The problem with the views discussed in the previous sections is that they set
incorrect thresholds for permissible action; I think that the relevant constraint is
comparative: how does the action in question compare to other available actions?
In his recent discussion of the Canberra Suburbs and related cases, Johan E.
Gustafsson suggests a comparative constraint:
Avoid indeterminate worseness.If possible, choose an optionxsuch that it is
determinate that no option is better thanx.
23Recast in terms of permissibility: if it is determinate that for every actionψ,ϕing
realises at least as much utility asψ, then one ought toϕ. This suggestion is, I think,
correct: it requires that one not choose an action that is indeterminately what one has
most reason to do, when there is another option that isdeterminatelysupported by reason
available. The comparative aspect here is that an actionϕwhich is mandated by some but
not all sharpenings may be permissible only if there is no other actionxthat is mandated
byallsharpenings. This constraint captures what needs to be done to render the right
verdict about the Canberra Suburbs.
These comparative considerations suggests that rules like Capricious Sufficiency are
22Williams (2014), p. 32.
on the wrong track: what matters to the permissibility of an actionϕis not just that there
is some sharpening (or some portion of sharpenings) on whichϕmaximises utility; it also
matters howϕdoes on theothersharpenings. In the case of the Canberra Suburbs, it
seems to matter not only are there some sharpenings that mandate choosing Exe, but that
even on those sharpenings which mandate choosing Wye, Exe isno worsethan Wye.
As a first step, consider a view that I used to defend:
Sharpening-Maximize.
ϕing is permissible if for any other actionψ, there are
at least as many sharpenings that supportϕing, as there sharpenings are that
supportψing.
A quick gloss on Sharpening-Maximize is that one ought to take the optionϕ, such
that it is ‘most true’ that one has most reason toϕ. This is a substantial strengthening of
Capricious Sufficiency: many actions which are permissible according to that principle
are not permissible according to Sharpening-Maximize. For instance, since there are more
sharpenings that mandate choosing Exe than there are that mandate choosing Wye, it is
only permissible to choose Exe.
Sharpening-Maximize is obviously a close cousin of Randomize, but it renders a
different (and to my mind, correctly so) verdict in cases of indeterminate superiority. In
less outr´e cases, the difference between a Randomizer and a Sharpening-Maximizer will
manifest in long-run frequencies. If 60% of sharpenings supportϕing and 40% support
ψing, then in the long run the Randomizer willϕ60% of the time, whereas the
Sharpening-Maximizer willϕevery time.
But notice that it doesn’t rule out a lot of freedom: in cases of flat indeterminacy, there
may still be a lot of options permissible according to the Sharpening-Maximizer. In the
Mushy Boats case, where the indeterminacy is ‘flat’, the Randomizer will in the long run
save each boat half the time. But we can make no such prediction about the
Sharpening-Maximizer’s behaviour, even assuming full rationality: we know that every
time, she will choose to save one of the boats, but can say no more about the long-run
frequencies of her choices than we can about the Ass of Buridan. Thus, surprisingly,
though Sharpening-Maximize is a more restrictive rule than Randomize in many one-shot
cases, in terms of which actions are permissible, it allows for more ‘long-run’ freedom.
Each of the views we have seen is non-aggregative, in the sense that all that matters to
the permissibility ofϕing is the existence or measure of sharpenings on whichϕing is
mandated. Though not strictly entailed by these considerations, we have been implicitly
assuming that aggregative considerations do not matter. But the following case should
bring this into question:
Imbalanced Utilities.The situation is flatly indeterminate, and there are two
sharpenings. On one sharpening,ϕing realises 1 more util thanψing; on the
other,ϕing realises 5000 utils less thanψing.
In this case:
•
if weϕ, then it is indeterminate whether we maximise utility or attain just one util
short of the maximial utility level;
•
if weψ, then it is indeterminate whether we maximise utility or fall 500 utils short
of the maximal utility level.
To make the case vivid, we can use JRG Williams’s device of the wholly self-interested
individual with indeterminate survival-conditions. Suppose that it is indeterminate (true
to degree 0.5) that the person occupying your (present) physical body in one month is
you. I offer you the following options: you mayϕ(take five dollars now), orψ(take five
million in a month). What ought you do? It seems plain to me that you ought to take the
five million dollars: doing so is indeterminately either vastly better, or slightly worse,
than taking the five dollars today.
When the question is posed directly—‘assuming flat indeterminacy, should you take
an option that is indeterminately vastly better or slightly worse, or an option that is
indeterminately either vastly worse or slightly better?’—the answer seems obvious. Once
again, epistemicism and its associated expected utility maximisation can render the
correct verdict: if you simply attach credence of 0.5 to the proposition that it will be you
who receives the five million dollars, and credence close to unity that it will be you who
receives the five dollars, then rational choice is clear on how you should pursue your
self-interest.
Alternatively, suppose that you are ill. If treatment A is either slightly better or much
worse than treatment B, but each has a fifty percent chance of being the better treatment,
then we should apply treatment B.
24Intuitively, at least, it doesn’t matter whether ‘either’
here is given an epistemic or an indeterminist gloss.
Caprice and Sharpening-Maximize will say that eitherϕing orψing is permitted—it is
permitted to take either sum of money—because each is mandated on at least one
sharpening, and they are supported on an equal measure of sharpenings. For similar
reasons, according to Randomize, we must choose randomly betweenϕing andψing. But
these are, I think, plainly the wrong permissibility verdicts. Simply counting sharpenings
is a step in the right direction, but it is not enough.
In document
Elson_unc_0153D_14716.pdf
(Page 79-82)