowledge with instruction in critical evaluation, systems thinking and effective human-AI collaboration. For employers, the research suggests that hiring strategies and workplace design should reward judgement-rich engagement with AI rather than raw reliance on it, and that training programmes should cultivate the collaborative habits that the experiments show lead to better performance. Policy makers, too, face a labour market in which the half-life of specific technical skills appears to be shortening even as the demand for durable human capabilities persists.

What makes this research resonate beyond academia is its timing and its tone. Arriving at a moment when generative AI has become a household presence and anxiety about technological unemployment runs high, the dissertation offers neither complacency nor alarm. It confirms that the labour market is changing quickly, that the composition of valued skills is shifting from building AI to applying it, and that the ability to work alongside intelligent systems is becoming a defining professional competency. At the same time, it demonstrates with empirical rigour that human creativity, analytical thinking and systems thinking are not casualties of that transition but its essential partners. The machines are getting better, but the evidence suggests that the people who thrive will be those who never stop thinking for themselves. Musazade defended his dissertation at Åbo Akademi University on 25 September 2026, and its findings arrive as a timely reminder that in the story of AI and work, the human role is being rewritten rather than erased.

Subject of Research: How artificial intelligence is transforming skill requirements in data analytics and AI-related occupations

Article Title: AI is reshaping the workplace, but not replacing human judgement

Article References: AI is reshaping the workplace, but not replacing human judgement. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: artificial intelligence, future of work, generative AI, skill requirements, labour market, human-AI collaboration, information systems, data analytics, analytical thinking, job advertisements, doctoral dissertation, workplace automation