A new approach using large language models (LLMs) can identify transformative technologies years before they become mainstream by tracking where separate scientific fields begin to overlap and share knowledge. Researchers from Switzerland's Institute of Entrepreneurship and Management and the Cyber-Defence Campus developed a data-driven pipeline that extracts patterns from nearly 280,000 scientific papers and nearly 10,000 patent applications to forecast technological disruption.

The challenge of predicting transformative technologies has long frustrated strategists and planners. Traditional forecasting methods, like expert panels and historical trend analysis, struggle to keep pace with the rapid innovation cycles in information and communication technologies. Consider how quickly large language models emerged; the term itself appeared in 2019, and within just three years, ChatGPT had become globally recognized. By the time major bibliographic databases caught up with stable terminology, the technology was already reshaping industries.

How Does This New AI Forecasting System Work?

The researchers built a multi-stage pipeline that leverages advances in large language models to extract meaningful relationships between technology concepts from unstructured text. Rather than relying on manually tagged metadata or citation counts alone, the system reads the full text of scientific papers and patents to understand how technologies relate to one another. The approach focuses on technological convergence, a concept introduced decades ago to describe how formerly separate fields begin to share technical knowledge and blur their boundaries.

• Semantic Triple Extraction: The system uses LLMs to identify relationships between technology-related concepts in scientific texts, creating a structured map of how ideas connect across papers and patents.
• Noun Stapling Technique: Researchers developed a novel method called "noun stapling" to group related technology terms that might be expressed differently across documents, improving accuracy when tracking emerging concepts.
• Graph-Based Convergence Detection: The pipeline constructs a large-scale graph of technology entities and relationships, then applies graph-based metrics to detect when distinct fields begin to intersect, signaling potential transformative breakthroughs.
• Temporal Trend Analysis: The system tracks how often technology terms appear together over time, identifying acceleration patterns that suggest emerging convergence rather than coincidental mentions.

What Data Did Researchers Analyze?

The team validated their methodology on two complementary datasets spanning seven years. They analyzed 278,625 arXiv preprints from 2017 to 2024 to capture early scientific signals, and 9,793 United States Patent and Trademark Office (USPTO) patent applications from 2018 to 2024 to track downstream commercial developments. This dual approach allowed researchers to see both where innovation begins in academic research and where it translates into practical applications.

The choice of datasets matters significantly. ArXiv preprints represent cutting-edge research before peer review, offering early signals of emerging ideas. Patents, by contrast, show which technologies companies are actually investing in and protecting. By analyzing both, the researchers could identify convergence patterns that appear in early-stage research and then track whether those patterns eventually lead to commercial development.

Why Does Technological Convergence Signal Disruption?

Transformative technologies rarely emerge in isolation. Instead, they typically arise at the intersection of multiple fields, where researchers and engineers from different domains begin to share tools, knowledge, and approaches. When machine learning researchers started applying neural networks to natural language processing, for example, they were drawing on decades of work in both fields. The convergence of these domains created the foundation for modern large language models.

By monitoring where scientific fields begin to overlap, the new forecasting system can identify these intersection points before they become obvious to the broader market. This gives organizations a crucial advantage: they can begin preparing for disruption years before competitors recognize the threat or opportunity.

How Does This Compare to Traditional Forecasting Methods?

Classical forecasting approaches like the Delphi Method, which relies on expert panels, and S-curve substitution analysis, which models technology adoption over time, have guided strategic planning for decades. However, these methods depend on stable terminology, well-defined application domains, and historical precedent. In fast-moving fields like artificial intelligence, these assumptions break down quickly.

The LLM-based approach sidesteps these limitations by working directly with full-text documents rather than waiting for standardized terminology to emerge. It can detect convergence patterns even when researchers use different terms to describe similar concepts, and it can identify emerging fields before they have established names or clear boundaries. This makes it particularly valuable for monitoring information and communication technologies, where innovation cycles compress faster than traditional forecasting methods can adapt.

The researchers demonstrated that their pipeline can identify both established convergence patterns, which validate the approach against known breakthroughs, and emerging patterns that may signal future transformative technologies. This dual validation suggests the system could become a practical tool for technology scouts, venture investors, and strategic planners seeking to stay ahead of disruption.

As organizations face increasingly rapid technological change, data-driven forecasting systems that can process vast amounts of scientific and patent literature offer a scalable alternative to expert-dependent methods. The ability to automatically monitor convergence patterns across hundreds of thousands of documents could reshape how companies anticipate and prepare for the next wave of transformative innovation.