Artificial Intelligence and the Imperative of Public Ethical Literacy

On a rainy Tuesday evening in March 2023, Mary Johnson, a 38-year-old automotive assembly line worker in Detroit, sat staring at her laptop screen in disbelief. The community college she had applied to for a part-time mechanical engineering program—her first step toward a higher-paying, less physically demanding career—had rejected her application. The official letter cited her 2.8 undergraduate GPA from a decade prior as insufficient to meet the program’s competitive threshold. But when Johnson appealed the decision, she uncovered a detail the admissions office had never disclosed: the rejection was not made by a human admissions officer, but by an algorithmic decision-making system that evaluated 17 distinct variables to rank applicants, from high school transcripts to zip code of residence. Johnson’s zip code, 48201, covers a predominantly Black, low-income neighborhood on Detroit’s east side, and the algorithm assigned it a 30% higher weight than her GPA, automatically downgrading her application before a human ever reviewed it. After a two-week review process that included Johnson’s personal statement and letters of recommendation, the college reversed the decision, admitting her to the program. But Johnson’s case was not an isolated incident: a 2024 internal audit of the college’s admissions algorithms found that 12% of rejected applicants from low-income majority-Black zip codes between 2020 and 2023 would have been admitted if their applications had been reviewed by human officers first.

Johnson’s experience is not an anomaly. It is a window into a quiet, accelerating revolution: the integration of artificial intelligence (AI) into nearly every layer of social, economic, and civic life. From the algorithms that filter job applications and set insurance premiums, to the systems that determine criminal sentencing and allocate public school funding, AI-driven decision-making now shapes outcomes for billions of people, often without their knowledge, consent, or ability to appeal. Unlike earlier waves of technological change that were confined to specialized professional or industrial contexts, contemporary AI systems operate as invisible gatekeepers to opportunity, resources, and even basic rights. And yet, for most people, the logic of these systems remains a mystery: a 2024 Pew Research Center survey found that only 17% of U.S. adults can accurately explain how machine learning—the core technology behind most modern AI systems—works, and just 12% are aware of the existence of algorithmic decision-making systems in the public services they use.

This gap between the pervasive influence of AI and the public’s low literacy about its functions and ethical implications is not merely a knowledge gap. It is a democratic deficit. The rules that govern how AI systems are built, deployed, and held accountable are being written today, by a small cohort of technology company executives, computer scientists, and policy elites, with little input from the communities most affected by algorithmic harm. The result is a landscape where AI systems routinely entrench and amplify existing social inequalities, where lethal autonomous weapons are being readied for deployment without global public consensus, and where personal data is harvested and exploited at scale with minimal regulatory oversight. Addressing these challenges requires more than technical fixes or top-down regulation: it requires broad, inclusive public engagement with the ethical dimensions of AI, what we will call *public ethical literacy*.

This chapter lays the foundational groundwork for that project. It begins by defining the core technical concepts that underpin all contemporary AI systems, demystifying terms that are often used interchangeably or misrepresented in public discourse. It then traces the historical evolution of AI ethics, from its origins in mid-20th century philosophical and computing debates to its current status as a defining public policy issue of the 21st century. Next, it makes the case for AI ethics as a public good, arguing that the stakes of AI governance are too high, and the interests at stake too broad, to be left to a small group of technical experts. Finally, it outlines the book’s three core thematic pillars—algorithmic bias, lethal autonomous weapons, and data privacy—and its formal, accessible analytical approach, designed to equip general readers with the knowledge and tools to participate meaningfully in the defining ethical conversations of our time.

1.1 Core Artificial Intelligence Concepts: Demystifying the Terminology

Before engaging with the ethical dilemmas of AI, it is essential to clarify the basic technical categories that define the field, as widespread confusion about these terms fuels both unwarranted fear and complacent acceptance of AI systems. For the purposes of this book, we will use four core definitions, aligned with the consensus of the global AI research community:

1. Narrow AI (Artificial Narrow Intelligence, ANI): The only form of AI that has been successfully built and deployed to date. Narrow AI refers to systems designed to perform a single, specific task (or a narrow set of related tasks) at or above human level of performance. Examples include facial recognition systems, large language models (LLMs) like ChatGPT, recommendation algorithms used by social media platforms, and autonomous vehicle navigation systems. Narrow AI systems have no general cognitive ability: a LLM that can write a coherent essay cannot independently learn to diagnose cancer or pilot a drone, unless it is explicitly trained and programmed to do so. All AI systems currently in commercial or public use are narrow AI.

2. Artificial General Intelligence (AGI): A hypothetical form of AI that would possess the ability to understand, learn, and apply knowledge across any intellectual task a human can perform, from creative writing to scientific research to emotional reasoning. AGI would not be limited to pre-specified tasks, and would be able to set its own goals and adapt to novel contexts without human intervention. As of 2024, no AGI system exists, and there is no consensus among AI researchers about when, or even if, AGI will be achievable. A 2024 Stanford AI Index Report survey of 2,778 AI researchers found that 76% believe AGI will not be achieved before 2040, with 42% estimating it will take 50 years or more, if it is achievable at all[^1].

3. Machine Learning (ML): A subfield of AI that enables systems to learn patterns from data without being explicitly programmed for every possible scenario. Rather than following a fixed set of hand-coded rules, machine learning models identify statistical correlations in training data and use those correlations to make predictions or decisions about new, unseen data. There are three primary categories of machine learning:

• *Supervised learning*: The model is trained on labeled data, where each input is paired with a correct output (e.g., a dataset of cat images labeled “cat” to train an image recognition model).
• *Unsupervised learning*: The model is trained on unlabeled data, and must identify hidden patterns or structures on its own (e.g., clustering customers into distinct groups based on purchasing behavior).
• *Reinforcement learning*: The model learns by interacting with an environment, receiving rewards for correct actions and penalties for incorrect ones (e.g., training an AI to play chess by playing millions of games against itself).

4. Algorithmic Decision-Making Systems (ADMS): Systems that integrate machine learning models (or rule-based algorithms) into real-world decision-making processes, replacing or augmenting human judgment in high-stakes contexts. ADMS are now used to determine who gets a job, who gets a loan, how long a criminal sentence should be, which patients get access to scarce medical resources, and even which children are flagged for child welfare interventi