Stanford University’s Institute for Human-Centered Artificial Intelligence (HAI) has published its 2022 AI Index annual report. The report identifies top trends in AI, including advances in technical achievements, a sharp increase in private investment, and more attention on ethical issues.
This year marks the fifth edition of the AI Index report, which is compiled by an interdisciplinary team in cooperation with government, industry, and academia. The report contains five chapters, and the editors have distilled several key takeaways from the Index, including: that in 2021, worldwide private investment in AI 2021 totaled around $93.5 billion, more than doubling since 2020; United States and China have the greatest number of cross-country research collaborations; and the number of AI patents filed in 2021 is 30x more than in 2015. Along with the report, HAI has updated their Global AI Vibrancy Tool which compares 29 countries across 23 AI-related metrics. According to HAI, the goal of the AI Index is:
[T]o provide unbiased, rigorously vetted, and globally sourced data for policymakers, researchers, executives, journalists, and the general public to develop a more thorough and nuanced understanding of the complex field of AI. The report aims to be the world’s most credible and authoritative source for data and insights about AI.
The report is organized into five chapters: Research and Development; Technical Performance; Technical AI Ethics; Economy and Education; and AI Policy and Governance. These chapters are based on publicly available data and contain dozens of charts. The first chapter examines trends in AI research and development, including journal articles, conference papers, and patents. The chapter also tracks the popularity of AI and deep learning software packages, with TensorFlow remaining the leader by a wide margin. The second chapter examines progress in different subfields of AI and machine learning, by tracking performance on benchmarks and challenge competitions across fields such as robotics, computer vision, and natural language processing (NLP). The report also analyzes the cost of training baseline models, and finds that since 2017, training an ImageNet computer vision model has decreased "by a factor of 223."
Chapter 3, Technical AI Ethics, examines efforts to measure progress in developing fairer AI models. The report notes that while large language models continue to improve their performance on NLP benchmarks, they also tend to have an increase in "elicited toxicity." Research in this area has expanded, with AI-industry researchers contributing 71% more publications at ethics conferences in 2021, compared with 2020. The fourth chapter examines how AI shapes the economy and education, tracking the growth of private industry investment as well as human capital, noting that "1 in every 5 CS students" who earn a PhD specializes in AI or ML. The final chapter examines government efforts to regulate and manage AI. The report finds that in a survey of 25 countries, the number of laws passed that mention AI "grew from just 1 in 2016 to 18 in 2021."
Several users reacted to a tweet from the Index's Twitter account announcing the report. Most praised the work and called attention to specific findings. Teemu Roos, a professor of AI at University of Helsinki, wrote:
One thing that caught my eye was that some of the curves showed a downwards turn since last year (esp. conferences). Is it a sign of a looming AI winter? Or are we just maturing beyond the hype and not labeling every other thing as "AI"?
The full report can be downloaded from the AI Index website. The report's raw data and high-res images are publicly available on Google Drive. The report is licensed under the Creative Commons Attribution-NoDerivatives 4.0 International license.