[cap-talk] Communicating conspirators (Re: Second ABAC Google talk is now up)

[Mark S. Miller]

John Carlson wrote:
> It seems to me that communicating conspirators is kind of like this:
> "If I told you, I would have to kill you."  I can see the case where
> you might tell one of the conspirators and lock them up in Pelican
> Bay until they die, but then why tell them at all?  They would
> be of no use to you locked up.

 > I guess I am thinking of a case where I have insured that the object
 > can only talk to me, that the implementation of the object is
 > transparent,
 > and I can see no way for the object to affect the state of an object
 > besides me.


Within a computational system, such a locked up conspirator may indeed do you 
a lot of good. But if it's locked up, it's not communicating; it's confined. I 
believe you've just described the confinement problem, whose overt subset is 
quite solvable.


> Perhaps this is a stateless object with no information
> to transmit?

E's approach to the confinement problem, essentially a simplification of 
KeyKOS's, is exactly along the lines you suggest: A transitively immutable 
object (in E, a DeepFrozen object) cannot be a communications channel.


> Which makes the object useless?  Would a mathematical  
> function
> which only had parameters and no side effects qualify?  I am talking
> about a function, not something implemented in the computer.  This
> is pure thought, not something practical.

Actually, DeepFrozen is a perfectly practical way to solve confinement. 
KeyKOS's solution is more complex but more flexible. If the extra flexibility 
justifies the extra complexity (which I think is plausible), then it is also 
more practical.


> Maybe security will change
> when we get quantum computers.   We need to be prepared.   Is there
> such a thing as private state that can only be shared with one particle/
> wave in physics? Entangled particles?

Interesting questions, but quite beyond me.

-- 
Text by me above is hereby placed in the public domain

     Cheers,
     --MarkM