Highlights:
⭐️ AI is still superior to humans in many tasks
⭐️ The cost of state-of-the-art AI models continues to rise
⭐️ AI boosts productivity, but there is a lack of regulation
Webmaster's Home (ChinaZ.com) April 24 News: The 2024 Artificial Intelligence Index released by the Human Center Artificial Intelligence Institute of Stanford University reveals eight major trends in artificial intelligence in the business field, covering key issues such as human advantages, costs, regulation, investment growth, and work efficiency improvement, providing important references and enlightenment for enterprises and decision-makers.
1. Humans are still superior to AI in many tasks
Studies have shown that AI is still inferior to humans in complex tasks such as advanced mathematical problem solving, visual common-sense reasoning, and planning. To reach this conclusion, the researchers compared the model to human benchmarks for a number of different business functions, including coding, agent-based behavior, inference, and reinforcement learning.
How AI models perform relative to humans in different tasks. Image: AI Index Report 2024/Stanford University HAI
While it's true that AI surpasses human capabilities in image classification, visual reasoning, and English comprehension, the results suggest that businesses have the potential to leverage AI to accomplish tasks where human employees actually perform better. Many businesses have begun to worry about the consequences of over-reliance on AI products.
2. State-of-the-art AI models are becoming more and more expensive
AI Index reports that by 2023, OpenAI's GPT-4 and Google's Gemini Ultra will cost around $78 million and $191 million to train, respectively. "At the current rate of growth, the cost of cutting-edge AI models will be around $5 billion to $10 billion by 2026, and few companies will be able to afford these trainings," Rahman, a data scientist, told TechRepublic in an email. ”
The cost of training AI models from 2017 to 2023. Image: Artificial Intelligence Index Report 2024/Stanford University HAI/Epoch, 2023
In October 2023, the Wall Street Journal reported that Google, Microsoft, and other big tech companies are struggling to monetize their generative AI products because of the high cost of running them. There is a risk that if the best technologies become so expensive that only large companies have access to them, their advantage over SMEs could increase disproportionately. The World Economic Forum pointed this out back in 2018.
However, Rahman stressed that many of the best AI models are open-source and therefore available to businesses of all budgets, so the technology should not widen any gaps. He told TechRepublic: "Open source and closed-source AI models are growing at the same rate. Meta, one of the largest tech companies, is open-sourcing all of its models, so those who can't afford to train the largest models on their own can download their models. ”
3. Artificial intelligence improves productivity and quality of work
By evaluating some existing research, researchers at Stanford University concluded that AI enables workers to complete tasks faster and improve the quality of their output. Observed occupations include computer programmers (32.8% of whom reported increased productivity), consultants, support agents, and recruiters.
The impact of artificial intelligence on customer support agent productivity. Image: 2024 AI Index Report/HAI, Stanford University/Brynjolfsson et al., 2023
In the case of consultants, the use of GPT-4 bridges the gap between low-skilled and high-skilled professionals, with a greater boost in performance for low-skilled groups. Other studies have also shown that generative AI in particular can act as an equalizer, as less experienced, less skilled workers can get more out of it.
Improve the job performance of low- and high-skilled advisors when using artificial intelligence. Image: AI Index Report 2024/Stanford University HAI
However, other studies do suggest that "the use of AI without proper supervision may lead to a decrease in performance," the researchers wrote. For example, there have been widespread reports of pervasive hallucinations in large language models performing legal tasks. Other studies have found that we may not be able to realize the full potential of the productivity gains that AI can bring over the next decade, as suboptimal output, complex guidelines, and a lack of proficiency continue to hinder worker development.
4. Artificial intelligence regulation in the United States continues to strengthen
The AI Index report found that in 2023, there are 25 AI-related regulations active in the United States, compared to just one in 2016. However, this is not a steady growth trend, with the total number of AI-related regulations increasing by 56.3% from 2022 to 2023 alone. Over time, these regulations have also shifted from expansionary to restrictive to restrictive for the development of AI, with the most prevalent themes being foreign trade and international finance.
The number of AI-related regulations in force in the U.S. between 2016 and 2023. Image: AI Index Report 2024/Stanford University HAI
The EU is also on the rise in AI-related legislation, with 46, 22 and 32 new regulations to be adopted in 2021, 2022 and 2023, respectively. In the region, regulation tends to take a broader approach, often covering science, technology, and communications.
See: NIST Forms AI Security Alliance
For businesses interested in AI, it's essential to stay up-to-date on the regulations that affect them, or they will be at risk of serious penalties for non-compliance and reputational damage. Research published in March 2024 found that only 2% of large companies in the UK and EU are aware of the upcoming EU AI Act.
5. Increasing investment in generative AI
From 2022 to 2023, funding for generative AI products that generate content based on prompts increased almost eightfold to $25.2 billion). OpenAI, Anthropic, Hugging Face, and Inflection, among others, have all received significant funding.
Total global private investment in generative AI from 2019 to 2023. Image: 2024 AI Index Report/Stanford University HAI/Quid, 2023
Generative AI capabilities may be built to meet the needs of businesses looking to incorporate them into their processes. In 2023, 19.7% of all earnings calls for Fortune 500 companies mentioned generative AI, and a McKinsey report revealed that 55% of organizations now use AI in at least one business unit or function, including generative AI.
After the launch of ChatGPT on November 30, 2022, awareness of generative AI grew rapidly, and since then, organizations have been racing to incorporate its capabilities into their products or services. A recent survey of 300 global enterprises conducted by MIT Technology Review Insights in partnership with Telstra International found that respondents expect the number of functions they deploy generative AI to more than double by 2024.
However, according to AI authority Gary Marcus, there is some evidence that the generative AI boom "could end soon" and businesses should remain vigilant. This is mainly due to the limitations of current technology, such as potential bias, copyright issues, and inaccuracies. According to a report from Stanford University, the limited amount of online data available to train models can exacerbate existing problems, limiting improvement and scalability. AI companies could run out of high-quality language data by 2026, low-quality language data in 20 years, and image data in the late 2030s to mid-2040s, the report states.
6. The benchmarks for LLM liability vary widely
According to the report, there are significant differences in the benchmarks used by tech companies to assess the credibility or accountability of their LLMs. This "complicates the work of systematically comparing the risks and limitations of top-tier AI models," the researchers wrote. "These risks include biased output and leakage of private information from training datasets and conversation history.
A responsible AI benchmark for developing popular AI models. Image: AI Index Report 2024/Stanford University HAI
Ruel, a PhD student at Stanford University's Intelligent Systems Lab, told TechRepublic in an email: "There are no reporting requirements at this time, and we don't have reliable assessments that would allow us to confidently say that if a model passes those assessments, then it's safe." "First place. ”
Without standardization in this area, there is an increased risk that some untrustworthy AI models may slip through the net and be integrated by enterprises. The report adds: "Developers may selectively report benchmarks that positively highlight the performance of their models. ”
Reuel told TechRepublic: "There are several reasons why harmful models are slipping through. First, there is no standardized or required assessment, making it difficult to compare models and their (relative) risks; Second, there is no reliable assessment, especially of the underlying model, to have a reliable and comprehensive understanding of the absolute risk of the model. ”
7. Employees are nervous and worried about AI
The report also tracks how attitudes towards AI have changed as awareness has increased. A survey found that 52% of people are nervous about AI products and services, and that number has risen by 13% in 18 months. The study also found that only 54% of adults believe that products and services using AI outweigh the harms, while 36% are concerned that AI could take their jobs in the next five years.
Global perception of the impact of AI on current jobs in 2023. Image: AI Index Report 2024/Stanford University HAI/Ipsos, 2023
Other surveys cited in the AI Index report found that 53% of Americans are currently more worried than excited about AI, with their most prevalent concern being the impact of AI on employment. As AI technology begins to incorporate into organizations, this concern can have a particular impact on employees' mental health, and business leaders should monitor it.
8. Most of today's most popular LLMs were created by the United States and China
TechRepublic's Ben Abbott describes this trend in the Stanford report in his article on building foundational models for AI in the Asia-Pacific region. In part, he wrote:
"U.S. dominance in AI continues throughout 2023. Stanford University's 2024 AI Index report found that 61 noteworthy models were released in the U.S. in 2023; This is ahead of China's 15 new models and France's 8 models, with the largest contributor coming from Europe (Figure 1). The UK and the EU as a region have produced 25 famous models – beating China for the first time since 2019 – while Singapore has 3 models and is the only other country in the Asia-Pacific region to produce well-known large language models. ”
The U.S. is overtaking China and other countries in the development of AI models.