AI has quickly permeated the general lexicon, but public understanding of what constitutes AI, and what its ethical implications are, lags behind its proliferation. This introductory blog post offers a primer on artificial intelligence, exploring its definition—or rather, the universal lack thereof.
No Universal Definition
It might surprise people to know AI has no universally accepted definition. See U.S. GOV’T ACCOUNTABILITY OFF., GAO-18-142SP 15 (2018) (stating “[t]here is no single definition of AI, but rather differing definitions and taxonomies”); Forrest E. Morgan et al., Military Applications of Artificial Intelligence: Ethical Concerns in an Uncertain World, RAND CORP. 8-9 & 9 n.4 (2020), www.rand.org/pubs/research_reports/RR3139-1.html [https://perma.cc/4XUX-7PQZ] (emphasizing experts’ aversion to providing definitive definitions for AI). This lack of consensus is partly due to the field’s quick evolution and its interdisciplinary nature. See David S. Rubenstein, Acquiring Ethical AI, 73 Fla. L. Rev. 747, 758 (2021) (“‘Artificial intelligence’ has no ‘universally accepted definition.’ The lexical rifts are largely attributable to the field’s evolution and multi-disciplinarity, which spans computer science, mathematics, psychology, sociology, neuroscience, philosophy, linguistics, and intersects with countless more.”). John McCarthy has offered a classic definition of AI as the science and engineering of making intelligent machines, especially intelligent computer programs related to the similar tasks of understanding human intelligence using computers. IBM, What is Artificial Intelligence (AI)? https://www.ibm.com/topics/artificial-intelligence (last visited Nov. 26, 2023); see Yijun (Jenny) Ge, AI and Corporate Governance: How May AI Assist the Board in Informed Decision-Making?, 24 U.C. Davis Bus. L.J. 115, 120 (2024).
AI as a Simulation of Human Intelligence
One common approach is to define AI as the ability of machines to mimic human intelligence or cognitive functions. See Tabrez Y. Ebrahim, Artificial Intelligence Inventions & Patent Disclosure, 125 Penn St. L. Rev. 147, 165 (2020). This includes tasks like problem-solving, decision-making, learning, perception, reasoning, and language understanding. See Scott J. Shackelford & Rachel Dockery, Governing AI, 30 Cornell J.L. & Pub. Pol'y 279, 285 (2020). The U.S. Congress defines AI as a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations [,] or decisions influencing real or virtual environments. H.R. REP. NO. 116-617, at 1164 (2020) (Conf. Rep.); see Haley Giaramita, AI Assistance in the Drug Development Process: Reaching for A Regulatory Framework, 54 Seton Hall L. Rev. 1239, 1243 (2024). AI systems, as Congress defines them, employ a combination of machine-generated and human-provided inputs to (A) perceive real and virtual environments; (B) abstract such perceptions into models through analysis in an automated manner; and (C) use model inference to formulate options for information or action. Id.
AI as a Set of Techniques
Another perspective views AI as a collection of computational techniques. See Ryan Calo, Artificial Intelligence Policy: A Primer and Roadmap, 51 U.C. Davis. L. Rev. 399, 404 (2017) (“AI is best understood as a set of techniques aimed at approximating some aspect of human or animal cognition using machines.”); Claudia E. Haupt, Artificial Professional Advice, 21 Yale J. L. & Tech. 55, 77 (2019). These techniques are often centered around machine learning, including supervised, unsupervised, and reinforcement learning. See Amanda Levendowski, How Copyright Law Can Fix Artificial Intelligence's Implicit Bias Problem, 93 Wash. L. Rev. 579, 590 (2018) (“When journalists, researchers, and even engineers say ‘AI,’ they tend to be talking about machine learning, a field that blends mathematics, statistics, and computer science to create computer programs with the ability to improve through experience automatically.”); Urs Gasser & Virgilio A.F. Almeida, A Layered Model for AI Governance, 21 IEEE INTERNET COMPUTING 58, 59 (2017) (“From a technical perspective, [AI] is not a single technology, but rather a set of techniques and sub-disciplines ranging from areas such as speech recognition and computer vision to attention and memory, to name just a few.”).
Artificial neural networks (ANNs) are one powerful computing technique often used in AI systems—inspired by the structure and function of the human brain, they use interconnected nodes or "neurons" to process information. See, e.g., N. F. Sussman, A Behavioral Theory of Robot Rights, 32 S. Cal. Interdisc. L.J. 113, 117 (2022) (“[M]y working definition of AI would include systems that execute via symbolic machine learning algorithms or ‘connectionist’ neural nets, which operate by means of programs that self-update according to the external ‘feedback such programs may receive on their prior outputs.). These networks "learn" by adjusting the weights of the connections between these neurons as they are exposed to input data. See David W. Opderbeck, Copyright in AI Training Data: A Human-Centered Approach, 76 Okla. L. Rev. 951, 958–59 (2024) (“Like the human brain, the algorithms include parameters that allow these systems to ‘learn’ as more and more data is processed, adjusting the algorithmic weights in various nodes[.]”); Walter A. Mostowy, Explaining Opaque AI Decisions, Legally, 35 Berkeley Tech. L.J. 1291, 1299 (2020) (“The neural network [] compares [incorrect] output to the correct output … and makes small adjustments to all of the neurons' parameters—the weights and the thresholds—to achieve a slightly better outcome. This process is then repeated for the entire training set, again and again, until overall accuracy no longer improves.”). ANNs, particularly deep learning networks, which consist of multiple layers of neurons, excel at recognizing patterns within datasets and can even make predictions based on those patterns. This capability to learn from data and make decisions based on recognized patterns aligns with a common understanding of AI as the simulation of human cognitive abilities using machines.
AI as a Practice
Some have proposed that AI is best understood as a practice—an applied science and engineering discipline that aims to instill qualities we consider intelligent into computer software. See Michael Veale et al, AI and Global Governance: Modalities, Rationales, Tensions, 19 Ann. Rev. L. & Soc. Sci. 255, 256 (2023). This definition emphasizes the human element: (1) Humans define the goals and applications of AI technologies; (2) humans design the tools and processes used in AI; thus, (3) the governance of AI must consider its human practitioners, their organizations, and the broader social, economic, and political structures surrounding its use. Id.
A Complex Endeavor
Ultimately, defining AI remains a complex and evolving endeavor. The lack of a single, universally accepted definition underscores the need to engage with AI's multifaceted nature when considering its ethical and legal implications. Future posts will move beyond these definitional explorations. My next post will delve into key distinctions within the field of AI, such as narrow (or weak) AI versus general (or strong) AI, providing a framework for understanding the varying capabilities and limitations of different AI systems. The subsequent post will then offer an initial survey of the inherent challenges in governing and defining AI, setting the stage for deeper explorations of specific ethical and societal implications.