国会研究服务部-人工智能：背景、选定问题和政策考虑（英文）-2021.5-47页_1mb

Summary of "Artificial Intelligence: Background, Selected Issues, and Policy Considerations"

Introduction

Artificial intelligence (AI), first introduced in the 1950s, refers to computerized systems that can learn, solve problems, and make decisions under uncertain and varying conditions. Over the past decade, AI has experienced rapid growth due to increased computing power, availability of big data, and advancements in AI methodologies. While AI offers significant potential for improving safety, quality, and efficiency in various sectors, it also raises concerns about trustworthiness, bias, ethics, and workforce impacts. The U.S. federal government has become more involved in AI through executive orders, legislative actions, and the formation of committees and working groups. This report outlines the background, terminology, historical context, and current policy considerations related to AI.

What Is AI?

AI encompasses a range of methodologies and applications, including machine learning (ML), natural language processing (NLP), and robotics. While there is no universally agreed-upon definition of AI, the National Institute of Standards and Technology (NIST) defines it as systems that can learn to solve complex problems, make predictions, and influence real or virtual environments. The U.S. Congress has attempted to define AI in legislation, such as the John S. McCain National Defense Authorization Act for Fiscal Year 2019, which categorized AI systems based on their ability to think like humans, act like humans, think rationally, or act rationally.

AI is currently considered to be narrow AI, meaning it is designed for specific tasks. Examples include email spam filtering, voice assistants, and financial lending decisions. Some experts prefer the term "augmented intelligence" or "human-centered AI" to emphasize the role of AI in enhancing human activities rather than replacing them. General AI, which would demonstrate intelligent behavior across a wide range of tasks, is still considered unlikely for the foreseeable future.

AI Terminology

The report highlights key AI-related terms and techniques, including:

• Machine Learning (ML): A subfield of AI that enables systems to automatically learn and improve from data or experience, without being explicitly programmed.
• Deep Learning (DL): A subset of ML that uses large datasets to recognize and classify previously unobserved data. It is commonly used in areas like autonomous vehicles and voice recognition.
• Generative Adversarial Networks (GANs): A type of ML system that consists of two competing networks—a generator and a discriminator. GANs are used in deepfake creation and can learn from less data than other deep learning algorithms.
• Supervised Learning: Uses labeled data to train algorithms to classify new, unseen data.
• Unsupervised Learning: Analyzes unlabeled data to find underlying patterns.
• Reinforcement Learning (RL): Enables systems to learn from experience, with rewards for achieving specified objectives.

Historical Context of AI

AI research has a long history, dating back to the 1940s, with formalization in the 1950s. The field has experienced periods of growth ("AI summers") and decline ("AI winters"), often due to theoretical challenges, limited data, and technological constraints. The resurgence of AI since around 2010 is attributed to the availability of big data, improved ML algorithms, and more powerful computing systems.

Waves of AI

DARPA has identified three waves of AI development:

1. First Wave: Handcrafted Knowledge – Systems based on predefined rules, such as expert systems, with limited learning and reasoning capabilities.
2. Second Wave: Statistical Learning – Systems that use statistical models and big data, but lack contextual awareness and explainability.
3. Third Wave: Contextual Adaptation – Systems that can adapt to new tasks and situations, with the ability to learn and reason in real-world contexts. This includes explainable AI (XAI) and other adaptive technologies.

Recent Growth in AI

AI has seen substantial growth in research and development, as evidenced by the increase in peer-reviewed publications. According to the AI Index group, the number of peer-reviewed AI publications in the Scopus database increased nearly 12-fold between 2000 and 2019. In 2020, China surpassed the U.S. in AI journal citations, though the U.S. still leads in AI conference citations. The number of AI-related preprint papers on arXiv increased over six-fold from 2015 to 2020, with machine learning and computer vision being the most common subfields.

Federal Activity in AI

The U.S. federal government has taken significant steps to address AI, including:

• Executive Orders: President Trump issued two executive orders, the American AI Initiative (E.O. 13859) and the promotion of trustworthy AI in the federal government (E.O. 13960).
• Committees and Working Groups: The National Science and Technology Council has established committees to coordinate federal AI activities and develop strategic plans.
• Legislation: Several laws enacted during the 116th Congress addressed AI, including the National Artificial Intelligence Initiative Act, the AI in Government Act, the IOGAN Act, and P.L. 116-94, which included a financial program for AI exports.

Selected Issues for Congressional Consideration

Congress has focused on several key issues related to AI, including:

• Trustworthiness and Bias: Concerns about algorithmic bias and the need for transparency in AI systems.
• Workforce Impacts: Potential job displacement and the need for reskilling and upskilling.
• Ethics and Fairness: The importance of ensuring AI is used ethically and fairly.
• International Competition: The U.S. must maintain its leadership in AI R&D and standards development.
• Standards Development: The need for national and international standards to ensure consistency and trust in AI technologies.

Conclusion

AI presents both opportunities and challenges for society, the economy, and national security. While it has the potential to drive innovation and economic growth, it also raises ethical, legal, and social concerns. Congress continues to debate how to regulate AI, balance innovation with oversight, and ensure that the U.S. remains competitive in the global AI landscape. The report underscores the importance of continued research, investment, and policy development to address these complex issues.