AI-enhanced robotics patenting took off after 2010, marking a structural shift in innovation dynamics; China shows the tightest integration between AI and robotics with substantial university/public-sector patenting, while the U.S. remains more market-driven and less integrated.

Main Finding

Patenting in core AI, traditional robotics, and robots embedding AI are distinct technological trajectories. AI-enhanced robotics accelerated sharply from the early 2010s, structural breaks cluster after 2010 (consistent with an AI diffusion regime shift), and long-run co-evolution between AI and robots varies systematically by national innovation system — with China showing strong integration (and a sizable university/public-sector contribution), the United States showing a more market-driven and weaker AI–robot integration, and Europe, Japan and South Korea occupying intermediate positions.

Key Points

• Novel disaggregation: the paper separates patenting into three mutually exclusive domains:
  • Core AI (foundational capabilities: learning, perception, reasoning, decision-making, ML).
  • Traditional robots (industrial, service, social robots of conventional automation).
  • AI-enhanced robots (robotic inventions that structurally embed AI functionalities).
• Data period and source: PATSTAT (EPO) patent-family data covering 1980–2019 (Spring 2021 release).
• Identification approach: combination of CPC codes, keyword-based retrieval, and document-level text mining to classify patents into the three domains.
• Time-series econometric approach:
  • Tests for non-stationarity / unit roots of the series.
  • Structural-break detection (breaks concentrated after 2010).
  • Pairwise cointegration analysis to assess long-run relationships between domains and across countries.
• Empirical findings:
  • Core AI: rapid expansion overall but heterogeneous across classes.
  • Traditional robots: steadier, gradual growth trajectory.
  • AI-enhanced robots: marked acceleration beginning early 2010s — consistent with AI diffusion into embodied systems.
  • Country heterogeneity: China exhibits strong cointegration between core AI and AI-enhanced robots and substantial public/university involvement; the U.S. displays weaker integration and more private-sector driven patenting; Europe, Japan, South Korea show mixed/intermediate patterns.

Data & Methods

• Data
  • Source: EPO Worldwide Patent Statistical Database (PATSTAT), Spring 2021.
  • Coverage: patent families 1980–2019, bibliographic fields (CPC classes, titles/abstracts, applicants/inventors, filing/publication dates).
• Domain construction
  • Core AI: patents linked to foundational AI subfields (machine learning, knowledge representation, perception, etc.).
  • Traditional robots: patents matching robotics taxonomies (industrial, service, social robots).
  • AI-enhanced robots: patents flagged where AI functionality is structurally embedded in robot systems.
  • Mutually exclusive classification achieved by combining CPC filters, keyword queries, and document-level text mining.
• Empirical strategy
  • Time-series diagnostics: stationarity/unit-root testing for each series.
  • Structural-break analysis to identify regime shifts (notably post-2010).
  • Pairwise cointegration tests to detect long-run stochastic relationships between domains at aggregate and country levels.
  • Cross-country comparisons focusing on major innovation systems (China, United States, Europe, Japan, South Korea).
• Scope and caveats
  • Patents used as proxy for inventive output — informative for emerging, competition-intensive domains but subject to strategic filing, variation in propensity to patent, and non-patent appropriation modes.
  • Analysis stops at 2019 (pre-2020s developments not covered).

Implications for AI Economics

• Measurement and evaluation
  • Aggregating “robots” masks an important heterogeneity: AI-enhanced robots follow different dynamics than traditional robots. Economic studies relying on aggregate robot adoption should distinguish embodied-AI robots to correctly assess GPT diffusion and complementarities.
• AI as embodied GPT
  • The acceleration of AI-enhanced robotics after 2010 supports the view of AI acting as a general-purpose technology when embodied in sectors (robotics), with implications for cross-sector productivity effects and task-level complementarities.
• Role of innovation systems and policy
  • Institutional arrangements shape the co-evolution of AI and robotics. State-coordinated or public-intensive systems (e.g., China) can generate tighter long-run integration between core AI and AI-embodied applications; market-led systems (e.g., U.S.) may see more fragmented pathways. This suggests policy design (industrial policy, public funding, university engagement, coordination mechanisms) materially affects how AI translates into embodied technologies and downstream economic outcomes.
• Labor and productivity effects
  • Faster diffusion of AI-enhanced robots implies potentially different impacts on tasks, occupations, and sectors compared with traditional automation. The embodied-AI dimension likely raises the scope of complementarities/substitutions across more complex cognitive and perceptual tasks — altering the distributional and productivity consequences usually attributed to “robotization.”
• Research and policy priorities
  • For researchers: integrate disaggregated patent-based measures of embodied AI in empirical studies of adoption, productivity, and labor impacts; extend analysis post-2019 to capture recent large-language-model and robotics developments; link patenting dynamics to adoption and firm-level outcomes.
  • For policymakers: designs that consider how institutional configuration (public funding, university–industry links, standards and regulation) shapes AI’s embodiment in physical capital can steer industrial upgrading, competitiveness, and distributional outcomes.
• Limitations to bear in mind
  • Patenting intensity is an imperfect proxy (strategic patenting, sectoral differences in patent propensity).
  • The analysis captures inventive activity up to 2019; rapid changes since then (foundation models, cloud robotics, new regulation) warrant updated analyses.

Overall, the paper argues that distinguishing AI-enhanced robotics from traditional robotics is crucial for understanding the technological and institutional dynamics of the “AI gold rush,” and that national innovation systems play a central role in shaping how core AI becomes embodied in robotic technologies — with important implications for productivity, labor markets, and policy.