| |
| |
| |
|
Page: 1 2
Comments:
<0> Here is the new AI found at this medical college:
<0> http://www.med.uottawa.ca/patho/repro/poster15.jpg
<1> what's that?
<2> ewwwwww, what the heck is that?
<2> i think i'm gonna puke
<1> haha
<3> AI_coder: please don't do that again, ok?
<0> athena: Sorry, it's a ovarian cyst.
<3> AI_coder: that's irrelevant
<4> lol
<5> DanF_DrC, eventually
<6> kats: I'd claim that it's impossible to understand text without interacting with an environment.
<5> i understood that without interacting with an environment
<6> (and ideally an environment similar to that discussed by the text, in some fashion. Although if you have common experiences with other texts which you can use to "translate" it, that would work as well.)
<6> hmm?
<5> i understood your text meaning, i think
<6> hmm. is English not your native language?
<5> it has been for 49+ years, why?
<6> Your phraseology is unusual.
<6> and it's hard for me to figure out some of what you're saying. >.>
<5> ok, to restart:
<5> [01:47] <6> kats: I'd claim that it's impossible to understand text without interacting with an environment.
<5> [01:48] <5> i understood that without interacting with an environment
<5> what of that is a nonsequiteur?
<6> The way I parse that sentence is "I understood that" modified by "without [me] interacting with an environment". Which ... is nonsensical.
<5> i understood that [which you just said] whithout interacting with an environment [which is what you claimed is impossible].
<6> Ah. But no. You /have/ interacted with an environment in the past.
<5> yes
<5> but for plain text, and understanding it, when it references nothing outside itself?
<6> I'm not saying you have to interact /while reading/, I'm saying that you have to have had interactions in order to "understand" - you can't compare concepts without having something to compare them to.
<6> But it always does. Even the word "text" references several concepts.
<5> in this case, it references itself
<6> And then there's issues like the really common metaphors people use all the time. (More is up. Importance is weight. The future is forward.)
<5> anyhow,, give me an example where an outside reference is required
<5> metaphors are text too,,,,, hmm
<5> it's 2am, maybe that's why i am not getting a grip on what you are talking about
<6> How would you understand the word "outside" without experiencing the world? ;)
<6> Or "getting a grip" ...
<5> "outside" is a difefrent place, like "room", but with other conditions
<6> People use a lot of non-literal language. Not to mention that there's very few purely self-referential texts - after all, there's little (if any) point to a text unless it refers to something.
<6> and what's a room, without having experienced it? Contemplate a dictionary - pick any word that references the world in some way, then follow every word in the definition that also does likewise.
<5> room is an enclosed space
<6> Keep going until you either give up or have a word defined - what's /space/? and /enclosure/? - with no reference to the world at all.
<5> when you think of something, do you follow all the possible understandings to the end , as you seem to be suggesting is necessary?
<6> No - but only because I have a mental construct of the concept on its own.
<6> But that construct was built from others.
<3> i.e. it is "grounded"
<5> room is place, outside is place
<6> I'm familiar enough with the constructs that pop up frequently that I don't have to break them down, but I often /can/ - and when I first learnt them, I had to.
<6> what is "place", to a computer program?
<5> a word
<5> with conditions for existance
<6> but the conditions themselves are meaningless.
<5> it must be that simple unless you have vision systems on the computer, and tactile abilities
<6> Okay, let me try another example: I ***ume you don't speak Swahili. If I gave you a Swahili dictionary, written for Swahili speakers... would it be of any use to you at all?
<6> just becuase you don't have the ability to do something doesn't mean it's not necessary. ;) There's a /reason/ NLP is very hard.
<5> well, <cough> "hard" is a relative term,, i wrote an exelent syntactic parser 15 years ago, just that no one was interested, and i could afford more puter power to switch to semantics
<5> couldn't
<5> i'm tired, (see 2am, mentioned before)
<6> syntax is moderately easy ... give or take the m***ive ambiguities. ("Time flies like an arrow" - something like five plausible parses) Semantics are where things get tricky. "Sit in the apple juice seat". does that have meaning?
<5> more than 5 :-)
<6> I might be thinking of the other cl***ical ambiguous sentance on the number of plausible parses. But you know what I mean. ;)
<5> yes, MIT got 5
<5> i am way out of steam on trying to work with others on Ai, i tried to cooperate for decades, and got nowhere, so i am not accustomed to friendy discussions about it
<5> i'd like to, however
<5> but it is 2am
<5> i'll stay up if you wish to talk
<6> Anyway. What athena said back there is a pretty good summary of what I'm trying to say - concepts need to be "grounded" to be meaningful. There's some words that have very little/simple meaning which are realllly easy to "understand" - mostly the little glue words like, well, "the", and "and", and so on. But for anything with concrete meaning, I think you need some sort of experience - not necessarilly with the real world, but with /some/ s
<6> ort of environment - to understand them.
<6> as for talking ... it's nice. but it's even later for me than it is for you. ;-)
<6> So I should sleep shortly as well.
<5> why must the Ai have a "hands on" experience with them tho?
<5> Cyc has no hands-on
<6> Cyc?
<5> yes
<5> Lenat's baby
<3> Doug Lenat's fact collection
<5> or i should say $baby,000,000,000
<6> Aaah. That.
<6> Does it understand, though? ;)
<5> i have never tested it
<6> If understanding is a mental state.. mm. I dunno. I view understanding as having a mental construct and the ability to compare experiences to it. ... which is sort of a circular definition vis a vis my claims. Hrm.
<5> hand coding facts, "common sense"
<6> But I'm not sure how else to define it.
<5> well, humans believe in dog(s), and have no scientific fact to back up anything they say about them
<5> err, god(s)
<3> but how's that related to understanding?
<6> right, but they defined them in terms of other concepts.
<3> oh, experince
<5> Sotek, simpler concepts, and even simpler, etc,,, tilly ou get as simple as i said before
<6> I mean, the cl***ical image of God is "A big bearded man in the sky who can do whatever he wants."
<5> Satan Claws?
<5> Sants Claus
<5> err,, nm
<6> Mm. hmm. Might be possible to define things /that well/ ... and just hardcode the "experiences" of the most fundamental things. Maybe. Still ... that seems like it could get a lot uglier than interacting with an environment. And I'd suggest that at some level you're going to wind up creating a simulated environment if only to explain the constellation of concepts along the lines of "space", "position", "location", etc. and of course time wo
<6> uld probably be directly experienced, too...
<3> and many of these concepts are relative to "self", which may need embodiment, etc.
<5> embodiment would be good, but is problematic
<5> if the Ai has a diffeent body that a typical human, most human physical concepts will be void too
<5> ask anyone in a wheelchair, or blind
<5> or deaf, or mute
<3> i think the more basic concepts are more important
<5> the simpler the better
<5> it simply must be so
<6> well, deaf/mute not as much as blind. sound is not as fundamental to the way most of us think as sight is. but ... yeah. and embodiment does have difficulties of the sort you pointed out, I suppose...
<6> ... but a lack of embodiment has other difficulties. And I think embodied would be simpler.
<5> i don't like the idea of an intelligence trapped in a puter box, unable to hear, see, feel
<3> can you give some examples of difficulties with embodiment related to different body?
<5> to me, hearing is a fantastically more difficult job than NLP, and i'd like to attempt it some day, but i'd never give a clue i did it, if i did
<6> hearing is scary. but it's probably easier than NLP. At least gauging from evolution.
<6> (there's a lot more things that can hear than can engage in NLP >.>)
<3> easier for evolution doesn't necessarily mean easier for us to duplicate
<5> hearing, afaict, is as easy as ***iging a cpu for every hair in the cochlea :-|
<6> ... I just got owned.
<6> athena: No, but it's the only real indicator we have, I think.
<6> I mean, all we know about NLP is that we haven't succeeded yet. ;)
<5> processing a 20khz signal, plus everything else in white/pink noise, to determine if anyone is talking int he background, and then ***iging a tag to everyone talking, and determining speech matched to stored templates,, and if one of them has a cold, etc etc
<5> understaning your partner on the other side of a table at a "new age" bar with the 110db "music"
<5> it's doable, but NLP may be far easier
<6> humans are really good at extracting certain kinds of patterns from certain kinds of noise.
<5> humans, yeas, Ai, no
<6> I suspect that if we can figure out how that works, we'd be a hell of a lot closer to having "true AI".
<3> yes, that's really the basic function needed
<5> you do realise the human cochlea has a trigger for each frequency we each can hear? and a feedback mechanism from the brain to each hair?
<5> to compare that to NLP, you'd need a motherbd for each word in the language
<6> kats: I wasn't aware of that, but I'm not surprised. Vision is ... pretty scary too.
<6> (actually I /know/ vision uses even more processing, but.)
<5> vision doesn't even have feedback to each rod/cone
<3> katsmeow: what does the feedback from brain to hair do?
<6> well, it's hard to envision how they could work "actively", while I can see how the hairs can get more information "actively".
<5> athena, throttle movement
<6> so to reasonably replicate hearing would probably require special hardware. Hrm. but I doubt the hardware would be that complex, just specialized. How many neurons are linked to a given hair?
<5> the feedback is approximately logarithmic too
<5> Sotek, i don't kow the count
<3> seems strange to feedback so far down... why not handle at a slightly higher level, as in vision?
<5> athena, the hairs are infinitely fragile, they break easily, to last a lifetime they must be bent as little as possible
<3> ah
<5> and a measurement of feedbackk times incoming signal is better than trying to ramp up the incoming signal
<6> I seem to remember log is easy somehow. anyway.
<5> the dynamic range of the neurons is limited
<6> moot point arguing which is harder. answer is: they're both hard. >.>
<5> i was gonna do a central set of oscilators to banks of LM566's, and banks of tuneable bootstrapped twin-T filters, etc etc,, the 100% hardware approach
<5> now the LM566 isn't even made
<6> heh. :P
<3> hmmm... Helen Keller only lost sight and hearing at 19 months age... i wonder if there are cases of children born without those senses and reaching same levels of "understanding" of concepts
<5> sure there are
<5> otherwise there'd be warehouses full of such people
Return to
#ai
or
Go to some related
logs:
ubuntu ifconfig setting static ip
#gimp
Eclipse +phpbrowser +firefox
#suse
kkkp linux
#php
gentoo missing libdri
#perl
mysql match against score stays 0
ubntus
|
|
