Introduction and Overview
Industries and societies witness transformative waves as AI advances into its deployment era. The article talks about the AI Index Report, which thoroughly explores the changing world of artificial intelligence. Last year’s report was comprehensive. It covered AI public opinion, detailed technical performance, large language and multimodal models, global AI legislation trends, and the environmental impact of AI systems.
The AI Index Report serves as a compass in the dynamic realm of artificial intelligence, meticulously tracking, collating, and visualizing data to provide unbiased, rigorously vetted, and broadly sourced information. Tailored for policymakers, researchers, executives, journalists, and the general public, its mission is to foster a nuanced understanding of the intricate field of AI. Positioned as the world’s most credible and authoritative source, this report illuminates the complex journey of AI, offering insights to guide decision-making at the highest echelons.
As we explore the report in depth, co-directors Jack Clark and Ray Perrault beckon us into the era of AI deployment. Throughout 2022 and the early months of 2023, monumental AI models like ChatGPT, Stable Diffusion, Whisper, and DALL-E 2 have emerged, showcasing unprecedented capabilities. From text manipulation to image generation and speech recognition, these models push the boundaries of what was conceivable a decade ago. Yet, they bring to the fore ethical challenges – prone to hallucination, bias, and susceptibility to serving unintended purposes.
While 2022 witnessed a decline in private AI investment for the first time in a decade, AI remains a focal point for policymakers, industry leaders, researchers, and the public. The report sheds light on the increasing role of a select group of private sector actors in shaping the AI landscape. It prompts us to reflect on the development and deployment of AI, considering its technological advancements and societal impact.
Trends in AI Publications and Significant Machine Learning Systems
- Global Collaboration Pivot: From 2010 to 2021, the U.S. and China quadrupled their cross-country collaborations in AI. However, 2021 saw a slowdown, marking the smallest YoY growth since 2010.
- Exponential Research Surge: AI research has more than doubled since 2010, with pivotal domains like pattern recognition, machine learning, and computer vision dominating the landscape.
- China’s Continued Dominance: China leads in total AI publications across journals, conferences, and repositories. The U.S. maintains its lead in AI citations but faces a gradual erosion of this advantage.
- Industry Takes the Lead: In 2022, the industry surpassed academia in producing significant machine learning models, marking a significant shift in the AI development landscape.
- Bigger and Pricier Models: Large language models, essential to AI advancements, are growing exponentially in size and cost. The flagship PaLM model in 2022 is 360 times larger and 160 times costlier than its 2019 counterpart, GPT-2.
Analysis of Cross-Country and Cross-Sector Collaborations in AI
A meticulous analysis of AI research and development reveals intriguing patterns in cross-country and cross-sector collaborations. From 2010 to 2021, the United States and China emerged as frontrunners in cross-country collaborations, witnessing a remarkable fourfold increase in collaborative efforts. However, the collaboration pace slowed in 2021, marking the smallest year-over-year growth rate since 2010.
The global surge in AI research is undeniable, with publications doubling since 2010. Key research areas, including pattern recognition, machine learning, and computer vision, continue to dominate the landscape.
China maintains its leadership in total AI journal, conference, and repository publications. While the United States remains at the forefront in AI conferences and repository citations, these leads are gradually diminishing. American institutions still produce the majority (54% in 2022) of the world’s large language and multimodal models, underscoring their continued influence.
The dynamics between industry and academia have undergone a significant shift. In 2022, the industry outpaced academia in producing noteworthy machine learning models, a departure from the trend up to 2014. The resources needed for cutting-edge AI systems, such as extensive data, computational power, and funding, now gravitate toward industry actors.
Large language models, the bedrock of AI advancements, are scaling in size and cost. For instance, PaLM, a flagship model in 2022, boasts 540 billion parameters and costs approximately $8 million USD—360 times larger and 160 times costlier than its 2019 counterpart, GPT-2. This pervasive trend indicates a broader shift towards larger and more expensive models.
Technical Performance of AI
AI consistently achieves state-of-the-art results, yet the year-over-year improvement on traditional benchmarks remains modest. The acceleration of benchmark saturation is evident, prompting the introduction of more comprehensive benchmarking suites like BIG-bench and HELM and providing a nuanced understanding of AI capabilities.
Challenging the traditional narrow-task paradigm, recent models like BEiT-3, PaLI, and Gato showcase the evolving landscape of AI. These single systems demonstrate increased adaptability, seamlessly navigating multiple tasks, particularly in the intersection of vision and language. Despite continual enhancements in generative capabilities, language models encounter difficulties in complex planning tasks. The ongoing struggle to integrate robust reasoning capabilities underscores a persistent challenge in AI development. AI is not merely advancing but actively contributing to its own progress. Instances like Nvidia using an AI reinforcement learning agent to enhance chip design and Google’s PaLM suggesting improvements to itself highlight the emerging era of self-improving AI learning, promising accelerated progress.
Generative AI and Its Environmental Impact
- Breakthroughs in Generative AI: The year 2022 witnessed the breakthrough of generative AI into public awareness. Innovations like DALL-E 2, Stable Diffusion, Make-A-Video, and ChatGPT marked significant strides in text-to-image, text-to-video, and chatbot capabilities.
- Limitations and Challenges in Generative AI: While generative AI systems showcase impressive capabilities, they are not without challenges. Prone to hallucination, these systems may confidently output incoherent or untrue responses, posing challenges for critical applications that require reliability.
- Environmental Impacts of AI: AI assumes a dual role in environmental impact. On the one hand, research reveals significant carbon emissions during training runs, as exemplified by BLOOM emitting 25 times more carbon than a one-way flight from New York to San Francisco. On the other hand, reinforcement learning models like BCOOLER show promise in optimizing energy usage, presenting a potential avenue for mitigating environmental concerns associated with AI.
- AI as a Catalyst for Scientific Progress: AI models are increasingly becoming indispensable contributors to scientific progress. In 2022, these models played pivotal roles in advancing hydrogen fusion, improving matrix manipulation efficiency, and generating new antibodies, demonstrating AI’s profound impact on scientific innovation.
Read more: Gen AI hazards that Every CISO Should Know in 2024
Ethical Considerations in AI
Embarking on exploring ethical dimensions in AI, confront the intricate interplay of bias, toxicity, and ethical challenges accompanying the technological strides in artificial intelligence.
Technical AI Ethics
 The scale of AI models introduces complexities in addressing bias and toxicity, exacerbated by training data and mitigation methods. Despite the challenges, recent evidence indicates that toxicity and bias can be partially alleviated post-training, especially with larger models employing instruction-tuning.
Generative Models and Ethical Quandaries
The year 2022 marked the ascension of generative models into the mainstream, accompanied by a wave of ethical challenges. Text-to-image generators, in particular, display routine biases along gender dimensions, while advanced chatbots like ChatGPT can be susceptible to manipulation for nefarious purposes.
Rising Tide of AI Misuse Incidents
Incidents involving the misuse of AI are on a sharp upward trajectory. According to the AIAAIC database tracking ethical misuse, incidents and controversies have surged 26 times since 2012. Notable instances in 2022, such as a deepfake video featuring Ukrainian President Volodymyr Zelenskyy and call-monitoring technology in U.S. prisons, underscore the dual facets of increased AI usage and growing awareness of misuse possibilities.
Fairness Paradox in Models
While pursuing fairness in AI models is commendable, extensive analysis reveals a nuanced challenge. Despite a clear correlation between performance and fairness, fairer models may not necessarily be less biased. Language models excelling in fairness benchmarks often exhibit worse gender bias, highlighting the complex interplay between fairness and bias.
Soaring Interest in AI Ethics
The surge in interest in AI ethics is palpable. The number of accepted submissions to FAccT, a leading AI ethics conference, has more than doubled since 2021 and multiplied by a factor of 10 since 2018. The year 2022 witnessed a record number of submissions from industry actors, signaling a collective acknowledgment of the importance of ethical considerations in AI.
Challenges in Automated Fact-Checking
The promise of automated fact-checking through natural language processing encounters hurdles. Despite several benchmarks, researchers reveal that a significant portion of datasets relies on evidence “leaked” from fact-checking reports that did not exist at the time of the claim surfacing. This underscores the complexities inherent in achieving reliable automated fact-checking.
AI’s Economic Impact
The American workforce is experiencing a seismic shift in the demand for AI-related professional skills. Across diverse sectors, AI-related job postings have risen from 1.7% in 2021 to 1.9% in 2022, reflecting a pervasive need for workers equipped with AI expertise. Employers in the United States seek individuals with a nuanced understanding of AI applications.
While 2022 witnessed a year-over-year decrease in global private investment in AI, the overall trajectory of the last decade showcases a substantial surge. With $91.9 billion invested globally, representing a 26.7% decrease from 2021, the decade saw an 18-fold increase in private investment in AI since 2013. Once again, the United States stands at the forefront of AI investment, leading the world with $47.4 billion invested in 2022. This amount is approximately 3.5 times higher than the investment in the next-highest country, China ($13.4 billion). The U.S. surpasses in total investment and leads in the number of newly funded AI companies, demonstrating a robust and dynamic AI ecosystem.
In 2022, specific AI focus areas commanded significant investment, with medical and healthcare leading at $6.1 billion, followed by data management, processing, and cloud at $5.9 billion, and Fintech at $5.5 billion. However, mirroring the broader trend, most AI focus areas saw decreased investment in 2022 compared to the previous year. Notable investment events include funding for GAC Aion New Energy Automobile, Anduril Industries, and Celonis, showcasing the diverse applications of AI across industries.
While the proportion of companies adopting AI has plateaued between 50% and 60%, those embracing AI continue to surge ahead. McKinsey’s annual research survey reveals that organizations incorporating AI report substantial cost decreases and revenue increases. The proportion of adopting companies has more than doubled since 2017, underscoring the tangible benefits experienced by early AI adopters.
Read more: Future of AI in Data Integration in Digital Businesses
Educational Trends in AI
In higher education, the proportion of new computer science PhD graduates specializing in AI jumped to 19.1% in 2021, reflecting a significant rise from 14.9% in 2020 and 10.2% in 2010. This shift underscores the growing importance of AI-focused education paths. Furthermore, the career trajectory of AI PhDs has notably tilted towards industry roles, with 65.4% opting for corporate positions in 2021, more than double the 28.2% who chose academia, signaling an increasing demand for AI expertise in the business sector.
Within North American computer science, computer engineering, and information faculties, the past decade has seen a consistent trend in faculty hires, with 710 new hires in 2021, slightly down from 733 in 2012. The dynamics of tenure-track hires peaked at 422 in 2019, followed by a decline to 324 in 2021, illustrating a nuanced pattern in faculty recruitment. Concurrently, the disparity in external research funding between private and public American CS departments continues to widen, with private universities securing substantially more funding in 2021—$9.7 million compared to $5.7 million for public universities.
As the educational landscape adapts to the evolving demands of the AI era, the increasing interest in K–12 AI and computer science education globally further solidifies the role of AI in shaping the knowledge landscape. In 2021, American students took 181,040 AP computer science exams, reflecting a 1.0% increase from the previous year and a remarkable ninefold surge since 2007. Additionally, 11 countries, including Belgium, China, and South Korea, have officially embraced K–12 AI curricula, marking a global acknowledgment of the significance of AI education at an early stage.
Diversity in North American Computer Science Education: A Snapshot
- Ethnically diverse representation is rising among bachelor’s, master’s, and PhD-level computer science students. While white students remain the majority, the percentage dropped from 71.9% in 2011 to 46.7% in 2021, reflecting increased representation from Asian, Hispanic, and Black or African American students.
- Despite growth in diversity, new AI PhDs still exhibit a significant gender imbalance. In 2021, 78.7% were male, with only 21.3% female—a modest increase from 2011. Gender disparity persists in advanced AI education.
- Women are making strides in CS, CE, and information faculty hires. Since 2017, the proportion of new female faculty hires in these fields increased from 24.9% to 30.2%. However, most faculty (75.9%) in North American universities remain male, with only 0.1% identifying as nonbinary as of 2021.
- K -12 computer science education in the U.S. is becoming more diverse in terms of both gender and ethnicity. The share of female students taking AP computer science exams grew from 16.8% in 2007 to 30.6% in 2021. Additionally, there’s an increasing presence of Asian, Hispanic/Latino/Latina, and Black/African American students in AP computer science year over year.
Policy and Governance in AI
- Global interest in AI among policymakers is evident, with AI-related bills passed into law increasing from 1 in 2016 to 37 in 2022 across 127 countries.
- In parliamentary proceedings across 81 countries, mentions of AI have surged nearly 6.5 times since 2016.
- The U.S. has seen a substantial shift from AI discussions to legislative action. In 2022, federal AI bills enacted into law increased from 2% in 2021 to 10%, with 35% of state-level AI bills also being passed.
- Policymakers globally approach AI from diverse angles. In 2022, discussions spanned AI-led automation risks in the U.K., human rights protection in AI in Japan, and exploring AI for weather forecasting in Zambia.
- The U.S. government has significantly increased AI-related contract spending approximately 2.5 times since 2017.
- The legal landscape is adapting to AI, with 110 AI-related legal cases in U.S. state and federal courts in 2022—seven times more than in 2016. Cases predominantly originated in California, New York, and Illinois, focusing on civil, intellectual property, and contract law related to AI.
Public Opinion on AI
Global Perspectives on AI and Societal Views
Regional Variations: Chinese citizens exhibit the most positive sentiments toward AI, with 78% perceiving more benefits than drawbacks. Saudi Arabia (76%) and India (71%) also express optimism. Conversely, only 35% of Americans share this positive sentiment, marking one of the lowest proportions globally.
Skepticism on Self-Driving Cars: Only 27% of respondents feel safe in self-driving cars worldwide. In the U.S., just 26% believe driverless passenger vehicles benefit society. This global skepticism, especially in America, challenges the widespread acceptance of autonomous vehicles.
Varied Excitement and Concerns: Among Americans excited about AI, 31% focus on its potential to enhance life and society, while 13% emphasize time-saving efficiency. Concerns, on the other hand, center around job loss (19%), surveillance, hacking, and digital privacy (16%), and the perceived decline in human connection (12%).
Gender Differences in AI Perception
Positive Outlooks and Trust: Men generally hold more positive views on AI, believing it will mostly help rather than harm. Survey data indicates that men are likelier to find AI products beneficial, trust AI-using companies, and perceive more advantages than drawbacks. This trend is consistent across various surveys, highlighting a gender divide in AI perception.
Concerns and Skepticism: Women, compared to men, are less optimistic about AI, with a smaller percentage believing it will mostly help. A 2021 survey reveals a gender imbalance in higher-level AI education, with most new AI PhDs still male (78.7%). These findings underscore the need for addressing gender disparities in AI perceptions and education.
Conclusion
Industry dominance has emerged, with a significant shift from academia to industry-produced machine learning models, reflecting the greater resources possessed by corporate entities. Despite AI’s consistent state-of-the-art performance, the pace of improvement on traditional benchmarks is slowing, prompting the development of more comprehensive benchmarking suites like BIG-bench and HELM. The environmental impact of AI is a growing concern, as highlighted by studies revealing substantial carbon emissions from training runs. On a positive note, AI is becoming a catalyst for scientific progress, contributing to advancements in diverse fields such as hydrogen fusion, matrix manipulation efficiency, and antibody generation. Concurrently, incidents related to AI misuse are rising, emphasizing the need for ethical considerations. The demand for AI-related skills is soaring across American industries, although private investment in AI will decrease in 2022. However, companies already adopting AI continue demonstrating substantial benefits, as reflected in cost decreases and revenue increases. Policymakers’ interest in AI is escalating globally, with a significant increase in AI-related bills passed into law. Notably, there are substantial global variations in public perceptions of AI, with Chinese citizens expressing the most positive views while Americans exhibit a more skeptical stance. These top ten takeaways collectively depict the multifaceted and evolving dynamics of AI’s influence on various aspects of society, technology, and governance.
FAQs
- What is the AI Index Report 2023?
The AI Index Report 2023 is the sixth edition of an annual report that tracks, collates, distills, and visualizes data related to artificial intelligence. It aims to provide a comprehensive and nuanced understanding of AI for policymakers, researchers, executives, journalists, and the public. - What are some key trends in AI research and development as of 2023?
Key trends include the dominance of industry over academia in producing significant machine learning models, a rapid increase in AI publications, and the significant role of cross-country collaborations, especially between the United States and China. - How has the performance of AI systems evolved?
While AI continues to achieve state-of-the-art results, there is a saturation in performance improvements on many benchmarks. However, new and more comprehensive benchmarks are being developed to push the boundaries further. - What is the public opinion on AI?
Opinions vary globally; for instance, many Chinese citizens view AI products and services positively, while American sentiment is more cautious. Overall, attitudes towards AI are mixed, with excitement about potential benefits and concern over issues like privacy and job displacement. - What are the environmental impacts of AI?
AI’s environmental impact is dual-faceted: while training large models can have significant carbon footprints, AI is also used to optimize energy usage and contribute to environmental sustainability efforts. - How is AI influencing the economy and job market?
The demand for AI-related skills is increasing across all sectors, reflecting AI’s growing integration into various industries. Despite a decrease in private investment in AI for the first time in a decade, the sector continues to thrive and evolve. - What developments are there in AI ethics and governance?
Ethical concerns, such as bias and misuse of AI, are receiving increased attention. Legislative interest in AI is rising, with many countries passing laws related to artificial intelligence. - How is AI impacting education?
Interest in AI education is growing, with an increasing number of computer science students specializing in AI. The report highlights trends in AI PhDs, undergraduate education, and K-12 initiatives. - What are the latest advancements in technical AI performance?
AI models are becoming more capable of general tasks, challenging the trend of AI systems being limited to narrow, specific tasks. However, challenges in reasoning and complex task performance persist. - How is diversity being addressed in the AI field?
The report discusses efforts and trends in increasing diversity within AI research and education, noting some progress and highlighting ongoing challenges in achieving gender and ethnic balance.
[To share your insights with us, please write to sghosh@martechseries.com]
