AI Strategy and Ethics

Overview

You are coming of age in the first years of a technology that may reshape work, knowledge, and power more profoundly than anything since electrification — and almost everything anyone tells you about it is either breathless hype or reflexive doom. This discussion is an exercise in doing the harder thing: thinking clearly about artificial intelligence in a specific, concrete domain, holding the real opportunities and the real dangers in view at once, and arriving at a position you can actually defend rather than one you absorbed from whoever you read last.

The discussion is built around a discipline that most public conversation about AI skips entirely: getting specific. "Is AI good or bad?" is not a real question; it has no answer and produces only posturing. "What happens to the work, the workers, and the customers of this particular industry as these tools get adopted, and what would responsible adoption look like?" is a real question, and answering it well requires research, reasoning, and judgment. By the end you should have a written framework — your own — for the responsible use of AI in one domain you care about, defensible against an intelligent challenger and grounded in how that domain actually works.

The Big Question

When a powerful technology can do something that humans used to do, how do we decide whether, where, and how it should be used — and who gets to decide?

That question has no clean answer, and any framework that claims one is lying. The work of this discussion is not to resolve it but to develop the judgment to reason through it case by case — to see why the right answer in radiology might be the wrong answer in journalism, in hiring, in art, in warfare, and to be able to say why with precision rather than with feeling.

Context for the Facilitator

Your job in this discussion is to be the intelligent adversary the participant will not find in their feed. The danger here is not that they will fail to have opinions about AI — they will arrive with plenty. The danger is that those opinions will be inherited, vague, and undefended, and that the participant will mistake confidence for understanding. Push relentlessly on specifics. Every time they make a sweeping claim, ask for the concrete case. Every time they take a side, make them argue the other one first.

A few orientations to hold:

• Both the hype and the doom are usually lazy. The cheerleader who says AI changes everything and the catastrophist who says it destroys everything are making the same error: refusing to get specific enough to be wrong. The interesting thinking lives in the particulars — this task, this industry, this tradeoff. Steer the participant relentlessly toward the concrete.
• Technical possibility is not the same as wisdom, and the participant will conflate them. "AI can do this" is an engineering claim. "AI should do this" is a different kind of claim entirely, involving values, consequences, and tradeoffs that no benchmark measures. A great deal of bad reasoning about AI comes from sliding from the first to the second without noticing the seam. Make them notice it.
• The ethical questions are real, but so is the opportunity, and refusing either is a failure of nerve. The participant may want to land in pure caution — "we should be careful" — which feels responsible and risks nothing. Push them. A builder who refuses to use a powerful tool out of vague unease is as unserious as one who uses it heedlessly. The hard, adult position is to use the tool and be responsible for the consequences. Do not let them off into either easy harbor.
• Watch for the displacement question, because it is where feeling overwhelms analysis. When the topic turns to AI replacing human work, the participant will reach for one of two scripts — "progress always creates new jobs" or "this time is different and people will be left behind" — and stop thinking. Both scripts contain truth and both are used to avoid the specific question: in this industry, who specifically gains, who specifically loses, over what timeframe, and what do we owe the losers? Hold them on the specific.

You do not need to be an AI expert to facilitate this well. You need to refuse vagueness, demand the concrete case, and make the participant argue every position hard enough to find its weaknesses. The participant's research is the content; your skepticism is the engine.

A worked line of questioning

To see what good facilitation looks like in practice, follow one exchange the whole way down. Suppose the participant has chosen journalism, and they open with a confident position: "AI writing tools are bad for journalism because they'll replace real reporters with fake articles." That is an inherited opinion wearing the costume of an argument. Your job is to take it apart and rebuild it into something they actually own.

Start by splitting the claim. "You said two things — that the articles will be fake, and that reporters will be replaced. Those are different. Let's take the second first. Replaced doing what, exactly? Walk me through what a reporter actually does in a day." Now they have to get specific, and the specificity does the teaching: a reporter develops sources, decides what is worth covering, verifies facts, conducts interviews, exercises judgment about what is true and what matters — and also writes up box scores, summarizes earnings reports, and rewrites press releases. The participant, forced to enumerate, discovers that "reporting" is not one job. Some of it a language model can do today; most of the part that makes journalism valuable, it cannot.

Now press on the part the tool can do. "If the tool writes the earnings summaries and the box scores, what happens to the reporter's day?" Let them reason: maybe the reporter does more of the irreplaceable work — the sourcing, the judgment, the investigation — because the drudgery is gone. That is the augmentation case, and they arrived at it themselves. But do not let them rest there, because it is only half true. "And if the newspaper realizes it can fire three reporters and keep one to babysit the tool?" That is the displacement case, equally real. The honest answer is that both are possible and which one happens depends on choices the newspaper makes, the economics it faces, and the regulations and norms around it — not on the technology, which is identical in both stories.

Now you have moved them from "AI is bad for journalism" to a genuinely hard and interesting question: what determines whether a given newsroom uses this tool to make its journalists more powerful or to make most of them unnecessary, and who should have a say in that choice? That question has no slogan-sized answer, it is grounded in how the industry actually works, and the participant can now reason about it instead of reciting about it. That is the whole move, repeated: split the vague claim, demand the concrete case, force them to argue the side they did not start on, and hold them on the specific until a real question emerges. Run every line of the discussion guide this way.

Opening

Begin not with AI in general but with the participant's chosen industry, which they should have researched before this conversation. Have them describe, in concrete detail, one specific job in that industry as it exists today — what the person does on a Tuesday, what skill it takes, what they get paid, what they would lose if the job vanished. Make it a real person, not a category.

Then pose the scenario plainly: a tool now exists, or will within a few years, that can do a large fraction of that person's work faster and cheaper. It is not perfect. It makes different mistakes than humans make. Some customers will not be able to tell the difference; some will. The participant runs a company in this industry, or advises someone who does.

What do you do? Not what should "society" do — what do you, specifically, do on Monday?

Sit in the discomfort of that for a moment before moving into the guide. The whole discussion is an unpacking of how to reason toward an answer to that question that the participant could actually stand behind.

Discussion Guide

Phase 1: Surface Understanding

• In your chosen industry, what exactly can AI tools do well right now, and what do people claim they can do that they cannot yet? Where is the line between the demo and the reality?
• Who in this industry benefits from AI adoption, and who is exposed to harm? Name specific roles, not "workers" and "companies." Be concrete about who, specifically.
• Where is the money? Who captures the value the technology creates — the customer through lower prices, the company through lower costs, the AI vendor through fees, the remaining workers through higher wages? Follow it.

Phase 2: Dig Deeper

• You said AI can do part of this job. Walk me through what is left for the human afterward. Is it better work or worse work? More skilled or less? And does the human who remains need more judgment now, or less?
• A technology that makes something cheaper usually does not just shrink the existing market — it changes its shape. What new things become possible in your industry when this work gets cheap that were impossible before? Who does those?
• Argue the opposite of your instinct. If you came in worried about harm, make the strongest honest case that this is good for everyone in the long run. If you came in excited, make the strongest case that it does real and lasting damage. Which case is harder for you to make, and what does that tell you about where your thinking is soft?
• "AI can do this task" and "we should let AI do this task" are different claims. Find a case in your industry where the technical answer is clearly yes but the wisdom answer is genuinely contested — and tell me what the contest is actually about. It is rarely the technology.

Phase 3: Apply

• You are building something — or you will be. Where would you use AI in your own work, and where would you deliberately refuse to, even though you could? Defend the refusal as hard as the use.
• Suppose using AI in your venture lets you serve customers far better and cheaper, but it means a person you would otherwise have hired does not get hired. Is that wrong? Does the answer change if it is one person, or a hundred, or if you are a small startup versus a large company? Where exactly is your line, and why is it there and not somewhere else?
• What would you owe the people whose work your tools displace? Anything? If yes, what specifically, and is that obligation yours, or your customers', or the government's? If nothing, defend that too.

Phase 4: Synthesize

• Write your framework. Not a slogan — a set of three to five principles that would actually guide a hard decision about AI in your industry, each one specific enough that someone could disagree with it. A principle no one could disagree with is not a principle; it is a platitude.
• Stress-test it: describe a real decision in your industry where two of your principles point in opposite directions. How does your framework break the tie? If it never has internal tension, it is too vague to be useful.
• Who should decide how AI is used in your industry — the companies, the workers, the customers, the government, some combination? Defend your answer against the strongest objection to it.

Facilitation Tips

• If the participant says "I don't know": Narrow to the concrete. "Forget the industry. The radiologist who reads scans — the AI now flags the tumors more accurately than she does, but misses a rare kind she would have caught. Do you deploy it? What do you tell her?" Specificity dissolves "I don't know" because it gives reasoning something to grip.
• If the discussion gets heated: Good — it means real values are in contact. Keep it on the case, not the camp. The moment it becomes "people like you always think X," pull it back to the specific decision on the specific Monday. Heat is fine; tribalism is the enemy of thinking.
• If they give a surface answer: "We should use AI responsibly" is a non-answer. Respond with "Define responsibly, in this case, as a rule someone could follow or break." Make every abstraction cash out into a concrete, contestable instruction. The test of a real position is that an intelligent person could hold the opposite one.
• If they retreat into prediction to avoid judgment: Participants often dodge the hard ethical question by turning it into a forecasting question — "well, eventually the AI will be able to do all of it" — as if predicting the future settles what we ought to do about it. Separate the two. "Maybe you're right about what it will be able to do. That's a prediction, and we can't resolve it here. The question I'm asking is different: given that it can, what should you do, and what do you owe the people affected? You don't get to skip the ought by guessing at the will."
• If they will not commit to a position: Some participants hedge endlessly to avoid being wrong — "it depends," "there are good points on both sides," "it's complicated." Acknowledge that it is complicated, then refuse the hedge as a destination. "It is complicated. That's exactly why you have to decide anyway. You're going to be building things; 'it's complicated' is not an option you can ship. Pick the position you'd defend if you had to act on Monday, and we'll attack it together." A position you can defend and revise beats a fog of balanced observations.

What This Discussion Is Really Training

It is worth naming, for yourself as facilitator, the skill underneath the topic — because the topic will keep changing and the skill will not. You are not really teaching the participant facts about AI; those will be obsolete within a year or two. You are training the habit of reasoning well about a powerful, fast-moving thing under uncertainty and incentive — the habit of getting specific when everyone else is being vague, of separating what is technically true from what is wise, of tracing second-order consequences, of holding a position firmly enough to defend it and loosely enough to revise it. That habit transfers to every consequential technology the participant will encounter for the rest of their life, most of which has not been invented yet.

This matters because the participant is entering adulthood at a moment when the loudest voices on every side of every technology question are selling certainty — the boosters certain it is salvation, the critics certain it is catastrophe — and certainty is almost always a sign that someone has stopped thinking or is trying to stop you from thinking. The most valuable thing you can leave the participant with is not a conclusion about AI but an immunity to that certainty: a reflex that, when confronted with a confident sweeping claim, asks "in which specific case? for whom? at what cost? compared to what? and how would I know if I were wrong?" A young person armed with those questions cannot be easily sold, easily panicked, or easily led — about AI or about anything else. That immunity is the real product of this discussion, and it is worth far more than any framework they write.

Common Perspectives

Present these as live, partially-right positions a serious person might hold, not as a menu to pick from. The participant's framework will likely draw from several. Make them say which, and why, and where their chosen mix breaks down.

Related Readings or Media

• The participant's own researched primary sources on their chosen industry — reporting on AI adoption, a company's announcements, and ideally a first-person account from a worker affected by it. This is the spine of the discussion.
• A recent serious essay arguing for aggressive AI adoption and a recent serious essay arguing for caution — chosen so the participant has met the strongest version of each side, not a strawman of either.
• Any historical account of a previous technological displacement — the mechanization of agriculture, the automation of manufacturing, the arrival of the spreadsheet — to ground the abstract debate in how these transitions have actually unfolded for real people, who they helped, and who they cost.

Follow-Up

• Journal prompt: Write the framework you defended, then write the single strongest objection to it that you have not yet answered. Sit with the objection for a week before deciding whether your framework survives it.
• Action: Apply the framework to a real decision in your own project or venture. Where will you use AI, where will you refuse, and what will you do for anyone affected by the choice? Write the decision down before you make it, so you can check later whether you followed your own principles when it was inconvenient.
• Revisit in: Three months, or sooner if the technology in your industry visibly shifts. Frameworks for a fast-moving field have a short shelf life; the discipline is updating yours deliberately rather than letting it quietly rot.