Inside The New AI Index: Expensive New Models, Targeted Investments, and More
![illustration of bright colors and intersecting lines](/https/hai.stanford.edu/sites/default/files/styles/media/public/2024-04/AIIndex_2024_WebGraphics_1200x628.jpg?itok=C47c-7NS)
A decade ago, the world's best AI systems were incapable of classifying images at a human level. They could not understand language, struggled with visual reasoning, and flunked the most basic reading comprehension tests. Today, AI systems routinely exceed human performance on standard benchmarks.
This is just one of the new findings of the 2024 AI Index, an annual study of trends in AI from the Stanford Institute for Human-Centered Artificial Intelligence (HAI). The AI Index, led by an independent and interdisciplinary group of AI thought leaders from industry and academia, is one of the most comprehensive annual reports on progress in AI; it tracks trends in research and development, technical performance, responsible AI, economics, policy, public opinion, and more. The report is grounded in quantitative data and designed to put key decision-makers, like policy leaders and business executives, in a position to best understand what is occurring in the world of AI.
Read the 2024 AI Index Report
Although AI was already showing exciting new capabilities in 2022, in 2023 the technology accelerated. The kinds of AI systems that were released a few years ago were limited in scope. For example, these models could process images but not generate text, and vice versa. However, recent models like Gemini, GPT-4, and Claude-3 have been impressively multimodal. They can generate fluent text in dozens of languages, process audio, and even explain memes.
"This year, we see more models able to perform across domains," said Vanessa Parli, Stanford HAI director of research programs. "Models can take in text and generate audio or take in an image and generate a description. An edge of the AI research I find most exciting is combining these large language models with robotics or autonomous agents, marking a significant step in robots working more effectively in the real world."
Industry Invests in AI
Business leaders have certainly paid attention. Global private investment in generative AI skyrocketed, increasing from roughly $3 billion in 2022 to $25 billion in 2023. Nearly 80 percent of Fortune 500 earnings calls mentioned AI, more than ever before. Moreover, there is robust data emerging from academic economists that suggests AI tangibly boosts worker productivity and can especially help low-skilled workers.
Industry also continues to lead in AI. In 2023, technology companies produced 51 significant machine learning systems, while academia produced only 15. Moreover, while 108 newly released foundation models came from industry, only 28 originated from academia.
Regulation Ramps Up
Policymakers are also responding. There were 2,175 mentions of AI in global legislative proceedings in 2023, nearly double the number from the previous year. U.S. regulators passed 25 AI-related regulations in the last year, a new record. Some of these regulations included copyright guidance for generative AI material and cybersecurity risk management frameworks. More broadly, 2023 was a banner year for major AI policy; the EU proposed its comprehensive AI Act and President Biden unveiled the Executive Order on AI.
Public Sentiment Dips Negative
The general public took notice of AI, too, and responded with nervousness. Survey data from Pew suggests that in 2022 only 38 percent of Americans reported feeling more concerned than excited about AI technology. In 2023, that figure rose to 52 percent. Across the globe, 52 percent of surveyed respondents felt nervous about AI, a 13 percentage-point increase from the previous year.
AI Gets More Expensive
Model training costs, as first reported in last year’s AI Index report, also continued climbing. New estimates suggest that certain frontier systems, like OpenAI’s GPT-4, cost $78 million to train. Google Gemini’s price tag came in at $191 million. By comparison, some state-of-the-art models released half a decade or so ago, namely the original transformer model (2017) and RoBERTa Large (2019), respectively cost around $900 and $160,000 to train.
Worldwide, America Dominates
Finally, from a global perspective, the U.S. continued to lead in AI. In 2023, substantially more significant AI models (61) came from U.S.-based institutions, compared with the European Union (21) and China (15). The U.S. also remains the premier location for AI investing. A total of $67.2 billion was privately invested in AI in the U.S. in 2023, nearly nine times more than the amount in China.
The new index shows China is America’s biggest competitor. China leads the world in robotic installations, having installed more robots in 2023 than the rest of the world combined. Likewise, the majority of the world’s AI patents (61 percent) originated from China.
Ultimately, the 2024 AI Index report makes clear that this technology continues to progress rapidly and the world including business leaders, policymakers, and the general public is beginning to react.
Nestor Maslej is the research manager and editor-in-chief of the AI Index.
The AI Index was first created to track AI development. The index collaborates with organizations such as LinkedIn, Quid, McKinsey, Studyportals, the Schwartz Reisman Institute, and the International Federation of Robotics to gather the most current research and feature important insights on the AI ecosystem.