1 Introduction
Quantum technology (QT) is a field of innovation attracting global attention in recent years. Two initiatives, each in the order of one billion Euro (€1B), were recently enacted by the EU (Riedel et al., 2019) and the USA (Raymer and Monroe, 2019). Similar programs exist, all initiated within the last decade, in Canada (Sussman et al., 2019), Japan (Yamamoto et al., 2019), Australia (Roberson and White, 2019), the UK (Knight and Walmsley, 2019), Russia (Fedorov et al., 2019), and China (Zhang et al., 2019). The combined planned investment through such initiatives worldwide is currently around €20B (Qureca, 2020). One of the main motivations for these initiatives is to support the commercialisation of quantum technologies, an activity that may be referred to as the 'out of the lab, into the market' approach.
These developments may be read as part of the ongoing 'second quantum revolution', a term coined in the early 2000s (Dowling and Milburn, 2003a), which covers (but is not limited to) technologies emanating from quantum information science. The main subfields of QT are generally characterised as quantum communication/cryptography, quantum computation/simulation, and quantum metrology/sensing. Besides the fact that QT has become a focal point for international investment and competition, why is this field of technology interesting for academic research about innovation? One reason is that while QT and quantum science have been heralded as one of the most important products of human society during the 20th century, with immense practical implications for society and the economy (Jaeger, 2018), exactly what constitutes QT is highly disputed and widely misunderstood. Concepts derived from quantum physics – such as superposition (in which a particle may be in two states, or places, simultaneously), quantum entanglement (whereby classically not-possible correlations can be shared between distant locations), quantum tunnelling (whereby a wave may pass through an ostensibly – according to classical mechanics – impenetrable barrier), or the replacement of absolute truth with probabilistic estimates – make QT not only difficult for people not deeply educated in quantum theory to understand, but also rather difficult to define. The intrinsic difficulty of reaching a widely accepted and precise definition of the concept of QT is amplified when seeking to operationally define QT for the purposes of industry analysis, public investment, or the assignment of intellectual property rights. If a national government wishes to allocate a billion Euro of public funds to QT, what are the actual decision parameters that a public official should use when allocating the funds? If a venture capitalist wishes to invest in QT, what criteria should be used to select appropriate ventures or projects for the firm's portfolio? In an attempt to address these intellectual-cum-practical challenges, we propose in this paper the use of a hybrid human-intelligence/artificial intelligence (AI) approach to 'naturally' defining the domain of QT based on the emergent way that the field has been treated by the world's patent offices in response to inventors in the field seeking exclusive rights for their inventions, i.e., for patents for the technological embodiments of quantum phenomena.
Although several bibliometric studies have been published focusing on the general field and subfields of quantum technologies (Bornmann et al., 2019; Dhawan et al., 2018; Olijnyk, 2018; Pande and Mulay, 2020; Seskir and Aydinoglu, 2021; Tolcheev, 2018), only a few academic studies investigating the patent landscape have appeared (Chang, 2005; Winiarczyk et al., 2013). One possible explanation is that patent data research relying on keyword searches and the use of cooperative patent classification (CPC) codes alone can easily yield 20% 'false positives' (and up to 80% in some cases), with the resulting risk that a significant number of patents belonging to other fields of technology may be falsely attributed to the domain of QT (Travagnin, 2019). To avoid this problem, we have utilised a recently developed supervised machine-learning method to create a new classification tool for quantum technologies to build a cleaner dataset for more accurate analysis.
Patent analysis studies can provide insights into certain aspects of a field such as technological maturity, commercial interest, market formation, and expectations by actors and stakeholders of returns. Furthermore, they can be used as inputs for higher-level analysis, for example on impact assessment for certain policies like industry-academia collaboration incentives. With these factors in mind, and in addition to the general motivation of our study – to employ a hybrid human-intelligence/AI approach to defining QT, focused on the analysis of patents – the research reported here was conducted with two primary aims. First, we aimed to paint a clearer picture of global patenting activities going on in QT, and to draw some preliminary conclusions from our initial analysis, especially on the distribution and the nature of patenting activities in and by the leading countries in the field. Second, we aimed to demonstrate that AI enhanced patent analysis can provide valuable insights about newly emerging fields of technology that are otherwise difficult or impossible to obtain through conventional searches using CPC codes and keywords alone.
2 Methodology
The methodology we employed to carry out our study relied upon three critically important procedures. The first was to develop a formal operational definition of QT based upon a comprehensive analysis of the semantic content of thousands of patents and published patent applications, using the actual substantive characterisation of the technology contained in the patent documents, rather than pre-conceived classifications based on standardised CPC codes or arbitrary or irregular use of keywords. We engaged in an iterative process, in multiple stages, involving the curated application of AI algorithms to gradually train an AI software system to recognise the distinctive features of an invention that made it a QT invention rather than a member of some other field of technology.
This process, a hybrid AI-human process, involved feeding the AI system with pertinent information in the form of patents in the field of QT obtained by well-known actors operating in the QT commercial landscape, and then systematically evaluating examples of QT patents identified by the AI system from a comprehensive global patent dataset covering all patents and patent applications published by patent offices from over 100 countries over the last several decades. In other words, the method required extensive interaction between intelligent humans who 'trained' the AI software to recognise examples of patents in the field of QT, and the AI system itself that presented to the human beings examples it found from published patent documents worldwide that fitted within the parameters of QT patents it had been trained to recognise at each iteration of the training process. The end result of this iterative process was a robust formal 'definition' of QT embedded in to the AI software as a set of characteristic features of QT against which any published patent document – either a patent or a patent application – could be checked. The QT definition, which in effect is a software-embedded algorithm-based classification tool, is labelled for convenience here as a QT 'classifier'. The overall process of training the AI system and producing the QT classifier – which entailed the careful checking of about 11,600 patent documents, and classifying over 5,000 of those documents as describing genuine QT inventions (i.e., as 'positives'), and over 6,000 as failing to do so (i.e., as 'negatives' or false positives) – is described in detail in the Appendix.
Following completion of the first procedure – the creation of a formal operational definition of QT (i.e., the creation of a QT 'classifier') – we then applied the QT classifier to search the whole population of all patent documents worldwide published digitally online, in all fields of technology, to identify which inventions described in the global patent system matched our AI-curated definition of QT. We calculated the cumulative number of active QT patent families worldwide for the most recent 35 years. A patent family was defined as a single set of pending and granted patents worldwide for one invention in which at least one patent or application in the set had been published. Each patent family in our dataset was allocated to a particular year according to the filing date of the first patent application in the family. In this manner we were able to identify over 14,000 patent families worldwide that with very high confidence may be classified as belonging to the domain of QT, conservatively defined.1 A detailed description of the second procedure is provided in Appendix.
For the third procedure, after having built the final dataset of verified QT patent families during the second procedure, we then applied the AI system to auto-generate technological subcategories from within the dataset itself, by utilising data from the title, abstract, and citations sections of patent documents. In other words, the AI system categorised the broad domain of QT in to sub-fields 'naturally' or endogenously using the internal language of the patent documents themselves – rather than by applying externally imposed classifications such as CPC codes – to characterise the diversity or complexity of the overall QT domain.
To view the full article click here
Originally Published by International Journal of Intellectual Property Management
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
