There is no neutral AI. Fourteen short lessons that turn that reframe into a working Moral Impact Review, a hierarchy of duties, a named owner with a kill switch, and a dashboard your board can read.
In 2019 a healthcare model went looking for the patients who most needed extra care. It used future cost as a stand-in for sickness, because sick people spend more. It worked exactly as designed, and it systematically underestimated how sick Black patients were, because the data recorded that they had received less care for the same illness. Nobody wrote a discriminatory rule. Nobody intended any of it. A system can be technically flawless and morally disastrous at the same time, and that is the problem this course exists to solve.
Every model you deploy is a hidden decision-maker already making moral choices in your name. It trades revenue against fairness, speed against dignity, and short-term margin against long-term trust, whether or not anyone chose those trades out loud. Design your values into the system, or the system inherits them from data that is a frozen record of how things used to be done.
This is a build rather than a survey. Each concept lesson has a worked partner where a real decision gets made, and every promised outcome ships as a filled artifact rather than a description of one. The course is built on Jeff Yelton's The Moral Code of AI, including the engagements he got wrong and says so.
The real deliverable is the Moral Impact Review: a four-question gate that no high-risk model passes without answering. Whose values are in the objective function, who bears the cost of the errors, who holds override authority, and what number takes the system offline later. Most models that reach a front page would have failed one of those four questions in a meeting nobody held.
CEOs and executives accountable for what their AI systems decide, including the decisions they never saw.
COOs, general counsel, and risk leaders turning "responsible AI" from a value statement into an engineered practice with a gate in it.
CIOs, CTOs, and product leaders who set the objectives models optimize and choose the trade-offs they make.
Expected time: about 70 minutes across fourteen short lessons (four to seven minutes each), plus the time you spend filling in your own artifacts.
The pocket question that reframes every AI review you sit in: when the system is wrong, who pays the price?
The three doors bias actually walks through, and why deleting the sensitive column closes none of them.
A provenance and credit standard for one generative use case, covering what you will never generate and what you disclose.
The Moral Stack as a diagnostic: which of engine, brakes, and steering wheel is missing from the system in front of you.
The three questions that expose the efficiency trap, plus a red-team prompt list you run before a line of code is written.
A Risk Appetite Statement that names your acceptable trade-off in explicit numbers, with a signature on it.
A hierarchy of duties, ranked, with the fairness definition you chose on purpose and the reasoning documented.
The Character Audit run on one live system, and the brand gap it exposes.
A drift monitoring plan with smoke alarms and a canary cohort, because a system audited fair on day one can be discriminatory on day 365 with identical code.
The four-question Moral Impact Review, filled in for a real system, positioned as a gate rather than a consultation.
A named model owner, an ambiguity zone where the machine hands over, and a kill switch delegated downward.
A three-line Moral Impact Dashboard your board can act on, including the correction cost that gets the CFO's attention.
The assembled Moral Architecture, and the three questions that replace "does the system work?"