[cap-talk] Is "Authority" Subjective?
toby.murray at comlab.ox.ac.uk
Mon Jul 2 07:43:46 EDT 2007
> On 6/30/07, Mark Miller <erights at gmail.com> wrote:
> > I am among those who don't know CSP. However, your explanation seems
> > quite clear, thanks. But it does leave open a remaining crucial
> > ambiguity: Does the expression "b -> ...", combined with your
> > description "b is performed by another object Bob" mean x) "If the
> > system is in a state where only b would be accepted and if Bob does
> > attempt b, then b will happen and cause the described state
> > transition." or y) "If the system is in a state where only b would be
> > accepted, then Bob will perform b and cause the described state
> > transition."?
> > Interpretation #y is more consistent with my vague sense of what CSP
> > is about. But this interpretation presumes to describe the actual
> > behavior of all the players, rendering the players too deterministic
> > to be considered players of a game, rendering any description of
> > authority mostly meaningless. As David Wagner says later on, causality
> > is about counterfactuals. A full characterization of everyone's
> > behavior leaves open where to draw the counterfactuals.
I think that usually one applies interpretation #x.
The counterfactual is "b doesn't occur", not "Bob chooses not to perform
Suppose we have the system
P = a -> b -> STOP
we usually reason that a causes b since b can't occur without a
occurring. The counterfactual test is
1. if A occurs, then b is possible.
2. If A doesn't occur, then b is not possible.
The states in whcih 1 is true is that represented by the process P' = b
The states in which 2 is true is the initial state, P. Since b isn't
possible in this state, we reason that a causes b to become possible to
occur. (We can't reason that a causes b to occur, since CSP has no
notion of when events are forced to occur, but merely when they become
possible. Hence why interpretation #x is the usual one.)
If a is performed by Alice, then she might have a choice here whether to
perform it (in which case the system evolves to P') or not (in which
case we stay in the initial state). I don't see that it is actually
relevant though. The counterfactual merely reasons about states, not
One couldn't reasonably argue that a doesn't cause b to become possible.
Hence, I don't see how one can argue that Alice doesn't have authority
to cause b to become possible. b is part of her authority.
> > Interpretation #x in effect says only what the bounds are on what the
> > players might do in a given state, without prescribing which of these,
> > if any, they will do. This interpretation is more consistent with your
> > arguments below, and with my sense of what authority analysis should
> > be about. Given this interpretation, I think your argument #2 is more
> > correct but misphrased. Alice can cause Bob not to be able to perform
> > b, therefore Alice can cause Bob to be able to perform b. But only if
> > we know that Bob will always attempt to perform b when it would be
> > accepted (as in interpretation #y), can we say that Alice can cause
> > Bob to perform b.
Like I said above, we can only reason about when Alice can cause Bob to
be able to perform b.
> > It seems that your arguments as stated require interpretation #x
> > regarding Alice's actions but interpretation #y regarding Bob's. This
> > is consistent with the kind of authority analysis done by SCOLL, where
> > Alice would be described as a non-relied-upon subject and Bob would be
> > described as a relied upon subject.
Since we're only reasoning about Alice causing b to become possible
(rather than causing Bob to be forced to perform b) interpretation #x
should work for both Alice and Bob.
To model a non-relied upon subject Bob, we model Bob as the most
nondeterministic subject that plays by the rules of the game. As I
explained to David Wagner, because our causation predicates
are /refinement closed/, they continue to hold no matter how
nondeterminisim is resolved. Hence, if any possible behaviour of Bob
leads to causation, then this will be picked up by our causation
> > >
> > > Argument 1: Before Alice acts, Bob can perform b -- the system won't
> > > refuse it.
> > The above sentence suggests interpretation #x, where Bob might not
> > perform b even when he could.
> > > After Alice finishes acting, Bob can perform b -- the system
> > > won't refuse it.
> > But in you next argument, you imply that Alice might perform a1 and
> > not perform a2, which means Alice might have the authority to prevent
> > Bob from being able to perform b.
Isn't this then interpretation #x for both Alice and Bob? Or am I
misunderstanding you. I think we agree.
> > > With and without Alice acting, Bob can still perform b.
> > But if Bob might not actually perform b even when he could, then Alice
> > can't cause him to perform b, and so Alice doesn't have authority to
> > cause Bob to perform b.
But she can cause Bob to be able to perform b. This is the sort of
authority I'm trying to reason about (since it's all that CSP can
capture). I'd also argue it is generally the only sane way to do
authority analysis since we want an overestimate of authority. If Alice
can cause Bob to possibly perform b, then reasoning that Alice can cause
b is a good overestimation of Alice's authority.
More information about the cap-talk