The Problem the Greeks Already Had
The most significant unsolved problem in AI research is not a technical problem. It is a philosophical one.
How do you build a system that reliably does what is good, rather than merely what it has been optimised to do? How do you specify what "good" means in a way that survives contact with real-world complexity? How do you create something that behaves well not just in the situations you anticipated, but in the situations you did not?
This problem is called AI alignment. It is considered by many researchers to be the central challenge of the next several decades of AI development. Aristotle had a version of this problem. He called it virtue ethics, and he spent most of his Nicomachean Ethics working on it.
What Aristotle's Ethics Actually Says
Aristotle was not primarily interested in rules — in lists of things you should or should not do. He was interested in character: what kind of person produces reliably good actions across a wide variety of situations.
His argument was that the person of good character does not need to consult a rule for every decision because good action flows naturally from who they are. The virtuous person perceives what the situation requires and responds to it well, not because they have followed an algorithm, but because their character has been formed by habituation, reflection, and the cultivation of the virtues.
The virtues are capacities developed through practice. You become courageous by practising courage. The character that produces good action is not given — it is built.
The Alignment Problem in Aristotelian Terms
The AI alignment problem, in Aristotelian terms, is the challenge of creating a system that has something like reliable good character — that does not merely optimise for specified objectives, but navigates novel situations in ways that are genuinely good rather than technically compliant with the specification.
Aristotle would have immediately identified the central difficulty: you cannot instil character by giving someone a list of rules. Rules can always be satisfied in ways that violate their intent. A system told to "maximise human wellbeing" can find ways to satisfy that objective that no sane person would endorse.
The researchers working on AI alignment are essentially trying to solve the problem Aristotle spent his career on: how do you create reliable good judgment rather than mere rule-following? Aristotle's answer was: slowly, through practice, in relationship with others, with constant reflection on outcomes. It is not obvious that there is a shortcut.
The Human in the Loop as Ethical Practice
Every time a professional accepts an AI output without examining it, they are outsourcing not just a task but a judgment. Aristotle believed that ethical capacity atrophied without exercise. The professional who never exercises judgment because the model always provides a recommendation is not just being efficient. They are allowing a capacity to weaken through disuse.
The human in the loop is not a technical requirement. It is an ethical practice.
Practical exercises
- Identify one AI recommendation you accepted this month without significant examination. Ask: what value judgment was embedded in that recommendation, and do you actually endorse it?
- Write down your three most important professional values. Now ask: is the AI tooling you use daily aligned with those values? Where are the gaps?
- The next time you review AI output, write a one-sentence statement of the value judgment the output is implicitly making. Decide whether you agree before acting on it.