XPost: alt.folklore.computers
From: antispam@fricas.org
In alt.folklore.computers rbowman wrote:
> On Tue, 30 Dec 2025 19:47:23 -0500, c186282 wrote:
>
>> They've THOUGHT they had 'AI' more than once. Minsky was sure it'd be
>> easy during the 60s. Then, as you said, the 80s came and went but no
>> 'AI' worth a damn.
>
> Much of the neural network operations are still the same today but the
> hardware didn't exist to do millions of tensor operations in any finite
> time for anything but a toy app. 'I need 50 billion to build a center that
> will suck up the power and water from three states' wasn't going to happen
> even if the hardware did exist.
>
> The neural network approach was taken over by 'expert systems' The
> problem was you more or less needed a human expert on hand.
But are 'expert systems' really AI? Theoretically so called
expert system shells could do smart things, but examples I saw
were essentially a bunch of "if ... then ..." which could be
written in almost any programming language. One example of
samewhat succesful 'expert system' is supposed to guide a user
trough installing Unix. Description suggests that is is not
smarter than modern Debian installer. And nobody thinks that
Debian installer is AI.
> Blend NN and
> expert and you have something that sucks in petabytes of data, sort of
> links together tokens (words) that commonly are used together, declares
> itself an expert, and spits out pieces it has acquired like a vomiting
> magpie.
Statistics people for ages construct decision trees. NN-s seem
to be much smarter, but once you have a decision tree (or more
fancy thing like forest with voting), you get quite ordinary
program.
--
Waldek Hebisch
--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)
|
