Differences

This shows you the differences between two versions of the page.

--- en:start [2025/12/13 19:10] – admin
+++ en:start [2026/06/05 12:37] (current) – [4.4 Symbolic Cognitive Service] admin
@@ Line 252: / Line 252: @@
 ----
-===== 5. The Three AIs Together =====
+==== 4.4 Symbolic Cognitive Service ====
-==== Why a single approach is never enough ====
-  * **ML** provides prediction.
+Amsafis has developed an executable knowledge layer that enables
-  * **ES** provides explanation.
+direct and operational consultation of expert models.
-  * **RAG-LLM** provides understanding of documents.
+It is not a chatbot nor a generative AI application.
+It is a service that exposes formalized knowledge so it can be used by:
+  * people (manual use)
+  * scripts and integrations
+  * enterprise software
+  * future automation systems or agents
+The objective is to transform expert knowledge into an executable,
+traceable, and reusable tool.
+=== How to use the technical sandbox ===
+In addition to the manual use available through the website, the service can be tested
+by executing commands from free tools such as:
+  * Git Bash
+  * Visual Studio Code (integrated terminal)
+(Other consoles may work, but syntax may vary.)
+These tools allow direct queries to be sent to the knowledge engine.
+==== Rare Disease Diagnosis ====
+=== Example of phenotype-based diagnosis ===
+<code bash>
+curl -X POST https://diseases-non-interactive.amsafis.com/mcp/dispatch \
+ -H "Content-Type: application/json" \
+ -d "{\"task\":\"diagnose\",\"payload\":{\"symptoms\":[\"Myopia\"]}}"
+</code>
+=== Example of comparative analysis between two diseases ===
+<code bash>
+curl -X POST https://diseases-non-interactive.amsafis.com/mcp/dispatch \
+ -H "Content-Type: application/json" \
+ -d "{\"task\":\"intersect\",\"payload\":{\"disease_a\":\"biotinidase_deficiency\",\"disease_b\":\"holocarboxylase_synthetase_deficiency\"}}"
+</code>
+=== Example of disease knowledge exploration ===
+<code bash>
+curl -X POST https://diseases-non-interactive.amsafis.com/mcp/dispatch \
+ -H "Content-Type: application/json" \
+ -d "{\"task\":\"review\",\"payload\":{\"disease\":\"biotinidase_deficiency\"}}"
+</code>
+==== GDPR Compliance ====
+=== Example of GDPR personal data compliance query ===
+The following query asks whether an IP address is personal data under GDPR and whether compliance obligations apply:
+<code bash>
+curl -s -X POST https://agent-gdpr.amsafis.com/mcp/dispatch \
+  -H "Content-Type: application/json" \
+  -d "{\"connector\": \"gdpr_check\", \"params\": {\"data\": \"ip_address\"}}"
+</code>
+The knowledge base encodes the personal data classification rules from the [[https://www.cnil.fr/en/sheet-ndeg1-identify-personal-data | CNIL Developer's Guide, Sheet 1]], which is an authoritative source for GDPR data identification in Europe.
+Five connectors are available:
+  * gdpr_check — is a data type personal data? Does GDPR apply?
+  * gdpr_consent — does it require explicit consent?
+  * gdpr_anonymous — is it truly anonymous?
+  * gdpr_pseudonym — is it pseudonymised, and what are the implications?
+  * gdpr_query — advanced: send any ground Prolog query directly
+Full connector list and parameter schema:
+<code bash>
+curl https://agent-gdpr.amsafis.com/mcp/connectors
+</code>
+This environment constitutes a **Knowledge Tool Provider**:
+a reusable component that can be integrated into more complex architectures
+without losing model transparency or knowledge governance.
+----
+===== 5. From AI components to Agents =====
+The previous section shows how a knowledge service can be accessed programmatically using a simple curl request based on the Model Context Protocol (MCP).
+This is not only a technical detail. It defines how different AI components can be **connected and executed as interoperable services**.
+An **Agent** builds on this capability.
+Rather than calling a single service, an Agent:
+  * interprets a request,
+  * selects the appropriate components,
+  * and coordinates their execution through structured interactions such as MCP calls.
+In this context, the **Model Context Protocol (MCP)** provides a standardized way to exchange structured inputs and outputs between components, enabling:
+  * reproducible execution,
+  * traceable interactions,
+  * and integration across heterogeneous systems.
+An Agent may combine:
+  * Expert Systems (rule-based reasoning),
+  * Machine Learning models (prediction),
+  * and LLMs (interpretation and generation),
+using MCP as the communication layer between them.
+This transforms isolated capabilities into a **coherent execution flow**, where each component contributes to the final result.
+For example:
+  * an LLM can interpret a clinical description,
+  * map it into structured concepts,
+  * trigger a knowledge service through an MCP request,
+  * and integrate the response into a final output.
+This is not a new AI pillar.
+It is a way to **operationalize the integration** described in the previous sections.
+In practice, each Agent is **designed for a specific use case**, with different levels of complexity:
+  * simple pipelines with minimal orchestration,
+  * or multi-step processes combining reasoning, prediction, and knowledge retrieval.
+This approach enables the transition from:
+  * isolated tools
+     to
+  * integrated, executable decision systems.
-In real organisations, high-stakes problems require all three:
-  * ML finds the signal
-  * ES encodes the decision logic
-  * RAG explains, documents and supports human interpretation
-**This tripartite architecture is the core of amsafis.**