Re: Darwinian extropy
N.BOSTROM@lse.ac.uk
Sun, 05 Jan 80 02:10:55 GMT
Dan Clemmensen has suggested that SIs don't colonise cosmos
because of their high discount rate for future benefits. The
discount rate would presumably be high because of the
rapidity of their subjective time and the slowness of cosmic
travels. The idea is interesting since, if right, it would
help resolve the Fermi paradox.
One reason why SIs could have a high discount rate would be
if they had a bias towards the near future, just as we
humans have. We tend to care more about an imminent pleasure
than about a similar one we would sure to get in a billion
years. Perhaps all SIs would be "irrational" in the same
way?
Another reason why an SI might discount the benefits of
colonisation is because they would only come about if it
diverted some of its resources to the space mission,
resources which could have been used for other purposes.
Suppose that the objective of the SI is to maximise the
amount of valuable computations it will carry out during its
life time. At a certain time t it has a given capital
(consisting perhaps of its mass or available energy). Part
of this capital could be invested into a project that would
yield returns at t'>t, but meanwhile that capital could not
be used to make valuable computations, i.e. there would be
an opportunity cost which would have to be subtracted from
the expected returns when considering whether the investment
is worthwhile. The question is what the function
f=valuablecomputationspower(capital) looks like. For
Clemmensen's argument to go through, it would not suffice
that f had a jerk at some point, because different SIs would
presumably start out with somewhat different amounts of
capital. If all SI had originally a capital less that the
critical amount, then they would invest in space missions,
but the benefit of space missions seems to come in chunks
(one chunk for every planet or solar system one arrived at),
and so there would be some overshoot: most SIs would obtain
a capital greater than the critical value, and they would
have little to lose from using the excess for new space
missions. f would rather have to have the shape of an
inverted exponential, so that for each amount of capital
(greater than some start-up value), a slight loss of capital
would reduce the computation power greatly, whereas a slight
increase would bring but a negligible increase. Not only
would the function have to be of an exponential character,
the constants would have to be rather great, considering
that the gains would be obtained after perhaps a thousand
years and be enjoyed for perhaps billions of years, while
the required investment would presumably be very small
compared to the SI's total capital whereas the gains could
be very substantial. For instance, if any considerable
degree of parallelization of valuable computations were
possible, then df/d(capital) would certainly not decrease
rapidly enough.
If Robin's objection, that not all knowledge can be obtained
simply by sitting back and think (making computations), is
directed against Clemmensen's basic idea rather than against
some other more specific claim Clemmensen has made, then I
don't think it carries very much weight, because there is no
obvious reason why SIs should be interested in the detailed
structure of distant cosmic regions. We must not forget that
a mere suggestion for where the Great Filter could be, even
if it falls short of being a proof, would be very helpful;
in any case there seem to be much greater difficulties with
Clemmensen's proposal than that it assumes that the
curiosity of SIs is rather limited.
Nicholas Bostrom n.bostrom@lse.ac.uk