2024-04-22-斯坦福大学-2024斯坦福AI指数报告（英文版）_502页_21mb

AI Index Report 2024 Summary

Core Content

The AI Index Report 2024 is the seventh edition of the report and the most comprehensive to date, offering a detailed overview of AI's technical progress, societal impact, and policy landscape. The report highlights both the rapid advancements in AI technology and the growing concerns surrounding its implications for employment, privacy, and ethical use.

Main Points and Key Information

Technical Advancements

• AI Outperforms Humans in Some Tasks: AI has surpassed human performance in image classification, visual reasoning, and English understanding, but still lags in complex tasks like competition-level mathematics and visual commonsense reasoning.
• Multimodal AI Emerges: AI systems like GPT-4 and Gemini Ultra are now capable of handling multiple modalities, including text, images, and audio.
• Performance on Harder Benchmarks Improves: New benchmarks such as SWE-bench, HEIM, and MMMU have been introduced, pushing AI models to tackle more complex tasks.
• AI-Generated Data Enhances Models: AI tools like SegmentAnything and Skoltech are used to create specialized data for tasks such as image segmentation and 3D reconstruction, which in turn helps improve AI performance.
• Human Evaluation Gains Importance: As AI generates high-quality content, human evaluations are increasingly used to benchmark performance, such as the Chatbot Arena Leaderboard.

AI Development Landscape

• Industry Dominates Research: In 2023, industry produced 51 notable machine learning models, while academia contributed only 15. There were also 21 models from industry-academia collaborations.
• AI Models Are Becoming More Expensive: Training costs for state-of-the-art AI models have reached unprecedented levels. For example, GPT-4 used $78 million in compute, while Gemini Ultra cost $191 million.
• U.S. Leads in AI Model Production: The U.S. produced 61 notable AI models in 2023, compared to 21 from the EU and 15 from China.
• Open-Source AI Grows Rapidly: The percentage of open-source AI models increased to 65.7% in 2023, up from 44.4% in 2022 and 33.3% in 2021.
• AI Projects on GitHub Surge: The number of AI-related projects on GitHub rose from 845 in 2011 to approximately 1.8 million in 2023, with a 59.3% increase in 2023 alone.

AI and the Economy

• Generative AI Investment Surges: Despite a decline in overall AI private investment, generative AI funding nearly octupled to $25.2 billion in 2023.
• U.S. Maintains Leadership in AI Investment: The U.S. saw AI investments reach $67.2 billion in 2023, nearly 8.7 times more than China. U.S. investment increased by 22.1% since 2022.
• AI Increases Productivity and Revenue: A McKinsey survey found that 42% of organizations reported cost reductions and 59% saw revenue increases from AI implementation.
• AI Job Market Shrinks: AI-related job postings in the U.S. decreased from 2.0% in 2022 to 1.6% in 2023, due to fewer postings from leading firms and a reduced proportion of tech roles.
• AI Adoption Grows: 55% of organizations now use AI in at least one business function, up from 50% in 2022 and 20% in 2017.
• China Dominates Industrial Robotics: China has become the world leader in industrial robot installations, with a 52.4% share in 2022, up from 20.8% in 2013.
• Collaborative and Service Robots Increase: Collaborative robots accounted for 9.9% of installations in 2022, and service robots (excluding medical) saw a rise in adoption across all categories.

Responsible AI and Ethics

• Lack of Standardized Evaluation: There is a significant lack of standardized benchmarks for assessing AI responsibility, making it difficult to compare risks across models.
• Deepfakes and Misinformation: Political deepfakes are becoming easier to generate and harder to detect, with AI tools like CounterCloud enabling the rapid spread of fake content.
• Complex Vulnerabilities in LLMs: Researchers have identified less obvious ways to manipulate AI models, such as infinite repetition of random words, to elicit harmful behavior.
• AI and Copyright Concerns: LLMs may generate outputs containing copyrighted material, raising legal questions about potential violations.
• Transparency Issues: The Foundation Model Transparency Index indicates a lack of openness in AI development, particularly around training data and methodologies.
• AI Bias Concerns: ChatGPT has been found to exhibit political bias toward Democrats in the U.S. and the Labour Party in the U.K., which could influence user opinions, especially during major elections.

Public Perception and Policy

• Global Awareness of AI Grows: A survey by Ipsos found that 66% of people believe AI will dramatically affect their lives in the next 3–5 years, up from 60% the previous year. 52% express nervousness about AI, up 13 percentage points from 2022.
• AI Regulations Rise in the U.S.: The U.S. passed 25 AI-related regulations in 2023, a 56.3% increase from 2022, and the number of AI-related legislative mentions has never been higher.
• Policymakers Seek Balance: Governments are increasingly involved in shaping AI's future by encouraging innovation while managing risks like employment displacement, privacy, and misinformation.

Key Areas Tracked

• Technical Progress: AI's capabilities in various domains.
• Community and Investment: Trends in AI research, funding, and development.
• Public Opinion: Societal views on AI's impact and risks.
• Policy Measures: Government actions to regulate and promote AI.

Supporting Partners and Contributors

The AI Index Report 2024 is supported by numerous organizations and individuals, including Google, OpenAI, Quid, GitHub, McKinsey & Company, and others. It is also contributed to by researchers from Stanford University and various academic institutions.

Citation

Nestor Maslej, Loredana Fattorini, Raymond Perrault, Vanessa Parli, Anka Reuel, and others, "The AI Index 2024 Annual Report," AI Index Steering Committee, Institute for Human-Centered AI, Stanford University, Stanford, CA, April 2024.

Tools and Resources

• Public Data and Charts: Available on Google Drive.
• Global AI Vibrancy Tool: An interactive tool to compare AI ecosystems across 30+ countries, to be updated in 2024.

Conclusion

The AI Index Report 2024 underscores the transformative potential of AI, highlighting both its capabilities and its risks. As AI continues to evolve, the report serves as a critical resource for understanding its impact and guiding responsible development and use.