Skip to main content

training data bias

Query use case​

Is there evidence that training data on which this system has (or is free of) bias?

Schemas used​

Pseudo code​

FUNCTION ai_system_has_data_bias(AI_System_ID)
    CREATE empty list Attestations

    // Step 1: Retrieve dataset verification credentials linked to the AI system
    SET Data_VC_IDs = get dataset verification credentials associated with AI_System_ID

    // Step 2: Identify bias attestations in each dataset verification credential
    FOR EACH Data_VC_ID in Data_VC_IDs DO
        SET Attestations_List = get bias attestations linked to Data_VC_ID

        FOR EACH Attestation in Attestations_List DO
            IF Attestation is of type "bias" THEN
                SET Component_Hash = Attestation's component hash
                SET Bias_Details = Attestation's details
                
                ADD ({"component": Component_Hash, "data_vc_id": Data_VC_ID}, Bias_Details) TO Attestations

    // Step 3: Return all bias attestations found
    RETURN Attestations
END FUNCTION

Explanation​

  1. Find relevant data sources:

    • Retrieve the configuration verification credential (ConfigVcId) for the AI system.
    • Extract the weights verification credential (WeightsVcId) used in training.
    • Ensure that the WeightsVcId is classified as "Weights".
    • Trace back to the training system that produced these weights.
    • Identify the datapack used in the training process.

  2. Extract the list of Data Verification Credentials (DataVcIds) used in training from the datapack.

  3. Identify attestations that indicate bias:

    • For each DataVcId, retrieve its bias attestations.
    • If an attestation is labeled as "bias", extract its component_hash and BiasDetails.

  4. Return a list of bias attestations:

    • Each entry consists of a tuple:
      • Component information (component hash and DataVcId).
      • Bias details describing the detected bias.

Query​

  • ai_system_has_data_bias(AiSystemId, Attestations) link to query
  • link to simulator

Notes​

This assumes we have a trusted method to identify bias on a data set.