Framework for trustworthiness
Method comprising: receiving a trust level requirement for a service; translating the trust level requirement into a requirement for at least one of a fairness, an explainability, and a robustness of a calculation performed by an artificial intelligence pipeline related to the service; providing the requirement for the at least one of the fairness, the explainability, and the robustness to a trust manager of the artificial intelligence pipeline.
1 . Apparatus comprising:
one or more processors, and
memory storing instructions that, when executed by the one or more processors, cause the apparatus to perform:
receiving an intent of a service;
translating the intent into a trust level requirement for an artificial intelligence pipeline related to the service, wherein the artificial intelligence pipeline comprises a data source manager configured for data collection, a model training manager configured for hyperparameter tuning, and a model inference manager configured for model evaluation;
providing the trust level requirement to an artificial intelligence trust engine related to the artificial intelligence pipeline.
2 . The apparatus according to claim 1 , wherein the instructions, when executed by the one or more processors, further cause the apparatus to perform:
translating the intent into at least one of a quality metrics of the service and a quality metrics of the artificial intelligence pipeline;
providing the quality metrics of the service to a management and orchestration function of the service if the intent is translated into the quality metrics of the service; and
providing the quality metrics of the artificial intelligence pipeline to a pipeline orchestrator of the artificial intelligence pipeline if the intent is translated into the quality metrics of the artificial intelligence pipeline.
3 . The apparatus according to claim 1 , wherein the instructions, when executed by the one or more processors, further cause the apparatus to perform:
mapping the trust level requirement on an artificial intelligence quality of trustworthiness class identifier, wherein the artificial intelligence quality of trustworthiness class identifier has one of a limited number of predefined values, wherein
the trust level requirement is provided to the artificial intelligence trust engine as the artificial intelligence quality of trustworthiness class identifier.
4 . The apparatus according to claim 1 , wherein the instructions, when executed by the one or more processors, further cause the apparatus to perform:
receiving an indication of a risk level of the service, wherein
the risk level indicates a risk caused by a failure of the service;
the trust level requirement depends additionally on the risk level of the service.
5 . Apparatus comprising:
one or more processors, and
memory storing instructions that, when executed by the one or more processors, cause the apparatus to perform:
receiving a trust level requirement for a service;
translating the trust level requirement into a requirement for at least one of a fairness, an explainability, and a robustness of a calculation performed by an artificial intelligence pipeline related to the service, wherein the artificial intelligence pipeline comprises a data source manager configured for data collection, a model training manager configured for hyperparameter tuning, and a model inference manager configured for model evaluation;
providing the requirement for the at least one of the fairness, the explainability, and the robustness to a trust manager of the artificial intelligence pipeline.
6 . The apparatus according to claim 5 , wherein the instructions, when executed by the one or more processors, further cause the apparatus to perform:
receiving at least one of an actual fairness in providing the service, an actual explainability in providing the service, an actual robustness in providing the service, an explanation, and an artifact from the trust manager;
storing the received at least one of the actual fairness, the actual explainability, the actual robustness, the explanation, and the artifact in a database accessible to a network operator.
7 . The apparatus according to claim 6 , wherein the instructions, when executed by the one or more processors, further cause the apparatus to perform:
providing the received at least one of the actual fairness, the actual explainability, the actual robustness, the explanation, and the artifact to the network operator.
8 . The apparatus according to claim 7 , wherein the instructions, when executed by the one or more processors, further cause the apparatus to perform:
monitoring if a request is received to provide the at least one of the actual fairness, the actual explainability, and the actual robustness to the network operator;
inhibiting the providing of the at least one of the actual fairness, the actual explainability, and the actual robustness to the network operator if the request is not received.
9 . The apparatus according to claim 5 , wherein the instructions, when executed by the one or more processors, further cause the apparatus to perform:
mapping the translated requirement for the at least one of the fairness, the explainability, and the robustness on a quality of trustworthiness, wherein the quality of trustworthiness has one of a limited number of predefined values.
10 . The apparatus according to claim 5 , wherein the trust level requirement is expressed by an artificial intelligence quality of trustworthiness class identifier, wherein the artificial intelligence quality of trustworthiness class identifier has one of a limited number of predefined values.
11 . The apparatus according to claim 5 , wherein the instructions, when executed by the one or more processors, further cause the apparatus to perform:
discovering the trust manager related to the service based on an identification of the service.
12 . Apparatus comprising:
one or more processors, and
memory storing instructions that, when executed by the one or more processors, cause the apparatus to perform:
receiving, from a trust engine, a requirement for at least one of a fairness, an explainability, and a robustness of a calculation performed by an artificial intelligence pipeline related to a service, wherein the artificial intelligence pipeline comprises a data source manager configured for data collection, a model training manager configured for hyperparameter tuning, and a model inference manager configured for model evaluation;
mapping the received requirement for the at least one of the fairness, the explainability, and the robustness on at least one of a specific requirement for a data source manager of the artificial intelligence pipeline related to the service, a specific requirement for a training manager of the artificial intelligence pipeline related to the service, and a specific requirement for an inference manager of the artificial intelligence pipeline related to the service;
providing the specific requirement for the data source manager to the data source manager and obtaining an actual value of the at least one of the fairness, the explainability, and the robustness from the data source manager if the received requirement is mapped to the specific requirement for the data source manager;
providing the specific requirement for the training manager to the training manager and obtaining an actual value of the at least one of the fairness, the explainability, and the robustness from the training manager if the received requirement is mapped to the specific requirement for the training manager;
providing the specific requirement for the inference manager to the inference manager and obtaining an actual value of the at least one of the fairness, the explainability, and the robustness from the inference manager if the received requirement is mapped to the specific requirement for the inference manager;
comparing at least one of the received actual values with the corresponding requirement;
issuing an error notice if the at least one of the actual values does not fulfill the corresponding requirement.
13 . The apparatus according to claim 12 , wherein each of the specific requirements comprises at least one of a required algorithm, a configuration, and a measuring target.
14 . The apparatus according to claim 12 , wherein the instructions, when executed by the one or more processors, further cause the apparatus to perform:
providing at least one of the obtained actual values to the trust engine.
15 . The apparatus according to claim 14 , wherein the instructions, when executed by the one or more processors, further cause the apparatus to perform:
monitoring if a request for the at least one of the received actual values is received from the trust engine;
inhibiting the providing of the at least one of the received actual values to the trust engine if the request is not received.
16 . The apparatus according to claim 12 , wherein the instructions, when executed by the one or more processors, further cause the apparatus to perform:
monitoring if the error notice is issued;
instructing, if the error notice is issued, at least one of the data source manager, the training manager, and the inference manager to perform a corrective action in order to make the actual values in future to fulfill the corresponding requirement.