In recent years, with the continuous development of artificial intelligence (AI) technology, the presence of AI has become increasingly prominent in various fields. Chatbot models like ChatGPT have gained global popularity, while AI-generated artwork has sparked ongoing debates.
How should we interpret this phenomenon, and what kind of future lies ahead for this hot field? Let’s take a look at the analysis provided by Stanford University.
In April, Stanford University’s Human-Centered Artificial Intelligence Institute (Stanford HAI) released the “2023 Artificial Intelligence Index Report.” Since 2017, a group of experts and professors from renowned institutions such as MIT, OpenAI (the developer of ChatGPT), Harvard, and McKinsey has come together to observe the development and future trends of this emerging technological field from various perspectives, including research and development, technology, ethics, economy, education, and governance.
The latest report reveals several key trends in the AI industry in 2022:
1. Industry Leads in Technology, but Training Costs are Increasing.
Previously, most significant machine learning models were released by the academic community. However, since 2014, the industry has taken the lead, with 32 models produced in 2022 compared to only three from the academic sector. Despite industry dominance, the report highlights a growing concern: the rising cost of training these models.
This change stems from the increasing need for large amounts of data, computing power, and funding to build state-of-the-art artificial intelligence systems, which is a natural advantage of the industry.
The Size and Cost of Large Language Models are Growing.
Many consider GPT-2, released in 2019, as the first large language model with 1.5 billion parameters, estimated to cost $50,000 to train. PaLM, one of the flagship large language models introduced in 2022, has a staggering 540 billion parameters, estimated to cost $8 million. The scale of PaLM in 2022 is 360 times larger than GPT-2 in 2019, with costs 160 times higher.
The significant increase in these numbers highlights the ever-increasing cost of training large language models.
2. China Rapidly Developing in the Field of AI, with a Considerable Number of Research Papers.
According to the report’s data, in 2021, the Chinese Academy of Sciences published the highest number of papers in the field of AI, with MIT ranking tenth.
Carnegie Mellon University (CMU) ranks second in the field of Natural Language Processing (NLP).
3. Limited Progress in Traditional Benchmark Tests.
Despite the continuous development of artificial intelligence technology, improvements in many benchmark tests have been minimal. As a result, new and more comprehensive benchmark test suites, such as BIG Bench and HELM, are set to be released soon.
4. AI as a Double-Edged Sword for the Environment.
The training of large AI models has an impact on the environment. Scientific research data indicates that in 2022, the carbon emissions from training BLOOM (an open-source language AI model developed by BigScience) were 25 times greater than the carbon emissions from a person flying from New York to San Francisco.
Nevertheless, new reinforcement learning models like BCOOLER suggest that AI systems can be utilized to optimize energy consumption.
5. Rapid Increase in Incidents of AI Abuse.
According to the AIAAIC database tracking incidents related to the ethical misuse of AI, the number of AI-related controversial events has increased 26-fold since 2012. This growth demonstrates both the widespread utilization of AI and the potential for its misuse.
Since the introduction of ChatGPT, due to concerns regarding its potential impact on academic integrity, an increasing number of universities have issued statements prohibiting or strictly limiting student use of ChatGPT for paper writing.
6. Increasing Demand for AI Talent and Emergence of New Job Positions.
Research indicates a growing demand for AI skills across various fields, leading to a significant increase in the number of related job positions compared to 2021.
7. Significant Decline in Investment, but Still Nearly Twenty Times Higher Than a Decade Ago.
For the first time in the past decade, investments in the AI sector have experienced a year-on-year decline.
In 2022, global AI investment amounted to $91.9 billion, a decrease of 26.7% compared to 2021. However, despite the decline, the investment in 2022 was still nearly 18 times higher than that of 2013.
8. Chinese Citizens Exhibit the Most Open Attitude towards AI Products.
According to an IPSOS survey conducted in 2022, 78% of Chinese respondents (the highest proportion among the surveyed countries) agreed with the statement that the benefits of using AI products and services outweigh the drawbacks. The countries with the next highest agreement rates were Saudi Arabia (76%) and India (71%). In contrast, only 35% of the sampled Americans (the lowest among the surveyed countries) believed that the benefits of using AI products and services outweigh the drawbacks.
As the AI era approaches, Stanford University’s AI report provides authoritative data support for understanding the current trends in the AI field. What are your thoughts on AI? Do you have an AI dream as well? Feel free to share your opinions in the comments section.
