From: Dan Clemmensen (email@example.com)
Date: Tue Aug 07 2001 - 20:08:50 MDT
Ben Goertzel wrote:
>>The more one
>>concentrates on building a real mind, and doing things right the first
>>time, the more complexity it takes to do simple things. It is better to
>>implement functionality with a thought than a fragment of code, but a
>>thought is vastly more work and requires a vastly larger base of
> yeah... you're preaching to the converted here!
> This is what I've always tried to explain to people who ask why, if Webmind
> is going to be so smart when it's all working, it can't beat Deep Blue at
> chess yet, or [insert your favorite achievement of a specialized piece of
While this is true at the philosophical level and at the architectural
level, the distinction may become moot at the implementation level.
This is not true now, of course, but may be true shortly after hard
takeoff. After having a thought (i.e., constructing a thought object
structure of some sort) the AI may elect to optimize that thought into
code, microcode, or hardware, depending on the projected or measured
likelihood of future use of the thought.
There are numerous examples of this, but note that the situation has
changed. In the past, the system included a programmer and a program,
not a self-programming programmer. In the examples I'm aware of, it
was easier for the programmer to implement the functionality at the
highest level of abstraction, but the result ran slowly. Examples:
early APL interpreters implemented the "domino" (matrix divide) in
APL, not in assembler. UNIVAC's cheap 360 clone implemented certain
obscure opcodes with registers in core memory rather than silicon.
FORTH interpreters implemented some FORTH primitives in FORTH.
In the case of the AI the program is the programmer. But we are saying
that we human programmers will be tempted to implement functionality
at the lower level rather than as thoughts. This is historically
unlikely. After we get used to thought objects, we will begin
programming in "thought." GISAI and especially CFAI specifically
call for the programmer to contribute only those thoughts that the
system will eventually be able to have derived for itself, assuming
that I read them correctly.
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT