Tuesday, March 15, 2005

Authentication of autonomous agents

Something I was thinking of today... I'm not sure our current forms of electronic authentication are really going to cut it when autonomous agents come into play. Imagine you work for a large company that restricts physical access to its facilities (in my case, a keycard), as well as to its network (a password while at work, and my keycard and PIN while away from work). What about artificially intelligent robots on my behalf? Obviously, this isn't a big problem at the moment. But, imagine a situation where you have robots doing simple tasks, such as delivering interoffice mail, or a personal robot who goes to get a CD from your car. How does authentication work in this case? There are several interests to take into account.

On one hand, the person wants the robot to be able to do everything they can. It should be able to access the same network resources, and open the same doors. They also want the assurance of security that if their robot is captured and cracked open, the thief can't suddenly drive their car or enter their house. On the other hand, the corporation wants to know that they can be safe from these same kinds of things—if Sneaky Bob captures Travis' robot, Sneaky Bob can't use it to let him into the building and access the source code to Longhorn, which should finally be shipping by the time that we have superintelligent robots.

So, first of all, any security credentials need to be stored safely. If the machine loses power, it shouldn't be able to get at those credentials without some kind of strong password—essentially, the key or password that the owner has assigned to the robot. This prevents someone from stealing the robot, turning it off, and then ripping out the hard drive or memory core or whatever and searching for the password. Those master credentials (your robot key) would unlock the other credentials (your house key) stored on the robot.

Also, the corporation would need some way to restrict your ability to copy credentials to another device. I don't know exactly how this is done, but presumably something exists with smart cards to at least deter this. The corporation could say that only the ID card that they issue to me can be used to open the front door, and I can't copy this over to a robot, and that my machines can never access source code. Some of this depends on trusted machines—someone could always create a machine that mimics a human or doesn't follow by the rules or duplicates a key that clearly has "DO NOT DUPLICATE" printed on the side. This seems to be what it always comes down to... technology breaks down when you hit the physical side of things. Someone could make a little mod chip for your robot that allows you to bypass your corporation's rules as to what robots can or cannot do. Eventually, someone would still come up with a way to steal someone's robot and robot key, and they'd break into the building. I'm not sure if there can ever be a good solution to this kind of problem.

Oh well. Not really finished thoughts. Just stuff that was on my mind this morning.

No comments: