# AI Ethics Washing: Are Corporate Principles Enough?

Author: Norbert Biedrzycki
Published: 2020-12-15
Language: en
URL: https://norbertbiedrzycki.pl/en/ai-ethics-washing-are-corporate-principles-enough/
Summary: Earlier this month Timnit Gebru, joint leader of Google's Ethical AI team and one of the most respected researchers in the field, left the company after a dispute over a research paper examining…

Earlier this month Timnit Gebru, joint leader of Google's Ethical AI team and one of the most respected researchers in the field, left the company after a dispute over a research paper examining the risks of large language models. Thousands of Google employees and academics signed letters of protest. Whatever the full story, the episode raised an uncomfortable question for the entire industry: what happens when ethical principles meet commercial priorities?

## A flood of principles

The last few years have produced an extraordinary number of AI ethics documents. A 2019 study in Nature Machine Intelligence by Anna Jobin and colleagues identified 84 sets of AI ethics guidelines published by companies, governments and institutions. AlgorithmWatch's global inventory now lists well over a hundred.

They converge on a familiar vocabulary: fairness, transparency, privacy, human oversight, accountability, beneficence. Almost every major technology company has published its own version.

The convergence looks like progress. It is also a warning sign. When everyone agrees on the principles, the principles are no longer what distinguishes responsible organizations from irresponsible ones.

## The implementation gap

The real test is not what a company says. It is what it is willing to give up. A principle that has never stopped a product launch, delayed a contract or cost revenue is a statement of intent, not a governance mechanism.

There are signs of genuine change. In June, IBM announced it would no longer offer general purpose facial recognition, and Amazon and Microsoft paused sales of the technology to US police. Those were decisions with commercial consequences.

But in many organizations the gap between principles and practice is wide. Ethics boards without decision rights. Review processes that are optional. Responsible AI teams located far from product teams and without influence over their roadmaps. Incentives that reward speed to market and nothing else.

That is what ethics washing looks like. Not hypocrisy, usually, but a structure in which good intentions have no mechanism to win an argument.

## From values to mechanisms

From a governance perspective, ethical principles become real only when they are translated into four things.

**Decision rights.** Someone with the authority to stop or change a system, and a clear process for using it.

**Controls.** Concrete checks built into the development and procurement lifecycle, not attached at the end.

**Measurement.** Tests and metrics that can show whether a system meets the principle.

**Accountability.** An executive whose performance depends on it.

This is the same distinction I would draw for any other corporate commitment. A company does not have a safety culture because it has a safety policy. It has one when a line manager can stop production without fear of consequences. The same will eventually be true of AI.

> *Leadership question: Is ethical AI a genuine organizational capability or simply a new form of corporate reputation management?*
