New research shows AI increases worker productivity, but it remains unregulated. His 2024 AI Index report, published by Stanford University's Institute for Human-Centered Artificial Intelligence, reveals the top eight AI trends for enterprises. Among them is that the technology still cannot outperform the human brain at every task.
TechRepublic takes a closer look at how these points impact your business, with insights from report co-authors Robi Rahman and Anka Reuel.
See also: Top 5 AI trends to watch in 2024
1. Humans still outperform AI in many tasks
Research shows that AI is still no better than humans at complex tasks such as advanced mathematical problem solving, visual common sense reasoning, and planning (Diagram A). To reach this conclusion, we compared our model to human benchmarks across a variety of business functions, including coding, agent-based behavior, inference, and reinforcement learning.
Diagram A
Although AI has outperformed humans in image classification, visual reasoning, and English language understanding, this result suggests that companies may leverage AI for tasks where human staff can actually perform better. It shows that there is. Many companies are already concerned about the impact of over-reliance on AI products.
2. Cutting-edge AI models are getting expensive.
AI Index reports that training costs for OpenAI's GPT-4 and Google's Gemini Ultra in 2023 will be approximately $78 million and $191 million, respectively (Diagram B). Rahman, a data scientist, told TechRepublic in an email: “At current growth rates, frontier AI models will cost around $5 billion to $10 billion by 2026. Few companies will be able to afford to perform these trainings at that point. ”
Diagram B
In October 2023, the Wall Street Journal reported that Google, Microsoft, and other major technology companies were struggling to monetize their generative AI products due to the huge costs of operating them. If the best technology becomes so expensive and available only to large companies, the advantage of large companies over small businesses may increase disproportionately. This was pointed out by the World Economic Forum in 2018.
However, Rahman emphasized that this technology will not widen the divide, as many of the best AI models are open source and available to businesses of all budgets. He told TechRepublic: “Open source and closed source AI models are growing at the same rate. One of the largest technology companies, Meta, has open sourced all of its models, so you can build your largest models yourself.” Those who can't afford to train can simply download their own models.
3. AI improves productivity and quality of work
After evaluating a large body of existing research, researchers at Stanford University concluded that AI enables workers to complete tasks faster and improves the quality of their output. Occupations where this was observed include computer programmers, consultants, and support agents, where 32.8% reported increased productivity (Figure C) and recruiters.
Figure C
For consultants, the use of GPT-4 bridged the gap between low-skilled and high-skilled experts, further improving the performance of the low-skilled group (Figure D). Other research has shown how generative AI may act as an equalizer, especially as generative AI can benefit more workers with less experience and lower skills. .
Figure D
However, other studies suggest that “using AI without proper supervision can lead to poor performance,” the researchers wrote. For example, hallucinations are widely reported to be prevalent in large language models performing legal tasks. Other research shows that unsatisfactory performance outcomes, complex guidelines, and lack of proficiency continue to hold workers back from realizing AI's full productivity potential. We know it could take another 10 years.
4. AI regulations in the US are on the rise
According to the AI Index Report, there were 25 AI-related regulations in place in the U.S. in 2023, compared to just one in 2016 (Figure E). However, this trend is not constant, as the total number of AI-related regulations increased by 56.3% from 2022 to 2023 alone. Over time, these regulations have also changed from broad to restrictive regarding advances in AI, with the most common subjects they touch on being foreign trade and international finance.
Figure E
The EU is also seeing an increase in AI-related legislation, with 46, 22, and 32 new regulations passed in 2021, 2022, and 2023, respectively. In this region, regulation tends to take a broader approach, most often covering science, technology, and communications.
See also: NIST establishes AI Safety Consortium
For companies interested in AI, it's essential to stay up-to-date on the regulations that affect them. Otherwise, you risk heavy penalties and reputational damage for non-compliance. A study published in March 2024 found that only 2% of large UK and EU companies were aware of the upcoming EU AI law.
5. Increased investment in generative AI
Funding for generative AI products that generate content in response to prompts increased nearly eightfold from 2022 to 2023, reaching $25.2 billion (Figure F). OpenAI, Anthropic, Hugging Face, Inflection, and more have all raised large funding rounds.
Figure F
Building generative AI capabilities could meet the demands of companies looking to implement it into their processes. By 2023, 19.7% of all financial statements for Fortune 500 companies will cite generated AI, and McKinsey reports that 55% of organizations now use AI, including generated AI, in at least one business unit or department. It became clear that it was.
Generative AI gained recognition after ChatGPT was announced on November 30, 2022, and organizations have been racing to incorporate its capabilities into their products and services ever since. A recent survey of 300 companies worldwide conducted by MIT Technology Review Insights in collaboration with Telstra International found that respondents expect to more than double the number of departments implementing generative AI by 2024. I found out that I expected it to be.
See also: Defining generative AI: How it works, its benefits, and its dangers
But according to AI guru Gary Marcus, there is some evidence that the generative AI boom “could be coming to an end pretty quickly” and companies need to be cautious. This is primarily due to the limitations of current technology, such as bias, copyright issues, and possible inaccuracies. According to the Stanford University report, the limited amount of online data available for training models can exacerbate existing problems and puts a cap on improvement and scalability. AI companies say they could run out of high-quality language data by 2026, low-quality language data within 20 years, and image data by the late 2030s to mid-2040s.
6. Benchmarks for LLM responsibilities vary widely
According to the report, there is wide variation in the benchmarks by which technology companies evaluate their LLMs for reliability and accountability (Figure G). The researchers wrote that this “complicates efforts to systematically compare the risks and limitations of top AI models.” These risks include biased output and leaking personal information from training datasets and conversation history.
Figure G
Reuel, a doctoral student at Stanford University's Institute for Intelligent Systems, told TechRepublic in an email. first place. ”
Without standardization in this area, there is an increased risk that some untrustworthy AI models will slip through the cracks and be integrated into the enterprise. “Developers may selectively report benchmarks that positively emphasize model performance,” the report added.
Reuel told TechRepublic: “There are multiple reasons why harmful models slip through the cracks. First, there is no standardized or mandatory evaluation, making it difficult to compare models and their (relative) risks. Second, the absolute There is a lack of robust assessments, particularly of the underlying models, that allow for a robust and comprehensive understanding of the risks involved.”
7. Employees are nervous and concerned about AI
The report also tracked how attitudes toward AI are changing as awareness of it increases. According to one survey, 52% feel uneasy about AI products and services, and this figure has increased by 13% in his 18 months. Additionally, only 54% of adults agree that products and services using AI have more benefits than drawbacks, and 36% do not expect to lose their job within the next five years. I understand that you are concerned about this.Figure H).
Figure H
Other research referenced in the AI Index report shows that 53% of Americans currently feel more anxious than excited about AI, and the most common concern they have is the impact on jobs. got it. As AI technology begins to be integrated into organizations, such concerns can particularly impact the mental health of employees, and business leaders will need to monitor this.
See also: 10 Best AI Courses of 2024
8. Most of the current popular LLMs are created by the US and China.
TechRepublic's Ben Abbott highlighted this trend from the Stanford University report in his article on building AI-based models in the APAC region. He writes in part:
“The U.S. lead in AI continued throughout 2023. The 2024 Stanford AI Index report found that 61 notable models were released in the U.S. in 2023. This exceeded China's 15 new models and France's eight models, the largest European contributor (Figure I). The UK and the European Union produced 25 notable models as a region, surpassing China for the first time since 2019, but APAC is the only other country producing notable large-scale language models with 3 Only Singapore had two models. ”
Figure I
