Stanford University’s Institute for Human-Centered Artificial Intelligence (HAI) has published its 2021 AI Index annual report. This underlying data for this year's report has been expanded compared to the previous year's, and the report includes several perspectives on the COVID-19 pandemic's impact on AI research and development.
To improve the latest report, over the last year the Institute requested feedback from over 140 members of academia, government, and industry. In response, the new report contains more data and analysis on technical performance, diversity, and ethics. The final report contains seven chapters, and the report summary distills nine key takeaways. In particular, AI applications for molecular biology and drug discovery received the most private investment, in part due to the pandemic. The pandemic also caused many AI conferences to switch to a virtual format, resulting in an increased participation. However, the pandemic has not adversely affected AI investment and hiring, and an increased percentage of PhD graduates in AI fields took industry jobs this year compared to previous years.
The report is organized into seven chapters: Research and Development; Technical Performance; The Economy; AI Education; Ethical Challenges of AI Applications; Diversity in AI; and AI Policy and National Strategies. These chapters are based on publicly available datasets and contain dozens of charts. The Research and Development chapter covers the growth of research papers and conferences over time and by geographic region. Technical Performance tracks AI accuracy on several benchmarks in computer vision (CV), natural language processing (NLP), and molecular biology. The report points out that AI systems can now generate speech, prose, and images of such quality that humans are often unable to identify the results as synthetic.
GAN Progress on Face Generation (Image Source: Stanford HAI AI Index Report)
The Economy chapter focuses on trends in jobs and investment by country, while the AI Education chapter looks at university course offerings and PhD graduates in AI; a key takeaway is that in North America, 65% of these new PhDs chose jobs in industry over academia, compared to 44.4% the previous year. The Ethical Challenges chapter notes that the team was "surprised to discover how little data there is on this topic," and in particular calls out a lack of benchmarks. The Diversity chapter also cites a lack of publicly available data, but does point out that the "AI workforce remains predominantly male."
The final chapter, AI Policy and National Strategies, shows trends in published versions of nations' AI strategies and international collaborations on AI. It also focuses on the U.S. government's role in AI investment and policy-making. The report notes that the 116th Congress is "the most AI-focused congressional session in history," with AI being mentioned more than three times as often as in the previous Congress.
The Institute has also updated its interactive Global Vibrancy Tool for comparing up to 26 countries across 22 metrics. The metrics measure performance on various research and development, economic, and inclusion factors, such as number of conference papers, number of patents, and investment. The tool can show an overall index for the full list of countries, or detailed metrics for a single country, and contains data from the year 2015 up to 2020.
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.