> For the complete documentation index, see [llms.txt](https://docs.oomus.org/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.oomus.org/compliance/ai-ethics.md).

# AI & Ethics

Oomus CampaignID integrates artificial intelligence components to improve service quality. This page transparently describes the use of AI, its limitations, and the ethical safeguards in place.

***

## Uses of AI in Oomus CampaignID

Oomus CampaignID uses artificial intelligence in **two specific, clearly bounded contexts** :

### 1. Anomaly detection — IsolationForest

**Component** : Premium module "AI Fraud Detection"

**Algorithm** : IsolationForest (unsupervised learning algorithm for anomaly detection)

**What it analyzes** :

* Card scan patterns (frequencies, volumes, times)
* Atypical distribution behaviors
* Unusual account activities (generation volumes, API requests)

**What it does NOT analyze** :

* Beneficiaries' personal data (name, date of birth, health)
* Content of generated cards
* Medical or clinical data

**Result** : Reporting of behavioral anomalies for human investigation. No automatic decision is made.

***

### 2. MPI deduplication — Probabilistic engine

**Component** : MPI Sovereign Identity Engine (included in all plans)

**Algorithm** : Composite similarity score calculation (Jaro-Winkler + African normalization + semantic matching)

**What it analyzes** :

* First name, last name (phonetic and orthographic similarity)
* Date of birth (exact or near match)
* Gender, geographic district
* External identifiers (DHIS2 UID)

**What it is NOT** :

* A biometric system (no facial recognition, fingerprints, iris)
* A profiling system (it does not "profile" the beneficiary)
* A predictive identification system (it does not predict behavior)

**Result** :

* Score ≥ 0.95 → automatic linking (high certainty)
* Score 0.75–0.94 → **mandatory human review** before any merge
* Score < 0.75 → creation of a new identity

***

## Ethical safeguards

### No profiling of beneficiaries

Oomus CampaignID does not build behavioral, socio-demographic, or medical profiles of beneficiaries for commercial or analytical purposes. The collected data is used solely for card issuance and verification.

### No automatic decisions affecting beneficiaries

In accordance with GDPR principles (Article 22) and ethical best practices, **no significant decision directly affecting a beneficiary is made automatically** by an Oomus CampaignID algorithm:

* The merging of two MPI identities always requires human validation (if score < 0.95)
* Card revocation is always an intentional human action
* Access to care never depends on an Oomus algorithmic decision

### Systematic human oversight

All outputs from Oomus AI systems are **recommendations** intended to be reviewed by a human operator:

| AI output                            | Level of human oversight                            |
| ------------------------------------ | --------------------------------------------------- |
| Scan anomaly detected                | Alert → human investigation required                |
| Probable MPI match (score 0.75–0.94) | Review queue → human decision required              |
| Certain MPI match (score ≥ 0.95)     | Automatic linking authorized (calibrated threshold) |

### Sensitive data handling

Before any analytical processing or model training, an **automatic safeguard** excludes attributes belonging to 7 sensitive categories:

| Category      | Why excluded                                  |
| ------------- | --------------------------------------------- |
| HIV/AIDS      | Highly stigmatizing data                      |
| STIs          | Highly stigmatizing data                      |
| Tuberculosis  | Sensitive health data                         |
| Mental health | Sensitive health data, risk of discrimination |
| Serological   | Sensitive biological data                     |
| Biometric     | Irreplaceable data, permanent security risk   |
| Financial     | Socio-economic data, risk of discrimination   |

***

## Algorithmic bias — Measures taken

The MPI deduplication engine was specifically designed to reduce algorithmic bias related to West African names:

### Identified issue

Standard text similarity algorithms (simple Levenshtein, English Soundex) perform very poorly on African names:

* Multiple spelling variants: Ibrahim / Ibrahima / Brahim / Braima
* First/last name order culturally variable
* Frequent compound names
* Transliterations from Wolof, Pulaar, Manding

### Implemented solution

The MPI engine includes a **specialized normalization layer** :

* West African name variants dictionary
* Detection and handling of compound names
* Robustness to diacritics and apostrophes
* Regression tests on representative anonymized datasets

***

## Algorithmic transparency

Oomus CampaignID is committed to transparently documenting the algorithms used and their limitations:

* The algorithms used are established, auditable methods (Jaro-Winkler, IsolationForest)
* Decision thresholds are documented and configurable by administrators
* Algorithm results are explainable (the fields that contributed to the score are indicated)

***

## Ethical roadmap

Oomus CampaignID is committed to maintaining and improving its ethical safeguards:

* **Regular audits** of algorithmic bias in the deduplication engine
* **Consultation** with end-user communities (health workers, beneficiaries) on AI use
* **Publication** of algorithm performance metrics (precision, recall, false positives)
* **Regulatory monitoring** on the European AI Act and African equivalents

***

## Contact

For any questions about the use of AI in Oomus CampaignID:

* **Email** : <ceo@oomus.org>
* **Documentation** : this page is updated with every significant evolution of the AI components

***

## Next steps

* [Legal compliance](/compliance/legal-compliance.md)
* [Data Protection](/security/data-protection.md)
* [Sovereign MPI identity](/features/mpi-sovereign-identity.md)


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.oomus.org/compliance/ai-ethics.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.