AI adoption in Central and Eastern Europe chiefly benefits senior, high-skill workers and firms in ICT and manufacturing, while junior employees face heightened displacement risk; uneven diffusion and weak trust and reskilling systems are widening wage and regional divides.

Main Finding

AI adoption in Hungary is at an early, uneven stage: senior professionals and select industries already capture measurable productivity gains (10–30%), while junior roles face displacement risks (“junior paradox”). Academic discourse shows a stable mild negativity bias about AI’s labour-market effects. Adoption is concentrated geographically (Budapest) and sectorally (IT, consulting, finance, automotive), and organisational rollout is strongly linked to perceived transparency and presence of evaluation tools.

Key Points

• Literature bias: A scientometric analysis of 250 papers reveals a statistically significant tilt toward cautious pessimism about AI’s labour-market impacts; five stable topical clusters persist since 2015 (job dynamics/wages, productivity, skill change, algorithmic management, robotics).
• Junior paradox: Interviews with 20 Central‑European software professionals report routine 10–30% efficiency gains for seniors using AI for boilerplate work, while entry-level tasks—needed for skill accumulation—are disproportionately eroded.
• Early, concentrated diffusion: From 1,382,827 Hungarian job adverts (2019–2024), 18,496 (1.34%) explicitly or semantically reference AI. AI-related vacancies cluster in Budapest and four sectors; generative‑AI terms (e.g., “ChatGPT”) appear only 128 times.
• Semantic methods matter: A DSPy language-model classifier recovered >3,000 AI-related ads missed by keyword/RegEx rules, showing the importance of semantic approaches for labour‑market surveillance.
• Organisational trust drives scale-up: A “Trust My AI” survey (n = 150) finds weekly AI use common but enterprise roll-out limited; perceived transparency and evaluation tooling (RAG, ARES, SageMaker Clarify, etc.) predict scaling intent. Usage–trust correlation ρ = 0.37 (p < 10−4).
• Policy risk: Without intervention, AI is likely to amplify existing inequalities (geographic, seniority, firm-size) rather than reduce them.
• New scientific theses (author): T1 — literature-sentiment asymmetry; T2 — validated junior paradox; T3 — low market penetration (1.34%); T4 — trust–adoption link.

Data & Methods

• Mixed-methods design:
  • Scientometric meta-analysis: 250 curated papers analyzed with LDA topic modelling, PageRank citation-network analysis and sentiment mapping; identified small-world citation network anchored by canonical works (e.g., Autor 2015; Acemoglu & Restrepo 2020).
  • Qualitative fieldwork: 20 semi-structured interviews with Central‑European software professionals, benchmarked against the 2023 Stack Overflow survey.
  • Large-scale NLP vacancy mining: 1,382,827 Hungarian job adverts (2019–2024) processed via RegEx pipelines and a DSPy language‑model classifier to extract AI-related vacancies and roles.
  • Survey: “Trust My AI” (n = 150) measuring adoption hurdles and confidence in evaluation tooling; statistical associations (usage–trust ρ = 0.37, p < 10−4).
• Key measurements: share of AI-related vacancies (18,496; 1.34%), generative-AI mentions (128), DSPy-added adverts (>3,000), reported senior efficiency gains (10–30%).
• Limitations noted by author: geographic focus (Hungary/Central Europe), early-stage diffusion period, modest interview and survey sample sizes, observational design for many claims.

Implications for AI Economics

• Measurement and surveillance: Keyword-based vacancy monitoring underestimates AI demand. Semantic / LM-based classifiers materially improve detection; large-scale labour-market monitoring should incorporate semantic methods to track adoption and skill signals.
• Tasks framework & labour dynamics: Evidence reinforces a tasks-based outcome where AI complements senior cognitive/judgment tasks but substitutes routine entry-level tasks, raising risks to traditional experiential training ladders and increasing within-occupation heterogeneity.
• Inequality and spatial concentration: Early diffusion centered in capitals and high-skill sectors suggests AI may amplify geographic and sectoral inequality unless policy offsets (regional upskilling, SME compute/data access) are implemented.
• Role of governance and trust: Organisational adoption is not primarily cost-constrained but trust- and governance-constrained. Evaluation frameworks and transparent tooling materially increase scale-up intent, implying that investments in evaluation, auditing and explainability can accelerate beneficial diffusion.
• Policy levers: To steer outcomes toward inclusivity, a mix of measures is suggested — preserving junior pipelines (internships, protected tasks), fiscal incentives for junior hiring, reskilling grants, AI curricula and critical-judgement skills, targeted sectoral pilots, and support for SME compute/data capabilities.
• Research agenda for AI economics:
  • Causal evidence on displacement vs complementarity (RCTs, firm-level panel studies).
  • Wage and career-path impacts for cohorts entering during AI diffusion.
  • Longitudinal monitoring of vacancy composition using semantic classifiers.
  • Evaluation of policy interventions (tax credits, training subsidies, governance standards) on adoption patterns and inequality outcomes.
  • Cross-country comparisons to test generalisability beyond Hungary/Central Europe.

Summary judgement: The booklet documents an early but consequential stage of AI adoption where measurable productivity gains coexist with meaningful distributional risks; economics and policy should prioritise better measurement, governance tools, and interventions that protect pathways for junior skill accumulation.