Only an Oath Speaks in the First Person: The Fourth Value of the DoGoDa Oath

A fire drill works because nothing is burning yet. You rehearse the decision before the day it costs something. If you ask me, that is exactly what an oath is for, and self-control, the fourth value in the data professional’s oath I have been reading and writing through since July, needs that rehearsal more than any of the other three I have covered so far.

What each instrument binds

Three instruments are on the table when a profession tries to hold itself to a standard, and they bind different parties. A code of conduct describes the good professional in the third person: you read it the way you read a job description, without feeling personally addressed. A law binds your employer, and it typically arrives years after the damage is done, landing on a legal department rather than on the person who made the call. An oath is the only one of the three written in the first person. In my own words, self-control’s sentence reads: I will weigh this capability against its alternatives before it ships. That distinction is the whole argument, and everything that follows is a test of whether it holds up.

The Dutch test, with teeth

The strongest evidence against me is Dutch, and I would rather carry it than route around it. Since 2015, under the Wijzigingswet financiële markten 2015, Dutch bank employees have been legally required to take the banker’s oath, backed by a disciplinary body, the Foundation for Banking Ethics Enforcement, that can investigate a breach and impose real sanctions. It is not a gesture. And the measured result is sobering: research published in 2016 (Loonen and Rutgers, Journal of Banking Regulation) found trust in the oath low, knowledge of the concept low, and bank employees themselves among the most opposed to it. I read that as a finding about the word mandatory, not about oaths in general. An obligation imposed on you by an employer and a regulator is the opposite of a commitment you take. The Dutch banking sector did not disprove the value of an oath. It ran the experiment on the version that removes the one thing that makes an oath work.

Handed over, or freely taken

That reading has research behind it, from two directions. McNamara, Smith and Murphy-Hill (ESEC/FSE 2018) ran 168 participants, students and professional developers, through eleven ethical vignettes and found that explicitly instructing them to consider the ACM code of ethics had no observed effect on their decisions. A document, handed over, changed nothing measurable. Commitment research points the other way, within its own scope: Jacquemet, Luchini, Rosaz and Shogren (Management Science, 2019) found that an oath taken freely, expressed publicly and signed shifted how truthfully people answered in laboratory settings, specifically when the situation made truthfulness meaningful to the person taking it. That condition matters and I am not going to drop it for a cleaner sentence: the effect showed up when the stakes were made to feel real, not automatically. Put the two studies together and the mechanism becomes visible. A code you are handed is background noise. An oath you take, in public, before the moment that tests it, is a rehearsal you have already run once when it finally costs something.

Restraint leaves nothing behind

Here is why self-control specifically needs a name and a standing that the other seven values in this oath do not need in the same way. Dignity shows up in how a data row treats a person. Interpersonal trust shows up in numbers someone can still explain. Every one of those values leaves an artifact behind: a documented model, a dataset handled with care, a decision you can walk someone through. Successful restraint produces no ticket, no deliverable and no thanks, and nobody defends a non-event in a room measured on what shipped. There is no line item for the capability you left in the pilot. That is precisely why it needs an instrument that survives the absence of a receipt, and a first-person sentence is the cheapest one available.

The act is real when someone actually performs it, and it is expensive rather than symbolic. Microsoft’s own announcement, dated 21 June 2022, retired general-purpose inference of emotion, gender and age from the Azure Face API, on the company’s stated Responsible AI grounds, with paying customers already using the capability. I am reading this as Microsoft describing its own decision, not as an independent verdict on the technology, and it is worth naming what makes the case land: the thing worked, people were paying for it, and the company took it out anyway. The European Union drew a related line in law almost three years later. Article 5(1)(f) of the AI Act prohibits inferring emotions of a person in the workplace or in education, with narrow medical or safety carve-outs, and that prohibition has applied since 2 February 2025. The sequence is the point, not the regulation itself: a vendor made this judgment on its own well before a legislature made it compulsory, which is exactly the order self-control asks a data professional to work in.

Yes, an oath needs a checklist

The sharpest objection to everything above comes from inside the field, and it deserves its full turn rather than a summary sentence. DJ Patil, Hilary Mason and Mike Loukides, in “Of oaths and checklists” (O’Reilly Radar, 2018), argue that an oath is a set of broad principles, and broad principles do not tell anyone what to do on a Tuesday deployment review. A checklist, built on Atul Gawande’s model, is what actually connects a principle to a practice: concrete, repeatable, checkable by someone other than the person who wrote it. All true. But a checklist can only enforce a standard that already exists somewhere, and the oath is where that standard gets set, stated in the first person, before the sprint that will eventually test it. The two are not competitors. The checklist is the Tuesday version of the sentence you already signed.

The arithmetic

Dignity, the first value in this oath, lives in how a data row treats a person. Interpersonal trust, the second, lives in numbers someone can still explain to the person acting on them. Justice, the third, lives in what your data product feeds into once it leaves your hands. All three are visible in something that exists: a model, a number, a system. Self-control decides whether the thing exists at all, which makes it the value that determines whether the other seven ever get tested in the first place. Of the eight values in this oath, this is the one I most want rehearsed like a fire drill, precisely because the day it counts gives no warning.

This series started by asking why a data professional needs an oath at all, when nobody is required to sign one to keep working. Self-control is the value that answers that question from the inside: the moment it decides is the one moment in the whole chain where you are unambiguously the person in the room.

Happy to think this through with you.