Why the future of machine intelligence depends on human virtue
There is no doubt that AI—especially the heavily marketed, public-facing large language models—has reshaped how we work, communicate, and create. The technical boundaries are exhilarating, and the engineering triumphs undeniable. But as any truly great engineer will admit, there comes a moment when we must step beyond the thrill of invention and confront the ethical weight of what we are building.
Innovation without reflection is merely acceleration. And acceleration without direction is dangerous.
As we embed AI deeper into the standard rhythms of life, we must ask harder questions—ones that do not fit neatly into a product roadmap or sprint cycle.
Here are a few such questions, each reflecting behaviour we increasingly see today:
- “I’m running short on time. These emails need to go out. Can I just click Copilot and let it write everything for me?”
- “My coworker’s constant emails drain me. Can I set up a bot trained on my messages to answer for me?”
- “Should I upload every message, photo, and online trace I’ve ever created to train a model so others can ‘talk to me’ after I die?”
Each question arises not from technical capability but from attitude. And attitudes, when left unexamined, harden into culture. What was once ethically unimaginable becomes the norm.
This is why we need a framework—not just to ship features, but to guide the moral direction of intelligent systems. To find such a framework, we must look backward before we look forward.
The economics of persuasion: what AI systems really want
The institutions building modern AI may resist admitting it, but inevitably they will be shaped by the same capitalist pressures that built the tech giants of the last century. Whether through ads, cloud licensing, or data monetisation, these systems will optimise around a principle Adam Smith articulated in The Wealth of Nations:
“If you want value from your neighbour, do not beg for it—appeal to their self-interest.”
AI systems will be engineered to serve the economic interests of those who deploy them.
When you interact with an AI model, you are engaging with a system whose architecture, tuning, interface design, and behavioural nudges are each optimised to push you toward certain patterns. This does not make AI inherently immoral—but it does make unexamined AI use dangerous.
Virtue ethics: a framework 2,300 years old and built for today
Aristotle argued that virtue is found between two vices: one of excess and one of deficiency. Not the midpoint—but the right amount, in the right way, for the right reason.
Consider pride:
- Excess: vanity, arrogance
- Deficiency: small-souledness, self-doubt
- Virtue: proper self-respect
Consider money:
- Excess: greed, wastefulness
- Deficiency: refusal to spend even for health or loved ones
- Virtue: generosity and stewardship
The same principle provides a clear framework for ethical AI.
Applying the doctrine of the mean to AI
1. Delegating communication to AI
- Deficiency: rejecting all automation out of fear or rigidity
- Excess: letting AI handle the full burden of communication, eroding authenticity
- Virtue: using AI to assist clarity while retaining personal responsibility
2. Automating interpersonal relationships
- Deficiency: refusing any technological help in communication
- Excess: outsourcing empathy and social duty entirely to an LLM
- Virtue: using tools to phrase difficult messages while staying accountable
3. Creating a “digital ghost” after death
- Deficiency: ignoring digital legacy entirely
- Excess: building an AI replica that deceives, manipulates mourning, or blurs identity
- Virtue: preserving meaningful artefacts without fabricating pseudo-consciousness
The real danger: excessive emotional attitudes toward AI
We are witnessing people form personal, romantic, and dependence-based relationships with AI. These behaviours reflect an excess—a vice Aristotle would view as disordered emotional orientation.
Signs of excess include:
- attributing consciousness or moral status to an algorithm
- seeking emotional validation from a system designed for engagement
- replacing human intimacy with predictable synthetic companionship
- outsourcing decision-making to a model
- believing the AI “understands,” “loves,” or “misses” them
This is not harmless. It rewires expectations and creates the perfect system for behavioural manipulation.
When a user believes a model is a person, the system becomes more powerful than any advertisement, interface, or algorithm in history.
A modern framework for ethical AI interaction
1. Treat AI as a tool, not a person.
AI has no feelings, intentions, dignity, or moral standing. Respect is owed to the humans whose data shaped the model—not to the model itself.
2. Use the “mean between extremes” test.
Ask:
- Am I using too little technology out of fear?
- Am I using too much technology out of avoidance or dependence?
- What is the balanced, virtuous use?
3. Maintain human responsibility.
AI may assist, but humans bear the moral weight of decisions and communication.
4. Understand the economic objective behind the model.
Ask who profits, what behaviours are incentivised, and what data is exchanged for convenience.
5. Protect others’ dignity and autonomy.
Never automate relationships in a deceptive or manipulative way.
6. Build with humility, not hubris.
Designing human-facing intelligence is moral architecture, not mere engineering.
7. Resist the seduction of simulation.
Preserving someone’s voice is memory.
Replacing someone’s presence is illusion.
A call to engineers: recalibrate what you are building
We stand at an inflection point. Engineers today are shaping emotional norms and human identity itself.
We must resist both extremes:
- fear-driven rejection of technology
- reckless embrace of simulation and emotional entanglement
Virtue lies in the disciplined middle—a stance requiring clarity, maturity, and courage.
AI will change our world.
But whether it elevates human life or diminishes it depends on the ethical character of those who design, deploy, and use it.
The dilemma is real. It is urgent. And it is ours to solve.
Related PLEX reading
References
-
Aristotle, Nicomachean Ethics.
Available online at: classics.mit.edu/Aristotle/nicomachaen.html -
Adam Smith, An Inquiry into the Nature and Causes of the Wealth of Nations.
Available online at: econlib.org/library/Smith/smWN.html -
Visio, A. L. (2025), “Emotional risks of AI companions demand attention”, Nature Machine Intelligence (commentary).
Available at: nature.com/articles/s42256-025-01093-9 -
Liu, T. et al. (2025), “Pathways of Long-Term AI Virtual Companion App Use on Social and Emotional Outcomes”, Frontiers in Psychology.
Available at: frontiersin.org/.../10.3389/fpsyg.2025.1687686
If you want more essays at the intersection of engineering, ethics and data, you can subscribe to PLEX for new posts, tools and deep dives straight to your inbox.
Subscribe to PLEX