On The Global AI Index. Interview with Alexandra Mousavizadeh
by Roberto V. Zicari on January 18, 2020
“The US is the undisputed leader in AI development, the Index shows. The western superpower scored almost twice as highly as second-placed China, thanks to the quality of its research, talent and private funding. America was ahead on the majority of key metrics – and by a significant margin. However, on current growth experts predict China will overtake the US in just five to 10 years.” –Alexandra Mousavizadeh.
I have interviewed Alexandra Mousavizadeh, Partner, Tortoise Media, Director, Tortoise Intelligence. We talked about “The Global AI Index”.
Q1. On 3rd of December 2019 in London, you have released “The Global AI Index” ranking 54 countries. What was the prime motivation for producing such an index?
Alexandra Mousavizadeh: Artificial intelligence is an engine of change, for better or for worse. Increasingly, our daily lives are impacted by technologies using machine learning, and businesses are using them to support more and more of their processes.
Our motivation for producing the Index here at Tortoise was to monitor and help explain this change on a global scale. The initial request came from three governments who were seeking a comprehensive and detailed index that would help them set and track their national AI strategy. As a news company focused on understanding what forces are driving geopolitical, environmental and social change we knew we needed to focus on artificial intelligence. At Tortoise Intelligence, our data and analytics team, the tool for doing this is the composite index.
Q2. How did you choose the 54 countries?
Alexandra Mousavizadeh: The 54 countries were chosen to represent those that had lifted artificial intelligence to the top of the national agenda in some way; publishing a national strategy, appointing a minister for artificial intelligence, setting up public and private sector collaborations and institutes. Ultimately the list of 54 was a selection that represents the countries in which data was beginning to be gathered on the relevant factors, and those that are stepping onto the world stage in terms of development.
Q3.Of the 150 indicators you have chosen, which one(s) are most relevant for the ranking?
Alexandra Mousavizadeh: Our overall approach was to represent the fact that:
Artificial intelligence is still the product of human intelligence; and therefore talent is a priority. Talented practitioners and developers who can innovate and implement new technologies. A leading indicator, and one that is very relevant to the ranking is the number of data scientists active in a given country. For this indicator we drew data from GitHub, StackOverflow, Kaggle and LinkedIn.
Research into artificial intelligence is also a leading factor; making skills researchers and the generation of new understandings and techniques another priority. A leading indicator, and another that impacts the rankings significantly is the number of researchers in top rated journals in a given country.
Finally, money remains the primary catalyst to activity on artificial intelligence. Talent, and research, come at a premium to businesses and other institutions. So commercial funding is another leading indicator, with total amount of investment into artificial intelligence companies being an impactful indicator.
Q4. What criteria did you use to weight each indicator for importance?
Alexandra Mousavizadeh: Throughout the course of our consultations with the advisory boards, and the many ThinkIns held at Tortoise during the development of the Index, we put together a model for explaining the significance of each sub-pillar in terms of building capacity for artificial intelligence. As described, the leading factors were talent, research and investment; mostly expressing that financial and intellectual capital currently trump all other factors.
Our experts were consulted across the full range of indicators, and we reached a consensus on the importance. We recognised that this remains a subjectively constructed set of weightings, which is why we have conducted testing to demonstrate that the impact of the weightings is relatively insignificant compared to the impact of the actual values themselves.
Q5. Why have you presented an index ranking on capacity?
Alexandra Mousavizadeh: At present the availability of information is growing rapidly, and the question as to how to manage and interpret this information is growing more urgent. Composite indicators meet the need to consolidation – through aggregation – a large amount of data into a set of simplified numbers that encompass and reflect the underlying complexity of the information. All indices constructed from composite indicators should be interpreted with caution, and scrutinised carefully before important conclusions are drawn out. In alignment with the OECD ‘Handbook on constructing composite indicators’; ‘capacity’ is the multi-dimensional concept and the underlying model around which the individual indicators of The Global AI Index are compiled.
Capacity – the amount of something that a system can contain or produce – is the organising concept of The Global AI Index. It is an appropriate means of considering the relationship between the different relevant factors that exist within a given nation. Increased capacity, in this case, can be understood as an increased ability to generate and sustain artificial intelligence solutions, now and in the future. The Artificial Intelligence for Development Organisation talk about ‘capacity’ for exactly this reason; it speaks both to the current organisation of productive factors that contribute to technological development, as well as future potential for generating new innovations in their use, and in the design of the technologies themselves.
Q6.Is it reasonable to compare nations of vastly different sizes when considering capacity?
Alexandra Mousavizadeh: We have constructed our data set to demonstrate both gross capacity, and proportional capacity – or intensity – with the intensity rankings being very different from the headline rankings for gross capacity. We believe that the answer to this question hinges on what you believe the purpose of a comparative index is; we think that they are an excellent tool for condensing a lot of complexity into a simpler conclusion that can be understood and tackled be experts and non-experts alike.
By creating a number of clusters within the 54 countries we have tried to present the rankings in a more like for like way. For example, the UK can be considered in relation to the full set, and to its closest competitors which we call the ‘Traditional Champions’ of higher education, research and governance. These nations; including Canada, France and Germany are facing some of the same challenges when it comes to development and adoption. In future editions we may choose to dig more deeply into the question of intensity versus raw capacity.
Q7.What data sources did use for The Global AI Index? What about Missing values or incorrect values?
Alexandra Mousavizadeh: The vast majority of sources used for The Global AI Index are publically available and open source; only one of which is proprietary. This was the Crunchbase API, which was drawn on for data in the ‘Commercial Ventures’ sub-pillar. A full list of the sources used in the The Global AI Index is available in the indicator table. Some headline sources are Crunchbase, GLUE, IEEE, GitHub API, LinkedIn and SCOPUS.
Missing values represent approximately 4.5% of the collected data-set for The Global AI Index. There was a limited amount of data available with which to train an imputation model – although this was strongly considered as an option – and as such there are a variety of imputation techniques employed.
Imputation by zero – used when data is not pre-defined but is the logical or necessary value; e.g, if the number of Kaggle Grandmasters is empty it is most likely because a country has never had one.
Imputation by average value – used when the variable in question is independent of a country’s population size or GDP; placing the mean or median value in place of a missing value.
Imputation by last observation carried forward – used when alternative data sources show only values from previous years; in some cases previous values are taken as indicators of a country’s current state.
Imputation by model – used in observation of obvious relationships between a country’s demographics – population, GDP, employment rates, etc. In some cases it was necessary to build a generalised linear model to predict what value should be used.
Q8. What are the key findings?
Alexandra Mousavizadeh: We believe that the key findings of the Index to date are: The US is the undisputed leader in AI development, the Index shows. The western superpower scored almost twice as highly as second-placed China, thanks to the quality of its research, talent and private funding. America was ahead on the majority of key metrics – and by a significant margin. However, on current growth experts predict China will overtake the US in just five to 10 years.
China is the fastest growing AI country, our Index finds, overtaking the UK on metrics ranging from code contributions to research papers in the past two years. Last year, 85 per cent of all facial recognition patents were filed in China, as the communist country tightened its grip on the controversial technology. Beijing has already been condemned for using facial recognition to track and profile ethnic Muslims in its western region.
Britain is in third place thanks to a vibrant AI talent pool and an excellent academic reputation. This country has spawned hugely successful AI companies such as DeepMind, a startup founded in 2010 which was bought by Google four years later for $500 million. Britain has been held back, however, by one of the slowest patent application processes in any of the 51 countries. Other countries are snapping at its heels.
Q9. What other findings did you find relevant or surprising?
Smaller countries – such as Israel, Ireland, New Zealand and Finland – have developed vibrant AI economies thanks to flexible visa requirements and positive government intervention. Israel’s Mobileye Vision Technology, which provides technology for autonomous vehicles, was purchased in 2017 by Intel for $15.3 billion.
More than $35 billion has been publicly earmarked by governments to spend on AI development over the next decade, with $22 billion promised by China alone. Many more billions may have been allocated secretly through defence departments which are not made public.
Countries are using AI in very different ways. Russia and Israel are among the countries focusing AI development on military applications. Japan, by contrast, is predominantly using the technology to cope with its ageing population.
Q10. What do we learn overall from this index?
Alexandra Mousavizadeh: We’ve learned more about the vast scale of activity on artificial intelligence and cut through some of the noise about how and why it is changing the world. We’ve been able to uncover a lot of information about collaboration between supposed rivals, informal learning of coding and machine learning skills, and a lot about the availability and competition for talent.
Q11. What were the main challenges in creating such an Index?
Alexandra Mousavizadeh: Building up a network of people who are sufficiently knowledgeable to scrutinise and comment on the process. Dealing with a vast number of data points that need to be normalised, and made comparable. Checking the provenance and robustness of the data points.
Q12. Where are the ethical considerations in this index?
Alexandra Mousavizadeh: Ethics have been a major focus in our conversation about artificial intelligence. We decided that The Global AI Index would solely measure capacity and activity. An index on AI Ethics is planned for this year.
Firstly, the most ethical model for developing and adopting artificial intelligence just hasn’t emerged yet, and perhaps it never will. This lack of consensus means that it is more difficult to select variable that show better or worse ethical considerations.
However, we have significant plans throughout 2020 to build upon our work on ethics and artificial intelligence. We hope this work will amount to another product within the year, one that reflects the complexities of governance in relation to artificial and where in the world the most is being done to safeguard good outcomes for all.
Q13 The fast-changing processes of innovation and implementation in artificial intelligence requires constant re-examination. How do you intend to keep up with such constant changes? and how do you plan to improve the index in the future?
Alexandra Mousavizadeh: We have planned a bi-annual refresh of the Index, drawing in new values for a range of our indicators to keep the rankings dynamic.
Our series of ThinkIns and events are Tortoise will also continue throughout the year. These represent fantastic opportunities to build upon our methodology and move the conversation into new areas. We are currently hoping to improve the index by:
Adding more data on importation and exports of computing hardware, chip designs and by expanding our data reach on patents.
Including data capture statistics, in an attempt to show which nations are building the largest and most useful data-sets. This will also fit into our investigation of data privacy and governance. Our most recent ThinkIn – which you can watch here – on ‘data rules’ focused on the various models for using data and the risks associated with each.
Q14. Are you planning to release the data open source?
Alexandra Mousavizadeh: We’ve already shared the underlying data-set with a range of partners and interested parties. Ultimately we hope the Index will be a tool for developing better understanding, and we will look to share the data as part of the ongoing conversation.
Alexandra Mousavizadeh is a Partner at Tortoise Media, running the Intelligence team which develops indices and data analytics. Creator of the recently released Responsibility100 Index and the new Global AI Index. She has 20 years’ experience in the ratings and index business and has worked extensively across the Middle East and Africa. Previously, she directed the expansion of the Legatum Institute’s flagship publication, The Prosperity Index, and all its bespoke metrics based analysis & policy design for governments. Prior roles include CEO of ARC Ratings, a global emerging markets based ratings agency; Sovereign Analyst for Moody’s covering Africa; and head of Country Risk Management, EMEA, Morgan Stanley.
Prof. Roberto V. Zicari is editor of ODBMS.ORG (www.odbms.org).
ODBMS.ORG is designed to meet the fast-growing need for resources focusing on AI, Big Data, Data Science, Analytical Data Platforms, Scalable Cloud platforms, NewSQL databases, NoSQL datastores, In-Memory Databases, and new approaches to concurrency control.
The portal was created to serve software professionals in the open source community or at commercial companies as well as faculty and students at educational and research institutions.
Roberto is Full Professor of Database and Information Systems at Frankfurt University. He was for over 15 years the representative of the OMG in Europe. Previously, Roberto served as associate professor at Politecnico di Milano, Italy; Visiting scientist at IBM Almaden Research Center, USA, the University of California at Berkeley, USA; Visiting professor at EPFL in Lausanne, Switzerland, the National University of Mexico City, Mexico and the Copenhagen Business School, Danemark.