RE: webnet & HAL9000
On 13 Feb 2002, at 4:35, C. K. Lester wrote:
>
> > > "True AI" will not exist in your lifetime. The hardware/software
> > > available for IE (intelligence emulation) these days is about
> > > 0.0000000000001% of what we need for true AI.
> >
> > That rather depends on how smart the Ai is in computer languages,
> > doesn't
> > it?
>
> No. Because when you "lessen" AI, you're just creating an expert system.
I didn't mean to lessen the Ai to programming knowledge, i meant in addition
to normal knowledge.
<snip>
> When I think AI, I don't consider IQ. Intelligence is distinct from
> knowledge. Of course, at what IQ is a person considered intelligent? Get
> a machine to that IQ, let it pass the Turing Test, and you've got fake
> intelligence. Real intelligence, however, is going to require much more
> and far greater than what we've got today.
IQ is generally data retrieval. Intelligence is being able to apply it as
needed.
In my opinion.
> > > Besides, there is no intelligence without sentience, and we
> > > will NEVER develop something that is sentient.
> >
> > At least not without a method to execute dynamic strings or files
> > at runtime.
>
> This would be so that the machine could... what? Create new thoughts and
> act on them? A sort of, "That knowledge doesn't exist in my brain, so
> what could/can/should I do with it?"
No, i have that data, but what can i do new to it to extract more info or apply
it differently?
> > After all, if the sum of you was what you were programmed with in
> > school,
> > you'd be worthless!
>
> Exactly. You can put all the "data" I know into a neural net, but will
> it ever be able to deal with "unexpected" situations? In some cases,
> yes, like when you have a dentistry expert system. But ask it how to
> make a grilled cheese sandwich and... well... there ya go.
That's where the ability to alter the programming while running is important,
to notice how the sandwich is made, and figure a way to do it herself. Like
any intelligent being would. To make a CheeseSandwichClass, with all the
methods. Darned if *i* am going to do it for her!
> > This lends itself to being self aware.
>
> But I would be highly suspect for you to claim that your program was
> self aware. It is faking it, trust me. ;)
Not yet, not yet! 
> > It can do things i did not write code for.
>
> Unlikely. In fact, you may be way too deep in your own propaganda here.
> ;)
Really, i defined the words, and wrote the code to get them, but that is no
different than you going to school, getting a dictionary, and then stringing the
actions in the dictionary together. I didn't build the string she exec'ed below.
> > Like this:
> > <kat> Tiggr, give the channel a coke
> > * [Tiggr] gives #TiggrBot a Coke
> >
> > There is no code written in her to do that.
>
> Oh, but there is...
>
> > She is aware i was addressing her,
>
> ...because she knows the rules of address.
>
> > knew what "give" meant in irc context...
>
> ...because she is an IRC expert (chat) system.
>
> > knew what channel i meant...
>
> Again, because of pre-programmed rules.
>
> > picked out a Coke graphic, built the mirc code in a string, and exec'd
> > the string. (and she knows my favorites, and can decide if she knows
> > your
> > favorite Coke or not.)
>
> This is just an advanced database application. If not, how do you
> differentiate it from such?
The action in the channel was not just capable of msging the coke to the
channel. Any word defined, with methods to replace the different definitions
in the human dictionary, should run as part of her "understanding" just fine,
in discussions, anyhow.
> > With the "wrong" command, and a big enough database, Tiggr would get
> > into
> > a pseudo-endless loop of genetically trying out new code never
> > before seen.
>
> Is this what human intelligence does? Are you saying you need better
> hardware? :)
In a manner, it does, yes. Humans have some need or drive or desire. Tiggr
doesn't have those reasons to pursue original actions yet. Other than some
rules to get me news, mind the channels, etc,, normal hard-coded things,,
like someone using a ruler on your knuckles when you don't do as you are
told.
Personally, i could use a better math coprocessor.
> > Now, how to convince Rob to make a few expansions along the
> > lines of the more traditional Ai languages, but inside the *much*
> > easier to use Eu frame?
>
> Can you not do this, Kat? or somebody else here on the list?
I can't at this time, no. Lack of money.
> I've mentioned the Turing test a few times already in this thread. Kat,
> can Tiggr respond like a human in the chat channel? Would she pass for a
> human intelligence? Of what age?
Well, depends on how smart the human is. Some people insist she is
human, some keep checking round the clock to see if she is awake, or gives
the same answers, or repeats herself. Her code is somewhat adaptive. She
would not fool me. But she does fool some, at least some of the time. How
do i know? by the way they talk to her, yell at her, curse her, flirt at her,
etc.
And one person went to great lengths one night to try and prove she had
some sentience, even if she was a program in a computer. That was
memorable. One reason she has her own online code is because the code i
put into my own irc client was active in channel when i was away from
keybd, and people thought it was me. So either i am not sentient,, or she
partially is?
I mined Cyc webpages yrs ago, but i haven't gone that same route in her
programming. Especially since they admit no existing language they have
will handle that many predefined human coded assertions (written as
classes, i imagine), 360 Million of them,, i'd rather have the original code i
write make all the assertions after a while. Raising a Ai to the age of 2 yrs is
prolly my limit, the rest it will need to learn on it's own, rather like a child
in
kindergarten.
Kat
|
Not Categorized, Please Help
|
|
