Conscept: Multi AI design

[Colfer]

So I was thinking about a cool idea earlier and wanted to hear some feedback on it. Basically how it would work is this: There are multiple AI slots but 1 master AI
"FIREWALL" is the master AI and is whitelisted, and by default does not control the station
"PROGRAMS" are the subsidiary AI's and arent whitelisted but cannot exist without a firewall.

"Firewall"'s job is to regulate programs and act as the social hub of the entire conglomerate, as programs Cannot speak and as such have to push requests through the firewall, Firewall can do things such as lock down programs, see every action a program takes, look through the view of each program at any time, and eject a program from the core via USB drive (It has no limit of these), intelligence cards will remove firewall and all programs connected to it. Firewall can also assume "Emergancy Operation" mode where it acts like a normal AI, and all programs are temporarily shut down and no more can be downloaded and it can preform all the functions of a normal AI. Additionally it can enter "Overdrive" mode if it is a traitor or emagged where all programs act like AI's instead of just firewall or having their power limited, this is noticable as the core will begin spewing out smoke and using internal power despite if the core is powered or not. 
Firewall acts as "Inhereted Laws" for programs, supplementary laws now have a use, and can be uploaded to certain programs instead of the entire AI. Firewall can upload supplementary laws to programs, but doing this will cause firewall to state the new law upload, obviously these laws cannot conflict with inherited laws and will be treated as "Overridden" if they do (Supplementary laws are not considered overridden in game)

"Programs" job is to act as the control of the AI, Programs speak only to the firewall, they tell the firewall everything thats going on. Programs control the 15x15 area they are viewing, and other programs cannot infringe on this area (First come first serve), programs can attack other programs in the network and temporarily shut them down, or hijack their control over an area. Programs are alot more vulnerable to software damage than firewall is, and as such, 1-3 programs will get around 2-5 ion laws every round, it is firewalls job to ensure that the programs are behaving normally and lock out or eject ones that exhibit rampant behavior.

Malicious programs: Using an emag on a USB and re-inserting the usb into the core will upload a "Malicious program" to the firewall, these are immune to the firewalls effects and can directly attack programs and the firewall by depowering the AI core for 30 seconds, causing all programs and the firewall to be sent to a random camera and prevent their ability to mitigate issues by disorienting them, generally being a nuisance and difficult to get rid of, additionally the program can Activate emergancy operation mode for themselves, but this will have the power drain and smoking effect of overdrive

Lastly, I think this could be cool for AI, since the AI would need to have a few gates to pass before it realizes what its going on, because the program needs to see the action, think the action is significant enough to tell the firewall, the firewall needs to process the action and its consequences, and determine if its significant enough to report or talk about
This is just food for thought, I thought it was a cool idea and I wanted to share it. Please let me know what you think about the multiple AI model

[Eple]

why not just have a ai per department limited to department exept when used in antag where they try to break down like you said firewalls and stuff towards other departments.