AI care and ethics at 0.338: bias and trust risks in the AI bubble

What does AI care and ethics at 0.338 measure in the AI bubble?

Care belongs to domain B, relation and living, with trust at 0.450 and communication at 0.305. It measures whether an AI system looks after the people it affects: employees, customers, and the public. A value of 0.338 puts the enterprise AI market in the fragile zone, and it is the second lowest of the five dimensions that map to failure causes. The study computed it in April 2026 from public data on the enterprise AI market.

The study reads the number as an intent. Companies deploy AI mainly to cut human costs and not to create value, so the system optimizes efficiency at the expense of an ethical principle. The score does not judge AI as good or bad. It records how much attention affected people receive in the way AI gets deployed.

Why do ethical issues and bias appear in 18% of failed AI projects?

The study ranks ethical issues and bias fifth among the causes of failure, at 18%. The cause ranks last of the five, yet one ethical incident can cost a company its customers’ trust. A discovered bias also reaches the communication dimension, which sits only 0.005 above the critical threshold.

Bias often enters through data and through the way models are used. The study notes that language models give false answers in 0.7% to 30% of cases, and up to 67% in a medical context (JMIR 2025). In sensitive domains such as health, hiring, or credit, a wrong answer harms a person and creates a legal and reputational cost for the company.

What does the care breakdown of 0.30 reveal about AI ethics?

The study measures the intent of care with a score called Care(S), which reads 0.30. Inside it, the absence of harm scores 0.40, the protection of impacted people scores 0.25, and regeneration, which means retraining, scores 0.20. AI is not harmful in itself, according to the study, yet the follow-up for people it affects is weak.

The pattern is clear. The first element looks acceptable, and the two elements that act after deployment sit far lower. A company can avoid direct harm and still leave employees without a path to a new role. Retraining at 0.20 is the weakest link in the whole care chain.

How do job losses without retraining weaken the AI bubble case?

McKinsey estimates that 30% of hours worked could be automated by 2030, and the study counts 300 million jobs disrupted. Fewer than 15% of companies have a retraining plan for the employees concerned. The gap between the size of the change and the size of the response is the care signal.

For the AI bubble, the issue is the durability of the gains. The projected productivity gain of +40% depends on people who stay engaged and skilled. If a company removes roles without building new skills, it risks losing the knowledge that makes its AI projects work. Skills built inside the company stay with the company, which is why retraining protects the investment as well as the people.

How large should an AI care retraining budget be?

The fourth recommendation of the study is a skills plan for each automated role, with a budget of 2% to 5% of total AI investment. The study states the principle in one sentence: AI must increase humans, not replace them. The plan starts by listing each role the project changes. Each plan names the new skill, the training route, and the date by which the person takes up the new role.

An illustrative case shows the scale. On an AI program of $10 million, a budget of 2% to 5% means $200,000 to $500,000 for retraining. The study expects the care dimension to rise from 0.338 to 0.50 with this plan, and Care(S) from 0.30 to 0.40, which raises the ethical safeguard significantly.

How does the ethical safeguard protect AI decisions in the EQOS Asia method?

The EQOS Asia engine applies an ethical safeguard to every result. In the AI market study the safeguard is active at 0.067, and the passage is authorized. If the safeguard turns negative, the engine issues no recommendation, since the study treats it as the absolute priority of the system.

This rule gives care a structural role. A recommendation that harms the people it concerns does not leave the engine, whatever its financial appeal. The method page describes how the safeguard fits with the audit, the modeling, and the prediction, so that ethics sits inside the calculation and not after it.

How should you read weak AI care as an AI bubble signal?

Weak care is a signal of fragile returns. Savings that come from cutting people without retraining them can erode when skills run short. If care stays near 0.338 while adoption grows, the gains of AI projects rest on a thin base. This page is general information, not investment advice, and it reflects the study as of October 2026.

To score care in your own organization, the Λ simulator estimates your φ(S) in about 10 minutes, and a 30-minute call with the EQOS Asia team can review your retraining plan. You leave with a care value and a budget range for the skills plan of each automated role. The score then becomes part of your project brief.

Frequently asked questions

About the author

Stéphane Jambu

President, EQOS Forecast Asia

Stéphane Jambu leads EQOS Forecast Asia and works as a consultant in e-commerce, search visibility, and AI. LinkedIn profile