By World Intellectual Property Organization (WIPO)

Artificial intelligence (AI) is fast becoming a part of our everyday lives. Where a few decades ago, only humans could play chess or read handwriting, both of these tasks are now routinely performed by AI-equipped machines. Today, researchers are working on ever more ambitious applications of AI, which will revolutionize the ways in which we work, communicate, study and enjoy ourselves.

Yet concerns exist about the nature of AI and the challenges it may pose to humanity. Any policy response to these concerns requires a common factual basis for discussions among decision makers. That is why the World Intellectual Property Organization chose AI as the first topic in its new WIPO Technology Trends research series. The WIPO Technology Trends: Artificial Intelligence report draws on data in patent applications and combines it with analysis of scientific publications to create a technical framework for understanding AI innovation. The data-analysis is complemented by interviews with experts in the field on what the future may hold for AI.

Artificial intelligence (AI) emerged in the 1950s, with the first mention of the term coming during the Dartmouth Summer Research Project on Artificial Intelligence in 1956. Since that time innovators and researchers have published over 1.6 million AI-related scientific publications and filed patent applications for nearly 340,000 AI-related inventions. But the history of AI hasn't always been smooth sailing. Periods of optimism, success and growth were followed by disappointment, contraction and regrouping; AI "summers" gave way to AI "winters" as the nascent discipline struggled to find its feet. Recent rapid growth in computing power and communications technologies has enabled the compilation and sharing of large volumes of data, opening up many new areas for AI technological development.

Techniques Machine learning is the dominant artificial intelligence (AI) technique. It is found in 40% of all AI-related patents studied and the technique grew at an average rate of 28% every year from 2013-2016. 40 percenttge of all AI-related patents mention machine learning. Within machine learning, the specific techniques currently revolutionizing AI are deep learning and neural networks. Both of these techniques are instrumental in transforming automatic translation, for example. Mentions of deep learning in patent filings grew annually at an average rate of 175% from 2013-16. Mentions of neural networks grew annually at an average rate of 46% over the same period. Case study Machine learning and perfumery The ability to craft a fragrance is something that takes master perfumers years of experience to develop. A group of IBM researchers and skilled perfumers at Symrise, a global producer of flavors and fragrances, got together to explore how to use AI to do just that. Mixing artistic and scientific thought into one big pot resulted in Philyra – an AI product composition system that can learn about formulas, raw materials, historical success data and industry trends. Philyra uses new, advanced machine learning algorithms to sift through hundreds of thousands of formulas and thousands of raw materials, helping identify patterns and novel combinations. As Philyra explores the entire landscape of fragrance combinations, it can detect gaps in the global fragrance market for which entirely new fragrance formulas can be designed. The first two fragrances to be produced using Philyra will be for Brazilian cosmetics company O Boticário and are scheduled for release in 2019. Applications Computer vision, which includes image recognition (critical for self-driving cars, for instance), is the most popular functional application of artificial intelligence (AI). It was mentioned in 49% of all AI-related patents and grew annually at an average rate of 24% over the period 2013-16. 49 Percentage of all AI-related patents mention computer vision. The other two top areas in functional applications are natural language processing (14% of all AI-related patents) and speech processing (13%). While computer vision, natural language processing and speech processing are the three most important functional applications in terms of the total number of patent filings, others such as robotics and control methods are emerging and growing fast. Case study Using speech processing to turn radio discussions into policy data In Uganda, where most of the population lives in rural areas, radio is a vibrant platform for public discussions, information sharing and news. Talk shows and phone-ins are popular ways for people to voice their needs, concerns and opinions. In this pilot project, UN Global Pulse and the Stellenbosch University in South Africa built speech recognition technology that uses machine learning to convert public discussions in radio broadcasts into text that can be read in several of the languages spoken in Uganda, including Luganda, Acholi, Lugbara and Rutooro. There is a wealth of data that can be extracted from public radio conversations and these data can be mined to support sustainable development and humanitarian efforts. Insights about the spread of infectious diseases, or the way people move during a disaster, or how they perceive healthcare campaigns or access to jobs and education, can be derived from radio talk. In order to protect the right to privacy, the project employs specific tools such as data anonymization, restricting access to the data during project implementation and destroying the data once the project is concluded. Findings from this pilot project continue to be analyzed to understand how the data gathered can be applied to advance the SDGs. Fields The top fields in which artificial intelligence (AI) technologies are employed are: telecommunications: computer networks/internet, radio and television broadcasting, telephony, videoconferencing, and VoIP transportation: aerospace/avionics, autonomous vehicles, driver/vehicle recognition, transportation and traffic engineering life and medical sciences: bioinformatics, biological engineering, biomechanics, drug discovery, genetics/genomics, medical imaging, neuroscience/neurorobotics, medical informatics, nutrition/food science, physiological parameter monitoring, public health 42 Percentage of all AI-related patents filed in telecoms, transportation or life and medical sciences Looking at the ten-year period from 2006-16, growth in transportation technologies stands out. Representing just 20% of applications in 2006, by 2016 transportation accounted for one-third of applications (with more than 8,700 filings). Telecommunications has remained at around 24% during this ten-year period, but the proportion of filings mentioning business, document management and publishing or life and medical sciences has decreased. Case study Saving lives with AI telecoms technology Sudden unexplained death in epilepsy (SUDEP) claims a life every seven to nine minutes. The Empatica Embrace is the first smart watch that uses AI to detect potentially life-threatening convulsive seizures. The watch continuously runs a seizure-detection algorithm, built using machine learning. The AI algorithm within is a support vector machine. This form of supervised learning is trained by collecting lots of data from wearables. An expert neurologist is then asked to provide a medically accurate label for each time chunk of the data. The labels and data are used to train the support vector machine, enabling it to learn how to map data sensed from the wearer’s wrist to labels likely to be given to that data by an expert human. The resulting trained support vector machine is programmed into every watch, where it runs continuously, looking for events that might be a dangerous seizure. When it detects such an event, it communicates with another piece of software (perhaps on a paired smartphone) that issues alerts and makes calls and text messages. In addition, the software logs the data and event timing, so it can be reviewed later by a medical professional. The device was cleared by the FDA (U.S. Food and Drug Administration) in January 2018, and has already been credited with helping save lives. Leaders Companies represent 26 out of the top 30 artificial intelligence (AI) patent applicants. Most of these are multinational firms active in consumer electronics, telecommunications or software. 26 of companies that feature in the top 30 AI patent applicants Acquisitions are relatively common, with 7 out of the top 20 companies having acquired AI firms. Among them, Alphabet has acquired the largest number (18) of AI companies. Most AI-related patent filings are made at the patent offices in the United States of America (152,981 filings) and China (137,010). Both countries combine a high number of innovations in AI and potential as a market for AI-related inventions. Filings under WIPO's Patent Cooperation Treaty (PCT System) represent 20% (67,662) of the total number of AI-related filings.