By Dr Gyan Pathak
Artificial Intelligence (AI) is transforming economies and promising new opportunities for productivity, growth, and resilience. Countries across the world are also responding with national AI strategies to capitalize on these transformations. However, no country today has sufficient data on, or a targeted plan for, national AI compute capacity. This polity blind-spot may jeopardise domestic economic goals.
This is the gist of a recent OECD digital economy paper titled ‘A Blueprint for Building National Compute Capacity for Artificial Intelligence.’ This report has provided the first blueprint for policy makers to help assess and plan for the national AI compute capacity needed to enable productivity gains and capture AI’s full economic potential.
It is worth noting that India is presently in the process of developing the National Artificial Intelligence Resource Portal under the aegis of the Centre of Excellence in Artificial Intelligence. The platform will offer a web-based system to search and browse AI resources, including training and a cloud-based compute platform. The other countries that have taken certain initiatives in the regard include Canada, Chile, Colombia, France, Germany, Japan, Korea, Slovenia, Spain, UK, US, Serbia, Thailand, and Europe (to be shared with member countries). However, all initiatives are far less than the requirement.
It is despite the fact that the Governments had committed themselves to the first intergovernmental standards on AI in the 2019 OECD Principles on Artificial intelligence, “fostering the development of, and access to, a digital ecosystem for trustworthy AI” including underlying infrastructure such as AI compute. The progress on the part of the governments across the world is obviously slower than the higher speed development in the field of AI.
Only few economies have supercomputers ranking as top computing systems, with emerging economies sparsely represented on the Top500 list. The November 2022 Top500 list shows 34 economies with a “top supercomputer”. The highest concentration (32%) of top supercomputers is in the People’s Republic of China, followed by the United States (25%), Germany (7%), Japan (6%), France (5%) and the United Kingdom (3%). The 17 countries on the list from the European Union (EU27) make up a combined 21% of top supercomputers. Beyond this group, the rest of the world makes up 12% of top supercomputers. Nearly 90% of top supercomputers were developed in the last five years. In recent years, supercomputer systems have been increasingly updated to also run AI-specific workloads, although the list does not distinguish supercomputers according to workload capacity specialised for AI. However, the simple count of Top500 list does not reveal the full picture due to variation in number and capacity in performance.
After defining AI compute, the report takes stock of indicators, datasets, and proxies for measuring nation AI compute capacity, and identifies obstacles for measuring and benchmarking nation AI compute capacity across countries. Then it suggests AI compute plan along three dimensions – capacity, effectiveness, and resilience. Capacity covers availability and use of AI; effectiveness covers people, policy, innovation, and access; and resilience covers security, sovereignty, and sustainability.
However, embracing AI-enabled transformation depends on the availability of infrastructure and software to train and use AI models at scale. Ensuring countries have sufficient such “AI compute capacity” to meet their needs is critical to capturing AI’s full economic potential.
Many countries have developed national AI strategies without fully assessing whether they have sufficient domestic AI compute infrastructure and software to realise their goals. Other AI enablers, like data, algorithms, and skills, receive significant attention in policy circles, but the hardware, software, and related infrastructure that make AI advances possible have received comparatively less attention.
Today, standardised measures of national AI compute capacity remain a policy gap. Such measures would give OECD and partner economies a greater understanding of AI compute and its relationship to the diffusion of AI, improve the implementation of AI strategies, and inform future policy and investments.
The demand for AI compute has grown dramatically for machine learning systems, especially deep learning and neural networks. According to research, the computational capabilities required to train modern machine learning systems, measured in number of mathematical operations (i.e., floating-point operations per second, or FLOPS), has multiplied by hundreds of thousands of times since 2012, despite algorithmic and software improvements that reduce computing power needs. The increasing compute needs of AI systems create more demand for specialised AI software, hardware, and related infrastructure, along with the skilled workforce necessary to utilise them efficiently and effectively.
As governments invest in developing cutting-edge AI, compute divides can emerge or deepen. An imbalance of such compute resources risks reinforcing socioeconomic divides, creating further differences in competitive advantage and productivity gains. Over the past decade, private sector led initiatives within countries have increasingly benefitted from state-of-the-art AI compute resources, particularly from commercial cloud service providers, compared to public research institutes and academia. The OECD. AI Expert Group on AI Compute and Climate advances collective understanding and measurement of AI compute to shed light on AI compute divides between countries and within national AI ecosystems.
Findings and measurement gaps are identified by the report to inform future work in developing AI-specific metrics to quantify and benchmark AI compute capacity across countries. They include: national AI policy initiatives need to take AI compute capacity into account; national and regional data collection and measurement standards need to expand; policy makers need insights into the compute demands of AI systems; AI-specific measurements should be differentiated from general-purpose compute; workers need access to AI compute related skills and training for effective AI compute use; and AI compute supply chains and inputs need to be mapped and analysed. (IPA Service)