Re: Opening chapters of Damien's new novel now online

From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Wed Dec 26 2001 - 11:08:39 MST


Spike Jones wrote:
>
> All wars should be pun wars. This could even continue
> even after all of humanity uploads (the horror). Then the only
> casualties would be those who intentionally erased their own
> ROMs, seeking in data-death the sweet myrrh-say of oblivion.

Spike, you know not the horror of what you say. As humans, our lexical
databases are not well optimized for the production of really horrible
puns, since it takes an extra effort to think of a similar sound for any
given word. A natural-language AI with an auditory database could
routinely search for all possible puns, selecting only the *really awful*
ones for production in a conversation. Furthermore, the rate of
production of really awful puns would probably be at least one per
paragraph. Imagine taking the time to go over every paragraph of an
otherwise ordinary conversation and spending a few minutes per sentence
trying to think of a really awful pun. Given an integrated punseeker
subsystem, this could be a trivial *and routine* effort for an AI. Thus
an AI could produce *horrible* puns, all the time, during every
conversation, with no letup. Furthermore, Moore's Law would cause the
puns to become continually worse, preventing the users from escaping even
into merciful numbness - given Moore's Law, this holds true even if an
exponentially greater amount of processing power is required for a linear
increase in pun horror.

An obvious corollary is that Michael Callahan is an AI.

Leaving that aside, it's clear that pun wars in the future will be much
grimmer than the relatively innocent affairs of today. Indeed, it may be
that humanity's only chance for survival will lie in avoiding pun war
entirely, as the puns of the future will be too horrible to be used.

-- -- -- -- --
Eliezer S. Yudkowsky http://singinst.org/
Research Fellow, Singularity Institute for Artificial Intelligence



This archive was generated by hypermail 2b30 : Sat May 11 2002 - 17:44:31 MDT