RE: Coding from Scratch: Jaron Lanier

From: Ben Goertzel (ben@goertzel.org)
Date: Sat Jan 25 2003 - 20:06:24 MST


> Does the SuperCompiler substitute in a different algorithm if the one we
> Code is nonoptimal? I'd be surprised if it was that sophisticated.
> In order for it to do this it would have to know specifics about the
> algorithm and the data being used by the program which is currently not
> Information known by the program. For instance a binary search is much
> faster than a sequential search if the data in the list is already
> sorted.

Interestingly, in many cases there *are* known restrictions on the inputs
given to a given software function, in the context of the larger software
system in which the function is embedded. This is a case where the
supercompiler can be very useful. This is a technique called "program
specialization" -- given the specific context the function is being used in,
there may be a specialized version of the function that's much more
efficient. The supercompiler is really good at figuring out such
context-specialized versions...

Of course, this is different from inventing whole new general-purpose
algorithms: for that you really do need an AGI...

-- Ben G



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:41 MDT