The G3 – AI areas of specialisation: comparative advantages in AI thematic areas indicator explores the specialisation of geographical areas in the AI field by means of their revealed comparative advantage (RCA). It measures a country’s specialisation in a thematic area in comparison with the global average specialisation for that area.
The RCA is a ratio calculated as share 1 / share 2: share 1 is the share of activities of a geographical area in a thematic area (or technological subdomain) in the total amount of activities in that geographical area; share 2 is the share of activities in that thematic area worldwide in the total amount of activities worldwide. For the calculation activities are assigned to the thematic area that best represents the activity’s content (resulting from a topic-modelling analysis). Since a RCA value of 1 represents the global average specialisation, this is taken as the benchmark (it is represented as a red dashed line in the graph). When the RCA is greater than 1 for a geographical area in an AI thematic area, that geographical area has a comparative advantage in that AI thematic area.
The types of AI activities tracked are: business activities (firms with a core business in AI), research activities (scientific articles in top AI and Robotics international conferences), and innovative activities (AI-related filed priority patent applications). Additionally, a fourth group is considered: the AI-related EC-funded projects. International comparability is granted when the first three groups are used. EC-funded projects are only used for the in-depth analysis of the EU and its Member States.
We use the textual information contained in the activities of the collected microdata to infer their technological content. Through a topic model we identify the following thematic areas or technological subdomains of AI.
- Audio & Natural Language Processing (NLP),
- Computer Vision Applications,
- Machine Learning (ML) Fundamentals,
- ML for Image Processing
- Internet of Everything (IoE),
- Autonomous Robotics,
- Connected and Automated Vehicles (CAVs),
- AI Services
— Audio and Natural Language Processing (NLP): Audio Processing AI systems facilitate the perception or generation (synthesis) of audio signals, including speech, and also other sound material (e.g., environmental sounds, music). Natural Language Processing is a machine’s ability to identify, process, understand and/or generate information in written and spoken human communications.
— Computer Vision applications are activities that identify human faces and objects in digital images, as part of object-class detection.
— Machine Learning (ML) Fundamentals are the ability of systems to automatically learn, decide, predict, adapt and react to changes, improving from experience, without being explicitly programmed.
— ML for Image Processing are machine learning methods used for image processing activities.
— The Internet of Everything (IoE): this refers to the interconnectivity of various technologies, processes and people. The human interaction in this context allows people to monitor or configure devices and processes through interfaces.
— Automation refers to activities related to the production or use of physical machines, computer software and other technologies to perform repetitive tasks, for which they are specifically designed and programmed. They can have several degrees of freedom, e.g., in terms of movement, and they may include intelligent control modules to interact with the environment in a controlled setting, e.g., using a temperature sensor. However, they are limited to a set of actions for which they are designed to operate, and have to be re-programmed for new or additional operations. The use of AI in automated machines is mainly related to the adaptation of the defined set of operations as a reaction to external parameters.
— Autonomous Robotics: activities related to the development or use of robotic systems that are meant to operate in a relatively complex environment involving interaction with other machines or humans. Autonomous robots perform multiple operations without any prior exact set of instructions, nor programmed sequence of actions. AI enables autonomous robots to have this higher degree of autonomy compared with automated machines.
— Connected and Automated Vehicles (CAVs) involve technologies for autonomous vehicles, connected vehicles and driver assistance systems, considering all automation levels and all communication technologies (V2X).
— AI Services are activities related to the provision of (online) AI services and applications, including infrastructure, software and platform services (e.g., cognitive computing, ML frameworks, bots and virtual assistants, etc.).
This indicator presents large values in the thematic area of AI Services, which refers to the provision of AI services and applications, including infrastructure, software and platform services. All major countries except Japan, South Korea and China have a value larger than 2. On the one hand, this indicates a clear orientation of western areas economies towards the development of businesses based on AI services. On the other, it also reflects the different involvement of China, Japan and South Korea in the AI landscape, which are more oriented towards the development of patents.
The advantage of the EU is especially evident in two thematic areas.
- The first is AI Services, highlighting the salient role of the EU AI players in the provision of services between firms (B2B) or to the end-consumers (B2C).
- The second is AutonomousRobotics, which is expected to positively affect the EU’s competitiveness and sustainability in industry and services, and is increasingly impacting many sectors, such as health, logistics or manufacturing, among others
The European Union
Regarding EU Member States, some Eastern European countries, such as Romania, Bulgaria, Slovakia, Poland and Croatia, show high RCA values in the thematic area of Automation. However, these countries are not involved in many AI activities, and in some cases this high RCA is the result of a relatively higher concentration in this AI thematic area, even if the number of related activities is small.
We note strong specialisations in Belgium and Ireland for Audio & NLP, in Spain for Automation, in Sweden for Connected and Automated Vehicles (CAVs), in Belgium for Computer Vision Applications, in Belgium and Ireland for Machine Learning Fundamentals, in Belgium and Finland for Machine Learning for Image Processing, and in Greece and Italy for AutonomousRobotics .