April 15, 2025 By Chu Chen-Tso
The Stanford Institute for Human-Centered AI (HAI), a top global AI research institution, has released its annual "2025 Artificial Intelligence Index Report." The 456-page report is divided into eight chapters covering Research and Development, Technical Performance, Responsible AI, Economy, Science and Medicine, Policy and Governance, and Education and Public Opinion.
Below are 18 key highlights selected from the full report:
1. Public Sector Still Lags Behind Industry in Frontier AI Development
Despite increasing global public sector investment in AI—led by the United States with an investment of $831 million in 2023—the private industry continues to lead in AI model research and development. In 2024, nearly 90% of significant AI models originated from the industry, a lead that has expanded from 60% in 2023.
2. Federal AI Legislation Slows, State-Level Legislation Surges
In the United States, the number of AI-related federal regulations in 2024 more than doubled compared to 2023, with a total of 59 AI-related rules issued by 42 different federal agencies. However, while legislative progress at the federal level has slowed, individual states are forging ahead: the number of state-level AI laws passed in 2024 more than doubled compared to the previous year, whereas federal proposals only increased by 29.2%.
3. Gap Between Open-Source and Closed-Source Models Closing
Competition in the AI field is intensifying as more developers release high-quality models. Over the past year, the performance gap between the first and tenth-ranked models has narrowed from 11.9% to just 5.4%. Furthermore, open-source models are catching up rapidly. By February 2025, the performance gap between open-source models and closed-weight models had narrowed to 1.70%.
4. US Leads in Frontier Models, but US-China Gap Narrows Significantly
The United States remains ahead of China and Europe across several key indicators, serving as the primary source of top-tier AI models (US: 40, China: 15, Europe: 3). However, across multiple important benchmarks, the performance gap between US and Chinese models has shrunk considerably. By the end of 2024, the disparity in model performance between the two nations was generally below 10 percentage points.
5. Technical Leaps Highlight Gaps in Standardized Evaluation
The development of AI is driving researchers to propose increasingly challenging benchmarks, such as "Humanity’s Last Exam," the mathematics benchmark "FrontierMath," and "BigCodeBench." However, research indicates that many benchmarks are poorly designed, highlighting an urgent need for standardized evaluation mechanisms to ensure the reliability of AI assessments and avoid misleading conclusions about model capabilities.
6. China Leads in Volume of Papers; US Dominates High-Impact Research
In 2023, China ranked first globally in both the volume of AI papers published (23.2%) and citations (22.6%). However, US institutions produced the majority of the top 100 most-cited AI papers over the past three years.
7. China Continues to Dominate Industrial Robotics
In 2023, China installed 276,300 industrial robots—six times that of Japan and 7.3 times that of the United States. Since surpassing Japan in 2013 to claim 20.8% of global robot installations, China's market share has risen to 51.1%. China now installs more industrial robots than the rest of the world combined.
8. Fastest Growth in AI Adoption in Greater China
While North America maintains the lead in the extent of AI usage by enterprises and organizations, the Greater China region showed the most significant annual growth in organizational AI adoption rates, increasing by 27 percentage points. Europe followed closely with an increase of 23 percentage points.
9. AI Boosts Efficiency, but Corporate Revenue Gains Remains Modest
Among companies that have adopted AI, 49% reported cost savings in service operations, 43% in supply chain management, and 41% in software engineering. However, most reported cost savings were below 10%. Regarding revenue, 71% of companies applying AI in marketing and sales reported growth, followed by 63% in supply chain management and 57% in service operations; yet, the majority of revenue increases were less than 5%.
10. Training Demands Double Rapidly
Currently, the computational volume required to train key AI models doubles approximately every five months, and the scale of datasets needed to train large language models doubles every eight months. Consequently, the electricity required for model training is rising significantly each year.
11. AI Drives Energy Shift; Nuclear Power Takes Center Stage
To support AI computational demands, Microsoft announced a $1.6 billion investment to restart the Three Mile Island nuclear power plant. Google and Amazon have also secured nuclear power supply agreements, indicating that nuclear energy has re-emerged as a component of the energy strategy for major tech companies.
12. Hardware Becomes Faster, Cheaper, and More Efficient
New research points out that machine learning hardware performance (based on 16-bit floating-point operations) improves by 43% annually, doubling approximately every 1.9 years. Price-performance ratio improves by 30% annually, and energy efficiency increases by 40% per year.
13. AI Outperforms Physicians in Key Clinical Tasks
A new study found that the GPT-4 model alone could outperform human physicians (whether assisted by AI or not) in diagnosing complex clinical cases. Other studies have shown AI performs excellently in cancer detection and identifying high-mortality risk patients.
14. Surge in FDA-Approved AI Medical Devices
Between 1995 (the first approval) and 2015, only six AI medical devices were approved. However, by 2023, the number of approvals had skyrocketed to 223.
15. US Public Remains Highly Distrustful of Autonomous Vehicles
According to a recent survey by the American Automobile Association (AAA), 61% of Americans fear autonomous vehicles, while only 13% express trust. Although this fear has decreased from the 2023 peak (68%), it remains higher than the 54% recorded in 2021.
16. US Teachers Want to Teach AI but Lack Preparation
Although 81% of computer science teachers in the US believe that "the use and understanding of AI" should be included in the foundational computer science curriculum, less than half of high school computer science teachers feel they possess the ability to teach AI.
17. AI Master’s Degrees Nearly Double in the US
While the boom in AI education has not yet shown significant growth at the bachelor's and doctoral levels, the number of people obtaining master's degrees in AI nearly doubled between 2022 and 2023.
18. Workers Expect AI to Change Jobs, Worry Less About Replacement
Globally, 60% of respondents believe AI will change the way people work within the next five years, but only 36% are worried that they will be replaced by AI within that same timeframe.
Read the full article:
SSL 256bit transmission encryption mechanism