Part I: Introduction to Ethical Frameworks
“Even if you know nothing about digital marketing analytics, digital marketing analytics knows plenty about you.”
When I started my book, Digital Marketing Analytics: In Theory and in Practice, with that line in 2020, the silent and invisible harvesting of consumer data was already reaching unprecedented levels. At the time, the data tracking landscape was defined by an extreme acceleration of collection capabilities, deep systemic anxieties, and a severe literacy gap among consumers.
That landscape has fundamentally shifted with the rapid deployment of high-stakes AI systems. While the data strategies of 2020 focused on the passive harvesting of consumer footprints, modern systems have advanced to autonomous decision-making and real-time execution. And because machines operate solely on formal mathematical directives, they do not possess moral agency, consciousness, or emotional intelligence.
As a result, the responsibility for ensuring that these systems protect human dignity and promote societal welfare rests entirely on us as humans—specifically the developers, corporate entities, and regulatory bodies that design and deploy them. To build and utilize responsible AI systems, we must move beyond a simple desire for “optimization.” We must introduce a moral code.
The Realities of Modern Data Governance
Yet, sadly, morality is rarely seen in the decisions of today’s technology leaders. Whether it is Meta monetizing risks to children, X failing to protect victims of harassment, or regulators lacking the courage to enforce meaningful boundaries on AI model deployment, the realities of leaders choosing profits over people are all around us.
To navigate today’s AI landscape, we must embed an approach to AI governance rooted in three relevant ethical frameworks:
- Utilitarianism (Maximizing Net Good)
- Deontology (Duties, Rights, and Absolute Rules)
- Virtue Ethics (Corporate Character and Integrity)
I’m kicking off a 5-part series of articles that will explore each of these frameworks and offer tactical ideas for how you can use them to ensure the AI systems you deploy are managed in a safe, respectful, and responsible way. Today, in Part I, we’ll begin with a high-level overview of each framework to set our baseline.
1. Utilitarianism: Maximizing Net Good
Utilitarianism focuses on consequences, net utility, and optimizing outcomes for the greatest number of stakeholders. Its core principle rests on the idea that an action is morally right if it produces the greatest balance of benefits over harms for everyone affected. It relies heavily on quantifiable impacts, making it highly attractive to data-driven business managers. With respect to AI, utilitarianism evaluates data models based on their systemic impacts, error rates, and aggregate benefits versus their localized harms.
Data science naturally defaults to utilitarianism because the mathematical architecture of machine learning optimizes for quantitative outcome metrics. Utilitarianism focuses entirely on maximizing aggregate welfare or net utility.
For example, consider a hypothetical (albeit very real) scenario where a major health system deploys a predictive algorithm to manage scarce ICU beds and specialized medical equipment. The AI is trained on historical data to predict survival probability, optimizing hospital operations and maximizing the absolute number of lives saved.
The model achieves a 15% increase in overall hospital survival rates, but it consistently deprioritizes patients from historically underserved communities due to systemic data voids and lower baseline access to preventive care.
A modern rule-utilitarian would reject this algorithm. They would argue that systemic discrimination erodes long-term public trust in healthcare systems, causing widespread societal harms that far outweigh short-term operational efficiencies. The framework forces managers to calculate the total, long-term consequences of algorithmic bias.
2. Deontology: Duties, Rights, and Absolute Rules
Deontology focuses on duties, universal rights, and absolute moral rules that must never be broken. Its core principle is derived from the philosophy that certain actions are inherently right or wrong, regardless of their consequences. It commands us to act out of duty and treat individuals as ends in themselves, never as a mere means to a corporate goal. From an AI perspective, the application of deontology means establishing unyielding baselines for consumer privacy and data governance that cannot be traded away for profit or efficiency.
As an example of deontology in action, take an AI startup that scrapes billions of public social media posts, private blog entries, and personal photographs without explicit user consent. They use this massive dataset to train a highly profitable, next-token prediction model for commercial use. The startup argues that the data was technically public, and the resulting model provides immense intellectual and economic value to millions of global users.
Deontology rejects this practice immediately. It states that scraping data without explicit, informed consent violates individual autonomy and treats human creators as mere fuel for a corporate machine. While the United States framework historically favored decentralized, market-driven data aggregation, deontology aligns perfectly with the strict, prescriptive architecture of the European Union AI Act, which treats data privacy as a fundamental human right rather than a negotiable asset.
3. Virtue Ethics: Corporate Character and Integrity
While rules and consequences establish boundaries, true algorithmic governance requires evaluating the human element through virtue ethics. Virtue ethics focuses on the moral character of the actor and the development of practical wisdom. Instead of looking at rules or outcomes, virtue ethics focuses on the character, motivations, and integrity of the moral agent (the developer, data scientist, or corporation). It asks: “What kind of person or organization do we become when we make this choice?” It prioritizes virtues like honesty, fairness, prudence, and courage.
For example, consider a company that discovers its customer service conversational AI agent occasionally experiences algorithmic hallucinations—meaning it invents false but highly convincing information about product warranties—yet the bot increases operational efficiency by 40%.
A virtue ethicist would look beyond the profitability or the lack of explicit anti-hallucination laws. They would argue that deploying a system known to generate fabrications lacks the essential virtue of honesty and damages organizational trustworthiness. To act with prudence and integrity, the company must pause deployment to refine the technical architecture, ensuring the AI aligns with the character of a truthful, reliable enterprise.
What Comes Next
Next week, in Part II of this series, we are going to dive deep into our first framework: Utilitarianism. Because machine learning naturally defaults to quantitative outcomes, data professionals are highly vulnerable to falling into a dangerous “efficiency trap.”
We will explore how you can mathematically audit your AI models for true net utility, identify hidden data voids, and ensure your pursuit of optimization doesn’t inadvertently cause systemic harm.
Until next week, Keep Analyzing!
