Generative AI shows characteristics of general-purpose technology

According to OECD analysis published in June 2025, generative artificial intelligence demonstrates characteristics that qualify it as a general-purpose technology. The research examines three defining features: pervasiveness across sectors, continuous improvement over time, and innovation spawning capabilities.

Patent data analysis reveals generative AI applications spanning over twenty-one distinct areas from 2000 to 2023. While software accounts for the largest share of generative AI patents at 32,780 applications globally, significant activity emerges in life and medical sciences with 5,994 patents, business solutions with 5,473 patents, and banking and finance with 1,420 patents.

The technology's improvement trajectory follows accelerated growth patterns. According to the research, training compute for AI models has grown 4.4-fold annually between 2010 and May 2024, compared to 1.5-fold yearly growth pre-2010. Frontier models including Google's Gemini Ultra 1.0 and OpenAI's GPT-4 demonstrate approximately 4-times yearly growth in computational scale.

Model capabilities show dramatic advancement across literacy, numeracy and logical reasoning benchmarks. OpenAI's GPT-4 correctly answered 85% of reading test questions and 84% of science questions in the Programme for International Student Assessment, outperforming student averages. The model's mathematics performance improved from 35% to 40% compared to its predecessor GPT-3.5.

Recent models achieve unprecedented performance scores. According to the document, OpenAI's o3 model released in early 2025 shows substantial improvements in mathematics and abstract reasoning benchmarks that previously challenged AI systems. The development of "deep thinking" capabilities enables multi-step processes for complex problem-solving through systematic approach variations.

Citation analysis of generative AI patents from 2010 to 2020 demonstrates innovation spawning across technology fields. More than 80% of generative AI patents received forward citations using a 3-year citation window. The number of technology fields citing generative AI patents consistently exceeds the fields of the original cited patents, suggesting cross-sector applications.

Patent quality metrics reveal increasing technological generality and originality over time. Generative AI patents increasingly rely on diverse earlier innovations while enabling breakthroughs across varied technology fields. Forward citations span 34 technology fields for patents published in 2018, including medical technology, telecommunications, transport, and logistics beyond core computer technologies.

The research identifies positive feedback loops between generative AI and application sectors. Forward citation trends from 2010 to 2022 show increasing numbers of generative AI patents being cited, non-generative AI patents citing earlier generative AI work, and generative AI patents citing non-generative AI innovations that previously referenced earlier generative AI technologies.

Current adoption rates remain limited but growing. Across OECD countries, 8% of firms used AI technologies in 2023, rising to 13.5% in the European Union by 2024. Individual adoption demonstrates faster growth, with ChatGPT reaching over 200 million weekly active users as of August 2024. Survey data indicates approximately 40% of US population aged 18-64 uses generative AI as of late 2024.

Occupational exposure analysis suggests widespread potential impact. According to the research, approximately 80% of US workforce could have at least 10% of work tasks affected by large language models, with 19% experiencing at least 50% task impact. Recent conversational data analysis from December 2024 to January 2025 finds 36% of occupations use generative AI for at least 25% of tasks.

Industry exposure correlates with economic output concentration. US industries most exposed to generative AI based on occupational composition also contribute larger shares to real GDP in 2023. Finance and insurance, professional services, and information sectors demonstrate higher generative AI exposure while accounting for substantial economic activity.

The technology shows characteristics of invention of a method of invention, potentially improving research and development productivity. Applications include novel research idea generation, social science hypothesis testing, data harmonization across sources, and drug discovery optimization. In education, generative AI enables adaptive learning experiences and personalized feedback systems.

However, productivity gains may not materialize immediately. The research notes potential productivity paradox effects similar to earlier general-purpose technologies, where benefits emerge after complementary investments, skill development, and organizational changes. Historical precedents suggest substantial productivity improvements could occur through innovation-spawning channels after transformation periods.

Policy implications emphasize comprehensive approaches across technology diffusion, continuous improvement support, and innovation encouragement. The analysis recommends fostering trustworthy generative AI diffusion, strengthening human capital development, supporting research and development investment, and embedding guiding values throughout innovation processes.

European regulatory developments add compliance complexity. CNIL finalizes recommendations for AI system development and Dutch data protection authority guidance establish detailed requirements for AI systems processing personal data.

For marketing professionals, these developments signal fundamental shifts in tool deployment. Companies using generative AI for content creation or customer service must demonstrate compliance with data protection requirements. The emphasis on data minimization could particularly impact marketing applications relying on broad data collection.

Recent industry developments demonstrate accelerating integration. Zeta Global's AI Agent Studio launched with agentic workflows enabling interconnected marketing automation beyond isolated tasks. Google's AI Mode expansion to UK users reflects systematic global deployment of conversational search capabilities.

Microsoft's free AI certification courses address skill development needs through comprehensive learning paths including 24-lesson curriculum covering neural networks, computer vision, and natural language processing. Professional development metrics show significant engagement with 1,674,807 learners completing Career Essentials in Generative AI courses.

The research suggests generative AI's transformative potential depends on policy implementation supporting widespread diffusion, continuous improvement, and innovation spawning. While productivity benefits may require complementary investments and organizational changes, the technology demonstrates characteristics qualifying it as a general-purpose technology with substantial economic implications.

Timeline

• June 2025: OECD publishes research analyzing generative AI as general-purpose technology
• March 2025: Google launches AI Mode for US users through Search Labs
• February 2025: Microsoft releases free AI certification courses with comprehensive curriculum
• Early 2025: OpenAI releases o3 model showing substantial improvements in mathematics and reasoning
• November 2024: ChatGPT launched in November 2022 reaches fastest-growing consumer application status
• July 2025: CNIL finalizes AI system development recommendations under GDPR
• May 2025: Dutch Data Protection Authority releases GDPR preconditions for generative AI consultation
• March 2025: Zeta Global announces AI Agent Studio general availability with agentic workflows

Summary

Who: OECD researchers Flavio Calvino, Daniel Haerle, and Sarah Liu from the Directorate for Science, Technology and Innovation analyzed generative AI characteristics. The study impacts policymakers, technology companies, and organizations implementing AI systems globally.

What: Research examining whether generative AI qualifies as a general-purpose technology through three characteristics: pervasiveness across sectors, continuous improvement over time, and innovation spawning capabilities. Analysis includes patent data, performance benchmarks, and adoption patterns.

When: Study published June 2025, analyzing data from 2000-2023 for patents, 2010-2024 for computational trends, and recent performance benchmarks through early 2025. Timeline covers technology development from foundational research through current commercial deployment.

Where: Global analysis focusing on OECD member countries with specific examination of United States patent data, European Union adoption statistics, and international deployment patterns. Regulatory developments span European jurisdictions including France and Netherlands.

Why: Understanding generative AI's general-purpose technology characteristics informs policy decisions, investment strategies, and regulatory approaches as the technology achieves widespread adoption. Research addresses questions about productivity impacts, innovation potential, and necessary policy responses for maximizing benefits while addressing risks.

PPC Land explains

Generative AI: Artificial intelligence systems that create novel content including text, images, audio, and video based on patterns learned from training data. Unlike traditional AI that analyzes existing data for predictions or classifications, generative AI produces new material often indistinguishable from human-created content. The technology enables natural language interaction through prompting, making it accessible to non-technical users across various applications from content creation to customer service automation.

General-Purpose Technology: Technologies characterized by pervasiveness across multiple sectors, continuous improvement over time, and ability to spawn innovations in application areas. Historical examples include the steam engine, electricity, and information communication technology. These technologies transform economic processes through widespread adoption and create innovational complementarities that drive aggregate productivity growth and economic development across entire economies.

Patent Citations: References to earlier patents within new patent applications, serving as indicators of technological influence and innovation flow between different fields. Forward citations measure how frequently later patents reference earlier work, providing quantitative evidence of innovation spawning capabilities. The analysis uses citation patterns to demonstrate how generative AI technologies influence subsequent innovations across diverse technological domains beyond their original application areas.

Innovation Spawning: The capacity of a technology to generate new innovations across multiple application sectors through spillover effects and complementary developments. This characteristic distinguishes general-purpose technologies from specialized tools by their ability to create positive feedback loops between the core technology and its applications. Innovation spawning includes both direct applications and indirect effects that stimulate creativity and problem-solving in unrelated fields.

OECD: Organisation for Economic Co-operation and Development, an international organization comprising 38 member countries that produces research and policy recommendations on economic and social issues. The OECD's Directorate for Science, Technology and Innovation conducts analysis on emerging technologies and their economic implications. Their artificial intelligence papers provide authoritative research informing policy decisions across member nations regarding technology adoption and regulation.

Large Language Models: Advanced neural networks trained on vast text datasets to understand and generate human-like language across multiple tasks. These models demonstrate capabilities in reading comprehension, writing, reasoning, and code generation through transformer architecture introduced in 2017. Performance improvements occur through scaling laws where increased computational resources, data, and model parameters lead to enhanced capabilities following predictable patterns.

Productivity: Economic measure of output per unit of input, typically calculated as goods and services produced per hour of labor or capital investment. For general-purpose technologies, productivity gains may experience delayed realization due to complementary investments needed in skills, organizational changes, and supporting infrastructure. The productivity paradox describes situations where technological advances initially show limited measured productivity improvement despite obvious capability enhancements.

Occupational Exposure: The degree to which job roles and tasks can be affected by artificial intelligence capabilities, measured through analysis of work activities susceptible to automation or augmentation. Research indicates 80% of US workforce could have at least 10% of tasks affected by large language models, with exposure varying by skill level, education requirements, and task complexity. Higher exposure often correlates with cognitive rather than manual work activities.

Forward Citations: Patent references that occur when newer patents cite earlier innovations, indicating technological influence and knowledge transfer across time periods. High forward citation counts suggest greater impact and broader applicability of original innovations. The metric helps identify breakthrough technologies and measure innovation spawning by tracking how foundational patents influence subsequent research and development across different technology fields.

Training Compute: Computational resources measured in floating-point operations required to train artificial intelligence models during development phases. Growth in training compute reflects increasing model complexity and capability advancement over time. Since 2012, machine learning training compute has doubled approximately every 3.4 months, significantly exceeding Moore's Law improvements in general computing hardware, demonstrating the technology's rapid scaling trajectory.