Stanford HAI Index highlights transformative developments in artificial intelligence that carry profound implications for societies worldwide—especially in regions across the Global South [1]. As we explore these insights, we recognize that AI is transforming industries, creating new opportunities, and driving economic growth. There are extraordinary opportunities that AI presents and a shared responsibility to ensure its benefits are accessible.
A Steep Drop in Costs and Barriers
One of the most remarkable shifts has been the dramatic reduction in the cost of AI model usage. The cost of querying an AI model that scores the equivalent of GPT-3.5 fell from $20 per million tokens in late 2022 to just $0.07 by late 2024. This more than 99% decrease is not merely a technical milestone—it’s a gateway to access. Innovators and entrepreneurs in low-resource regions can now harness powerful tools once restricted to the world’s largest companies, applying them to local challenges in healthcare, agriculture, education, and public service.
Closing the Performance Gap
The gap between open-weight and proprietary closed-weight models has also narrowed significantly. By 2024, open-weight models rival their commercial counterparts, fueling competition and innovation across the ecosystem. In parallel, the performance gap between the top frontier models has also compressed. Smaller models are achieving results once thought exclusive to massive-scale systems—Microsoft’s Phi-3-mini, for instance, delivers performance comparable to models 142 times larger, bringing powerful AI within reach of environments with constrained resources.
Persistent Challenges: Reasoning and Data
Yet challenges remain. Despite advances, AI systems still struggle with higher-order reasoning, such as arithmetic and strategic planning—capabilities that are essential in domains where reliability is critical. Continued research and responsible application are essential to overcome these limitations.
Another emerging concern is the rapid reduction of publicly available data used to train AI models. As websites increasingly restrict data scraping, model performance and generalizability may suffer—especially in contexts where labeled datasets are already limited. This trend may necessitate new learning approaches tailored to data-constrained environments.
Real-World Impact on Productivity and Workforce
Perhaps the most exciting development is AI’s tangible impact on human productivity. Last year’s AI Index was among the first to highlight research showing that AI meaningfully improves productivity. This year, follow-up studies confirmed and expanded those findings—especially in real-world workplace environments.
One such study tracked over 5,000 customer support agents using a generative AI assistant [2]. The tool increased productivity by 15%, with the most significant improvements seen among less experienced workers and skilled trade workers, who also boosted the quality of their work. Additionally, AI assistance helped employees learn on the job, improving English fluency among international agents, and even enhanced the work environment—customers were more polite and less likely to escalate issues when AI was involved.
Complementing these findings, Microsoft’s internal research initiative on AI and productivity compiled results from over a dozen workplace studies, including the largest known randomized controlled trial of generative AI integration[3]. Tools like Microsoft Copilot are already enabling workers to complete tasks more efficiently across roles and industries. The research underscores that the impact of AI is greatest when tools are adopted and integrated strategically—and that the potential will only grow as organizations recalibrate workflows to take full advantage of these new capabilities.
Expanding Access to Computer Science Education
As AI becomes more integrated into daily life, computer science education is more essential than ever. Encouragingly, two-thirds of countries now offer or plan to offer K–12 CS education, a figure that has doubled since 2019. African and Latin American countries have made some of the most significant strides in expanding access. However, the benefits of this progress are not yet universal—many students across Africa still lack access to computer science education due to basic infrastructure gaps, including lack of electricity in schools. Closing this digital divide is essential to preparing the next generation to not only use AI, but to shape it.
Our Shared Responsibility
At Microsoft, we view this moment as a significant inflection point—one that calls for thoughtful action as much as innovation. The rapid progress in AI brings enormous potential to improve productivity, solve real-world challenges, and drive economic growth. But realizing that potential requires continued investment in robust infrastructure, high-quality education, and responsible deployment of AI technologies.
To make the most of this moment, we need to support workers with learning new skills and tools to apply AI effectively in their jobs. Nations and businesses that invest in AI skilling will foster innovation and open doors to more people to build meaningful careers that contribute to a stronger economy. The goal is clear: to turn technical breakthroughs into practical impact at scale.
[1] “AI Index | Stanford HAI.” Accessed: Apr. 05, 2025. [Online]. Available: https://hai.stanford.edu/ai-index
[2] E. Brynjolfsson, D. Li, and L. Raymond, “Generative AI at Work*,” The Quarterly Journal of Economics, p. qjae044, Feb. 2025, doi: 10.1093/qje/qjae044.
[3] S. Jaffe et al, “Generative AI in Real-World Workplaces,” Jul. 2024, Accessed: Apr. 05, 2025. [Online]. Available: https://www.microsoft.com/en-us/research/publication/generative-ai-in-real-world-workplaces/