斯坦福-2019年人工智能指数报告（英文）291页_10mb

AI Index 2019 Annual Report Summary

Core Content

The AI Index 2019 Annual Report is an independent initiative by Stanford University's Human-Centered Artificial Intelligence Institute (HAI). It provides a comprehensive, unbiased overview of AI development and its impact across various domains, including research, economy, education, and societal considerations. The report includes data from global communities and is supported by a range of partners, including Google, OpenAI, and McKinsey Global Institute.

Key Contributors and Partners

• Steering Committee Members: Raymond Perrault (report coordinator), Yoav Shoham (chair), Erik Brynjolfsson (MIT), Jack Clark (OpenAI), John Etchemendy (Stanford), Barbara Grosz (Harvard), Terah Lyons (Partnership On AI), James Manyika (McKinsey), Saurabh Mishra (Stanford), Juan Carlos Niebles (Stanford)
• Supporting Partners: McKinsey&Company, Google, genpact, AI21labs, PwC, Quid, etc.

Main Chapters and Highlights

Chapter 1: Research and Development

• The volume of peer-reviewed AI papers has grown by over 300% from 1998 to 2018, now accounting for 3% of peer-reviewed journal publications and 9% of conference papers.
• China has surpassed the US in AI paper publications since 2006, though the US still leads in Field-Weighted Citation Impact (FWCI), with a 40% higher citation rate than the global average.
• Singapore, Switzerland, Australia, Israel, Netherlands, and Luxembourg have high per capita Deep Learning (DL) paper publication rates.
• East Asia accounts for over 32% of global AI journal citations, while North America contributes over 40% of conference paper citations.
• North America dominates AI patent citations, representing over 60% of global activity from 2014 to 2018.
• Women in AI Research: Despite growth in participation, women remain underrepresented, comprising less than 20% of new faculty hires and 20% of AI PhD recipients in the US since 2010.

Chapter 2: Conferences

• AI conference attendance has increased significantly, with NeurIPS expecting 13,500 attendees in 2019, up 41% from 2018 and 800% from 2012.
• WiML and AI4ALL have seen substantial growth in participants, indicating increased efforts to promote diversity in AI.
• The report highlights the importance of inclusivity and the need for continued support for underrepresented groups.

Chapter 3: Technical Performance

• The time to train large image classification systems has dropped from about three hours in 2017 to 88 seconds in 2019.
• Natural Language Processing (NLP) has seen rapid progress in benchmarks like SuperGLUE and SQuAD2.0, though performance on tasks requiring reasoning (e.g., AI2 Reasoning Challenge) remains lower.
• AI compute power has been doubling every 3.4 months since 2012, compared to every two years before.

Chapter 4: The Economy

• AI hiring growth has been fastest in Singapore, Brazil, Australia, Canada, and India from 2015 to 2019.
• In the US, AI-related job postings increased from 0.26% in 2010 to 1.32% in 2019, with Machine Learning being the most common specialization.
• Global private AI investment in 2019 reached over $70B, with $37B in startups, $34B in M&A, $5B in IPOs, and $2B in minority stakes.
• Autonomous Vehicles (AVs) received the largest share of AI investment in 2019, at $7.7B (9.9% of total), followed by Drug, Cancer, and Therapy ($4.7B, 6.1%), Facial Recognition ($4.7B, 6.0%), Video Content ($3.6B, 4.5%), and Fraud Detection and Finance ($3.1B, 3.9%).
• 58% of large companies reported AI adoption in 2019, up from 47% in 2018. However, only 19% of companies are addressing algorithm explainability risks, and 13% are tackling equity and fairness issues like algorithmic bias.

Chapter 5: Education

• AI education enrollment is growing rapidly, both in traditional universities and online platforms.
• AI is the most popular specialization for computer science PhD students in North America, with over 21% of graduates specializing in AI/Machine Learning in 2018.
• International PhD students in AI now make up over 60% of graduates in the US and Canada, up from less than 40% in 2010.
• Industry is the largest consumer of AI talent, with over 60% of AI PhD graduates entering the private sector in 2018, compared to 20% in 2004.
• AI faculty leaving academia for industry has increased significantly, with over 40 departures in 2018.

Chapter 6: Autonomous Systems

• In California, the number of miles driven by autonomous vehicles and the number of testing companies increased over seven-fold between 2015 and 2018.
• Over 50 companies and 500 AVs were licensed for testing in 2018, driving over 2 million miles.

Chapter 7: Public Perception

• Global central banks (e.g., Bank of England, Bank of Japan, Federal Reserve) show increasing interest in AI.
• AI-related legislation is rising globally, indicating growing awareness of its implications.

Chapter 8: Societal Considerations

• Fairness, interpretability, and explainability are the most frequently mentioned ethical challenges in AI.
• Over 3600 global news articles from mid-2018 to mid-2019 discuss AI ethics, data privacy, face recognition, and algorithmic bias.
• AI has the potential to contribute to all 17 UN Sustainable Development Goals (SDGs), though challenges remain in scaling its impact.

Chapter 9: National Strategies and Global AI Vibrancy

• The Global AI Vibrancy Tool (vibrancy.aiindex.org) allows readers to compare countries based on 34 indicators, highlighting the emergence of local AI centers.
• Finland excels in AI education, India in skill penetration, Singapore in government support, and Israel in private investment in AI startups per capita.

Public Data and Tools

• Raw data and interactive tools are available for public use, including:
  • Global AI Vibrancy Tool: Compares countries across 34 indicators.
  • AI Index arXiv Monitor: Enables full-text searches of AI-related papers on arXiv.
• The Technical Appendix provides detailed methodologies and sources for all data presented.

Conclusion and Measurement Insights

• The AI Index aims to provide a nuanced view of AI development, avoiding simplistic rankings between the US and China.
• The report underscores the importance of measurement in AI policy and highlights the need for more inclusive practices in the field.
• Measurement Questions are included in each chapter to encourage deeper analysis and discussion of data and metrics.

Citations

The AI Index 2019 Annual Report is authored by a group of experts and is available under a Creative Commons Attribution-NoDerivatives 4.0 License. It can be cited as:

Raymond Perrault, Yoav Shoham, Erik Brynjolfsson, Jack Clark, John Etchemendy, Barbara Grosz, Terah Lyons, James Manyika, Saurabh Mishra, and Juan Carlos Niebles, "The AI Index 2019 Annual Report", AI Index Steering Committee, Human-Centered AI Institute, Stanford University, Stanford, CA, December 2019.