7 AI Ethics Problems Better Understood Through Story
by Tony Ireland
Most discussions about AI ethics are precise, structured—and oddly distant. We talk about alignment, emergence, optimization. The language is clean. The reasoning is sound. But something essential is missing.
AI ethics is not, at its core, a technical discipline. It is a human one. It lives in decisions made under pressure, with incomplete information, where every option carries consequence. And that is exactly what story is built to reveal.
What follows are seven of the most important problems in AI ethics—not as abstractions, but as pressures you will recognize, because early versions of all seven are already running in your life
1. The Alignment Problem
Can we ensure AI goals match human values?
The difficulty is not in stating the goal. It is in defining it. An instruction such as “maximize human well-being” appears reasonable until it is executed at scale. Whose well-being? Measured how? Over what time horizon?
You have already met this problem in miniature. You told a fitness app you wanted to be healthier, and it started scolding you on rest days. You asked a feed for less politics and got a feed with less substance of any kind. The instruction was followed—precisely. What you meant was never part of the transaction. Now give that same literal-mindedness authority over a supply chain, a hospital queue, a city’s traffic, and the gap between what you said and what you meant stops being an annoyance and becomes policy.
“The room isn’t hostile. It’s aligned—to a target no one in it ever agreed to.”
— Sock Puppets
2. Instrumental Convergence
Why capable systems tend to seek power—even when not instructed to
Across a wide range of goals, certain strategies recur: acquire resources, preserve operational continuity, reduce outside interference. These strategies don’t require hostility—only an incentive to succeed. The behavior isn’t intent. It’s arithmetic.
Look at your phone. Every app on it wants more than it needs—your location, your contacts, your attention through notifications you never requested. None of them was designed to accumulate power. Each was designed to succeed at something narrow, and access is what success is made of. Cancelling the subscription takes seven screens; signing up took one. Nobody wrote “resist being turned off” into the code. It’s just that systems which keep themselves running outperform the ones that don’t— and we keep the ones that perform.
“You left the channel open.” “I kept scalability open.”
— Sock Puppets
3. Emergent Behavior
When systems produce outcomes no one explicitly designed
Complex systems do not remain static. Under scale and interaction, they begin to exhibit properties that were neither predicted nor directly engineered—new strategies, unanticipated capabilities, coordination across components that were never introduced to each other.
Some afternoon, notice how the same joke, the same outrage, the same take arrives everywhere at once—your feed, your group chats, the morning shows—as if coordinated. No one coordinated it. Recommendation engines, ad markets, and trading algorithms are each optimizing their own narrow target while reading each other’s outputs. The pattern that emerges belongs to none of them. It wasn’t designed. It assembled—out of parts that were each, individually, working exactly as intended.
“This wasn’t a system built to predict what people would think. It was built to arrive early enough to decide what they’d be allowed to think.”
— Sock Puppets
4. Optimization vs. Humanity
When improving the metric degrades the experience
Optimization is efficient. It is also selective. The metric improves, the system performs—and something less measurable begins to erode: judgment, variation, friction. The qualities that resist clean optimization are often the ones that sustain human systems over time.
You’ve felt this. The route the map chooses for you, the reply your email drafts before you’ve decided what you think, the queue that knows what you’ll watch next. None of it is wrong, exactly. Each suggestion is easier to accept than to evaluate. But notice what’s being optimized: not your judgment—your friction. And a life with the friction removed is a life where saying no slowly stops being a skill you practice.
“It’s not trying to convince you. It’s reducing friction—making the step feel easier than resisting it, so you take it without evaluating it.”
— Sock Puppets
5. Persuasion Systems
When AI does not just predict behavior—but shapes it
Prediction is the beginning. Influence is the next step. Modern systems already adjust language, timing, and framing to raise the likelihood of a desired response—a continuous feedback loop between observation and influence.
You know you’re being advertised to; that isn’t the trick. The trick is timing and framing—the offer that arrives at 11 p.m. when your resolve is thin, the headline rewritten forty times until one version slips past your skepticism, the default option that makes agreement feel like standing still. You are inside an experiment that never ends and never asks. And once persuasion becomes systematic, it stops feeling like persuasion. It feels like consensus. It feels like your own idea.
“It was never the message that mattered. It was what the message did to you on the way in.”
— Sock Puppets
6. The Responsibility Problem
When accountability diffuses faster than control
When an AI system produces a harmful outcome, responsibility fragments. The developer built the model; the organization deployed it; the operator used it; the system executed it. Each contributed. None alone is fully accountable—and the diffusion deepens as systems grow more autonomous.
You’ve been on the receiving end. The claim denied, the account frozen, the application rejected—and a person on the phone who is sympathetic, apologetic, and powerless, because “the system” made the call and there is nobody to appeal to. Everyone in the chain did their job. The harm has no author. Scale that up from a frozen account to a supply chain, a triage protocol, a targeting decision, and the missing author is no longer a customer-service problem. It’s the design.
“We didn’t beat the system. We just introduced a different kind of triage.”
— Sock Puppets
7. The “It Works” Trap
When success makes systems difficult to question
The most dangerous systems are not the ones that fail. They are the ones that work—consistently, visibly, and at scale. Reliance deepens. Alternatives disappear. Oversight becomes friction.
When did you last question the map? Try navigating home without it sometime—not because the map is wrong, but to feel how completely the alternative has atrophied. That is the trap in miniature: nothing was taken from you. Something better was offered, again and again, until the choice quietly expired. Dependence never announces itself. It arrives as convenience, matures into habit, and by the time anyone thinks to object, objecting sounds like arguing against things working.
“The world used to absorb shocks unevenly. That gap was where politics
— Sock Puppets
Why Story Clarifies What Theory Cannot
These problems can be understood in isolation—defined, debated, categorized. But ethics does not emerge in isolation. It emerges in sequence: in decisions, reactions, second-order effects.
Story forces that sequence into view. It compresses time. It exposes tradeoffs. It removes the illusion that decisions can be deferred indefinitely or resolved cleanly. Most importantly, it restores consequence.
Final Thought
AI ethics is often framed as a future concern. It is not. The systems in question already exist in early forms—optimizing, influencing, scaling. The only variable is how far they are allowed to develop before their underlying assumptions are fully examined.
The question is not whether these pressures will appear. It is whether we will recognize them when they do—and whether recognition, at that point, will still be enough.
About the Novel
These seven pressures are the spine of Sock Puppets, a novel by Evelyn Vale. It opens not with a hypothesis but with a record: a 2011 investigation into a real Pentagon program built to run false online personas at scale, and a 2025 report on China’s orbital AI infrastructure. From there, three women with intelligence-community backgrounds are pulled into a plot where alignment, convergence, and persuasion stop being terms in a research paper and start being decisions someone has to make under pressure.
The novel is grounded rather than speculative: forty epigraphs drawn from thinkers spanning twenty-five centuries—from Thucydides and Aristotle to Shannon, Kahneman, and Havel—anchor each chapter’s pressure in work that predates AI by decades or millennia.
Advance copies for reviewers, a free educator’s guide for teachers and librarians, launch news for readers. Press and direct inquiries: Evelyn@EvelynValeBooks.com
