Artificial Intelligence: Overview, Recent Advances, and Considerations for the 118th Congress

Introduction ..................................................................................................................................... 1 Artificial Intelligence Background and History .............................................................................. 1 Recent Artificial Intelligence Advances .......................................................................................... 2 Benefits and Potential Risks of Artificial Intelligence Technologies .............................................. 3 Artificial Intelligence Technologies in Selected Sectors ................................................................. 4

Health Care................................................................................................................................ 4 Education .................................................................................................................................. 5 National Security ....................................................................................................................... 6

Artificial Intelligence Laws and Legislation ................................................................................... 7

Current Federal Laws Addressing Artificial Intelligence .......................................................... 7 Federal AI Legislation Introduced in the 117th and 118th Congresses ....................................... 9

Perspectives on Regulating Artificial Intelligence ........................................................................ 10 Other Considerations for the 118th Congress ................................................................................. 12

Contacts Author Information ........................................................................................................................ 12

Introduction As the use of artificial intelligence (AI) has grown across a wide range of sectors, so too have strategies to influence the growth of AI and mitigate potential risks. This report provides a brief background on AI technologies and recent advances, including in generative AI (GenAI); benefits and risks of AI tools; current federal laws addressing AI; perspectives on regulating AI; and selected other considerations for the 118th Congress. This report also provides an overview of AI in selected sectors; however, it does not attempt to address all applications of AI. Information on the application of AI technologies in additional sectors can be found in separate CRS products.1

Artificial Intelligence Background and History AI can generally be thought of as computerized systems that work and react in ways commonly thought to require intelligence, such as the ability to learn, solve problems, and achieve goals under uncertain and varying conditions, with varying levels of autonomy. AI is not one thing; rather, AI systems can encompass a range of methodologies and application areas, such as natural language processing, robotics, and facial recognition.

Common terms used in the field of AI include machine learning (ML), deep learning (DL), and neural networks. ML, often referred to as a subfield of AI, examines how to build computer programs that automatically improve their performance at some task, through experience, without relying on explicit rules-based programming to do so.2 One of the goals of ML is to teach algorithms to successfully interpret data that have not previously been encountered. DL systems learn from large amounts of data to subsequently recognize and classify related, but previously unobserved, data. Neural networks, a type of DL often described as being loosely modeled after the human brain, consist of thousands or millions of processing nodes (i.e., computational units). DL approaches have been used in systems across many areas of AI research, from autonomous vehicles to voice recognition technologies.3

Historically, there has been debate over which technologies should be classified as AI. For example, robotic process automation (RPA) has been described as the use of rules-based software to automate highly repetitive, routine tasks normally performed by knowledge workers.4 Because it automates activities performed by humans, it is often described as an AI technology. However, some argue that RPA is not AI because it does not include a learning component. Others discuss RPA as a basic tool that can be combined with AI to create complex process automation, or intelligent process automation, along an “intelligent automation continuum.”5

1 For example, additional products cover AI in consumer lending, defense, and copyright law. See CRS In Focus IF12399, Automation, Artificial Intelligence, and Machine Learning in Consumer Lending , by Cheryl R. Cooper; CRS Report R46458, Emerging Military Technologies: Background and Issues for Congress , by Kelley M. Sayler; and CRS Legal Sidebar LSB10922, Generative Artificial Intelligence and Copyright Law , by Christopher T. Zirpoli.

2 Adapted from Erik Brynjolfsson, Tom Mitchell, and Daniel Rock, “What Can Machines Learn, and What Does It Mean for Occupations and the Economy?,” AEA Papers and Proceedings , vol. 108 (May 1, 2018), pp. 43-47, https://dspace.mit.edu/bitstream/handle/1721.1/120302/pandp.20181019.pdf.

3 Larry Hardesty, “Explained: Neural Networks,” Massachusetts Institute of Technology (MIT) News , April 14, 2017, http://news.mit.edu/2017/explained-neural-networks-deep-learning-0414.

4 For more information, see IBM, “What Is Robotic Process Automation (RPA)?,” https://www.ibm.com/topics/rpa. 5 IBM Global Business Services, “Using Artificial Intelligence to Optimize the Value of Robotic Process Automation,” September 2017, at https://www.ibm.com/downloads/cas/KDKAAK29.

The term artificial intelligence was coined at the Dartmouth Summer Research Project on Artificial Intelligence, a conference proposed in 1955 and held the following year.6 Since that time, the field of AI has gone through what some have termed summers and winters—periods of much research and advancement followed by lulls in activity and progress. The reasons for the AI winters have included a focus on theory over practical applications, research problems being more difficult than anticipated, and limitations of the technologies of the time. Much of the current progress and research in AI, which began around 2010, has been attributed to the availability of large datasets (i.e., big data), improved ML approaches and algorithms, and more powerful computers.7

Recent Artificial Intelligence Advances One of the most notable areas of advancement in AI has been in GenAI models and applications, such as ChatGPT.8 GenAI refers to ML models developed through training on large volumes of data in order to generate content. ChatGPT is an AI chatbot9 from a company called OpenAI, underpinned by a type of AI called a large language model (LLM).10 LLMs are trained on massive amounts of data, largely collected from public internet sites. When a user provides a prompt, usually a text prompt, the model can generate words or paragraphs with human-like quality. Other models can create different types of outputs from text prompts, such as images, music, videos, and computer code. GenAI models work to match the style and appearance of the underlying data, and they have shown what has been called “capability overhang,” meaning hidden capabilities of AI systems that researchers have not uncovered or thought to test for yet. While GenAI tools are not new, recent advances—particularly since the introduction of the transformer architecture11 in 2017 and improvements in “generative pre-trained transformer” (GPT) models since 2019—combined with the open availability to the public of these tools (2022) have led to widespread use.