- The Stanford report highlights that AI excels in English language understanding, image classification, and visual thinking, but still lags behind humans in advanced mathematics, logical thinking, and planning.
- Training AI models is becoming increasingly expensive, with costs reaching tens of millions of dollars, yet the integration of AI in workplaces can enhance productivity and bridge educational gaps among employees.
- There is growing public nervousness about AI, reflected in the increase of AI-related regulations in the U.S., from just one in 2016 to 25 in 2023.
A comprehensive report from Stanford University delves into the capabilities of artificial intelligence and how they measure up against human skills, offering a nuanced view of both the strengths and weaknesses of AI technology.
Titled “Artificial Intelligence Index Report,” the annual release by the Stanford Institute for Human-Centered AI, now in its seventh edition, is the largest yet in terms of size, scope, and reach. The report, spread across 458 pages, provides a detailed assessment of 51 machine learning models launched by companies, 15 by research institutions, and 21 developed through cooperative efforts in the year 2023.
The analysis within the report focuses on several key areas where AI has proven to be superior to human abilities, such as language understanding of English, image classification, and visual thinking — making thought processes both comprehensible and visible. However, the report also highlights areas where humans continue to outperform AI, particularly in tasks that involve advanced mathematics, logical thinking, and planning.
These findings were gathered from various evaluations and studies carried out on the latest models available, which are discussed in detail in the report, specifically in Chapter 2 under “Technical Performance.”
According to the report, while AI shows exceptional prowess in certain domains, humans still excel in more complex tasks, which involve higher-order cognitive capabilities. This ongoing competition between man and machine underscores the evolving landscape of AI technology.
The report also touches upon the economic aspects of AI development, noting the increasing costs associated with training AI models. For instance, OpenAI’s GPT-4 model required an estimated $78 million in computing power for its training, while Google’s Gemini Ultra model consumed $191 million worth of computing resources.
Moreover, the report suggests that integrating AI into workplaces can be highly beneficial. It emphasizes that AI not only boosts productivity and improves the quality of work but also helps bridge the gap between varying levels of employee education, making a compelling case for the potential symbiosis between humans and machines.
Public perception of AI is also covered in the report, with data from a 2023 survey showing that 52 percent of respondents now feel nervous about AI applications, up from 39 percent the previous year. This growing nervousness is likely influencing the regulatory landscape, as evidenced by an increase in AI-related regulations in the USA — from just one in 2016 to 25 in 2023.
