Safeguarding artificial intelligence-based network control
An Artificial Intelligence (AI)-based network control system includes an AI system configured to obtain data from a network having a plurality of network elements and to determine actions for network control through one or more Machine Learning (ML) algorithms; a controller configured to cause the actions in the network; and a safeguard module between the AI system and the controller, wherein the safeguard module is configured to one of allow, block, and modify the actions from the AI system.
1. An Artificial Intelligence (AI)-based network control system comprising:
an AI system configured to obtain data from a network having a plurality of network elements and to determine actions for network control through one or more Machine Learning (ML) algorithms;
a controller configured to cause the actions in the network, wherein the actions include configuration changes across one or more layers of the network; and
a safeguard module executed on a one or more processors located between the AI system and the controller, wherein the safeguard module is configured to obtain the actions from the AI system, one of allow, block, and modify the actions from the AI system based on analysis of the actions performed independently of the AI system to determine uncertainty in the actions, and provide the allowed or modified actions to the controller,
wherein the controller is configured to automatically implement the allowed or modified actions in the network based on the independent analysis by the safeguard module.
2. The AI-based network control system of claim 1 , wherein the safeguard module is further configured to obtain its own view of the network independent from the AI system and develop deterministic decisions which are used to compare with the actions from the ML algorithms.
3. The AI-based network control system of claim 2 , wherein the safeguard module is configured to allow the actions if the actions are within the deterministic decisions, block the actions if the actions are not within the deterministic decisions, and modify the actions based on overlap with the deterministic decisions.
4. The AI-based network control system of claim 1 , wherein the safeguard module is further configured to obtain operator input before the one of allow, block, and modify the actions, and wherein the operator input is provided to the ML algorithms for feedback therein.
5. The AI-based network control system of claim 1 , wherein the safeguard module is further configured to compare the actions from the AI system to a result from a deterministic algorithm.
6. The AI-based network control system of claim 1 , wherein the safeguard module is further configured to determine that the actions from the AI system do not violate predetermined conditions.
7. The AI-based network control system of claim 1 , wherein the safeguard module is further configured to interact with a second safeguard module associated with another network.
8. The AI-based network control system of claim 1 , wherein the safeguard module operates on the one or more processors independent from one or more processors for the AI system.
9. An apparatus configured to safeguard an Artificial Intelligence (AI)-based control system comprising:
a network interface communicatively coupled to i) an AI system configured to obtain data from a network having a plurality of network elements and to determine actions for network control through one or more Machine Learning (ML) algorithms and ii) a controller configured to cause the actions in the network, wherein the actions include configuration changes across one or more layers of the network;
a processor communicatively coupled to the network interface; and
memory storing instructions that, when executed, cause the processor to obtain the actions from the AI system via the network interface,
analyze the actions independently of the AI system to determine uncertainty in the actions, and
one of allow, block, and modify the actions from the AI system to the controller based on the determined uncertainty,
wherein the controller is configured to automatically implement the allowed or modified actions in the network based on the independent analysis by the safeguard module.
10. The apparatus of claim 9 , wherein the memory storing instructions that, when executed, further cause the processor to
obtain a view of the network independent from the AI system, and
develop deterministic decisions which are used to compare with the actions from the ML algorithms.
11. The apparatus of claim 10 , wherein the memory storing instructions that, when executed, further cause the processor to
allow the actions if the actions are within the deterministic decisions,
block the actions if the actions are not within the deterministic decisions, and
modify the actions based on overlap with the deterministic decisions.
12. The apparatus of claim 9 , wherein the memory storing instructions that, when executed, further cause the processor to
obtain operator input before the one of allow, block, and modify the actions, and
provide the operator input to the ML algorithms for feedback therein.
13. The apparatus of claim 9 , wherein the memory storing instructions that, when executed, further cause the processor to
compare the actions from the AI system to a result from a deterministic algorithm.
14. The apparatus of claim 9 , wherein the memory storing instructions that, when executed, further cause the processor to
determine that the actions from the AI system do not violate predetermined conditions.
15. The apparatus of claim 9 , wherein the memory storing instructions that, when executed, further cause the processor to
interact with a second safeguard module associated with another network.
16. The apparatus of claim 9 , wherein the safeguard module operates independent from one or more processors in the AI system.
17. A method comprising:
in a processing device having connectivity to i) an Artificial Intelligence (AI) system configured to obtain data from a network having a plurality of network elements and to determine actions for network control through one or more Machine Learning (ML) algorithms and ii) a controller configured to cause the actions in the network, wherein the actions include configuration changes across one or more layers of the network, obtaining the actions from the AI system via the network interface;
analyzing the actions independently of the AI system to determine uncertainty in the actions; and
one of allowing, blocking, and modifying the actions from the AI system to the controller based on the determined uncertainty,
wherein the controller is configured to automatically implement the allowed or modified actions in the network based on the independent analysis by the safeguard module.
18. The method of claim 17 , further comprising
obtaining a view of the network independent from the AI system; and
developing deterministic decisions which are used to compare with the actions from the ML algorithms.
19. The method of claim 18 , further comprising
allowing the actions if the actions are within the deterministic decisions;
blocking the actions if the actions are not within the deterministic decisions; and
modifying the actions based on overlap with the deterministic decisions.
20. The method of claim 17 , further comprising
obtaining operator input before the one of allow, block, and modify the actions; and
providing the operator input to the ML algorithms for feedback therein.