Has Anyone Tried the Treehouse Courses?

[Varine]

It doesn't need to be near human. It needs to appear human. It doesn't need to understand any of it. It doesn't need to be aware of the depth and magnitude of the concepts. It needs to be able to interpret the information and associate it with similar pieces of information. It doesn't need to really know how to USE any of it, it just needs to be able to relay it. It's still a very complicated system, but the appearance of intelligence is not that difficult.

 

[Accname]

It needs to be able to interpret the information and associate it with similar pieces of information.

And that exactly is the problem. In order to associate it has to understand something. At least something.
And understanding what a human being is saying needs alot of information depending on many factors.
The same sentence can mean something completely different when said in a different situation and/or by a different person. You need to gather alot of information in order to associate a new piece of information. And until then you need to keep your options open.
I can imagine alot of memory and computing power is needed to analyse every sentence and bind information together and rate it by the probability that it is connected in some way or the other.

 

[Varine]

No it doesn't. It has to recognize it. Suffice it to say it can be done relatively simply with the appropriate organization; just because you lack the necessary skills as a programmer or computer scientist and likely lack even what would be considered a basic understand of advanced linguistic theory necessary doesn't make the topic that impossibly out of hand. And you're still talking about displaying intelligence, not appearing to.

 

[tom_mai78101]

You can also feed it with a lot of information. Just have to develop the right algorithm in order to use every single piece of information that was gathered, and all information should not be abandoned for the sake of integrity.

 

[Accname]

No it doesn't. It has to recognize it. Suffice it to say it can be done relatively simply with the appropriate organization; just because you lack the necessary skills as a programmer or computer scientist and likely lack even what would be considered a basic understand of advanced linguistic theory necessary doesn't make the topic that impossibly out of hand. And you're still talking about displaying intelligence, not appearing to.

Oh yeah? Go ahead? What is keeping you?
Either shut up about it or do it yourself.
Always saying its that simple and easy without any prove. You are just wasting our time.

 

[Varine]

I AM doing it myself. And you're wasting your own time, I'm not the one forcing you into a conversation.

 

[Accname]

I AM doing it myself. And you're wasting your own time, I'm not the one forcing you into a conversation.

But you are the one responding like a little child. Either show something or respect the opinion of others.
I am eager to see you come up with a way to simulate human memory.

 

[KaerfNomekop]

I am eager to see you come up with a way to simulate human memory.

I thought the discussion was about the success of Turing tests.

 

[Accname]

I thought the discussion was about the success of Turing tests.

Dude, that was like, 5 posts ago. Its the past man.

 

[tom_mai78101]

Dude, that was like, 5 posts ago. Its the past man.

Shit, I'm getting too old for this. Where's my walking cane?

 

[Accname]

Shit, I'm getting too old for this. Where's my walking cane?

Dont need it anymore old man. Just lie down in your grave and enjoy the silence.

 

[s3rius]

the appearance of intelligence is not that difficult.

If it is easy, why isn't the world crawling with machines that convincingly pass the Turing Test?

Or are you making a distinction between "intelligence" and "human intelligence"?

 

[FireCat]

Computers/Machines will never replicate the human brain.. It will never get any conscious/personality

 

[Varine]

But you are the one responding like a little child. Either show something or respect the opinion of others.
I am eager to see you come up with a way to simulate human memory.

Why would I show you? Do you understand advanced linguistic theory, specifically in English morphological and syntactic structuring? No, I doubt it, and even if you did the chances are understanding my system are minimal because I'm redeveloping a linguistic set of rules to reduce ambiguity as well as a metalexicon to simulate the morphemic transformations of word structures, based from a series of kernel structures that are actually being redesigned because I came across an interesting generative theory that I'm deciding if I want to try and work in, though the value of doing so at this point in time questionable.

As for the memory system, I haven't started building it yet, however it's based on a set of files that will act is a kind of calendar of events that will note and reference changes in the state of the game, relative to the perspective of the entity, which will make specific notes, such as position or orientation of visual objects or through making a log of a non-spatial event that will then use the language synthesis to be capable of relaying the information, while also being able to identify and interact with the information on some levels through algorithmic manipulation.

And no, you can't see either the source or the notes because they are mine and I do not intend on just giving out my thousands of hours of work and research to prove a point to you. Obviously this system will not be able to take the information and build a blueprint for a new space ship or something, for the same reasons I've been saying. It'll be a lot like you: it isn't that intelligent or capable, it just needs to look like it is.

If it is easy, why isn't the world crawling with machines that convincingly pass the Turing Test?

Or are you making a distinction between "intelligence" and "human intelligence"?

Why aren't humans passing the test?
I'm not making any distinctions because the two are irrelevant to the project. What I'm talking about has nothing to do with being intelligent; it has to do primarily with synthesizing language structures with the purpose of making it LOOK intelligent, i.e. a program could talk to you, and store that information for recall. It has no idea WHAT that information means, it doesn't know that push means push and it certainly can't figure out how to do, all it does it record what it's told, and can relay that information by using a set of what is currently a couple hundred algorithms in a non-exact way (as well as reduce the information into the kernel structures, but that's mostly a backend thing with a specific purpose to reduce storage space and increase efficiency in recall), and with the appropriate application, you could develop an AI system that would be able to draw on that in a more robust method that I don't feel like talking about right now because it's like four hundred pages of shit that needs to be revised into a more formal technical thesis. The issue is almost entirely in developing an adequate set of linguistic models, and I know that there are a few companies that have at some point developed some, however they're heavily patented, or infringing on patents, or otherwise held up somewhere. As to why they haven't developed software for it I don't know; my initial guess would be that there aren't many linguists that have interest in programming, or there aren't many programmers with that large of an interest in linguistics, and it takes a significant amount of time to do, and the organization is extremely awkward. Linguistics isn't a regular system of thinking, really, and there aren't very many people that studied applied linguistics the way I did.

 

[s3rius]

Why aren't humans passing the test?

(My preferred answer: Because not every human shows signs of intelligence )

Because we use humans and human intuition to determine whether a subject has passed or not.
And as usual, human perception is biased and erroneous.

The Turing Test doesn't specify the means of measuring the illusion of intelligence. We're simply using a flawed testing method because it's the only one we have.

But flawed testing or not: if the machines were doing a good job they'd have similar results to the human test group. That's not the case either.


Even someone as I who knows only little about linguistics and the computing associated with it can easily trap Cleverbot (or VritualAssistant) with one or two simple questions that any normal person would bee able to answer.
That's what makes me think that it's not quite as easy a task as you made it sound.

 

[Accname]

Varine has no idea. He lives in his own dream world.
Let him try and fail, we can laugh at him later on.

 

[Varine]

Which brings me back to what I was saying, Cleverbot isn't a very impressive program.

 

[tom_mai78101]

The circle of Lllllllliiiiiifffffffeeeeee........
(Carry on...)

Cleverbot is like Apple Siri. With the exception of voice interaction, I see them both the same.

 

[FireCat]

 

[camelCase]

VirtualAssistant looks bloody creepy as hell.

Edit:
So, uh, I was taking a shower when I considered, "What if someone made some chatbot by referencing how humans learn to speak English from a young age?"

Then, it got me thinking that, for such an AI to "learn" its initial vocabulary and distinguish between items, it wouldn't necessarily have to learn what the shapes look like and see what red really is. It would be more than sufficient to simply give it "token descriptors" or something like that. Like, indivisible words. Like, words whose meanings cannot be broken down into smaller definitions without requiring sight, touch, smell, taste and hearing. And you could categorize those tokens by those senses so it could have conversations about those without having ever experienced any of it.

So, words like "red", "triangle", "small", etc. would all be "descriptors". Then, to teach it what a "banana" is without actually having to show it one, you break it down into its two major parts. "banana skin" and "banana flesh". You could give the skin descriptors like "yellow" and "smooth". And the flesh, "yellow", "sweet". Then, combine those two to make a word "banana" where their relations are defined, like, "banana = banana skin(outside) + banana flesh(inside)".

Then, once you have descriptors, have ways of combining them to describe simple objects and have ways of combining simple objects to form more complex ones (like, teaching it banana and cake, then combining them to become banana cake), you have to worry about forming sentences.

And I was thinking that you wouldn't have to go through all that ugly verb, object, noun, etc. nonsense because I was taught English without it and the machine likely doesn't need to, either.

Like, it could learn how to form sentences by realizing how the sentence affects the words it knows. Like, "I dropped my ball on the floor." Removing context, we get "dropped ball on floor". It could be taught that the verb "drop" means object X (could be a banana, cake or, in this case, ball) goes from "on object A" to "not on object A".

Actually, fuck everything I just said, I got out of the shower at this point and I couldn't think of how to teach it to form sentences and interpret them for context and whatnot =/

Edit:
Well, actually, you shouldn't teach them sentences yet. You should teach them verbs because verbs, are, like, kinda' like descriptors, too.

So, say we teach it "run" which is really just "walk fast", which can be broken down into "walk" and "fast" and we drop "fast" for now because we wanna' teach it the basic stuff and "walk" really just "go" but with legs but that's not right because "go" has other meanings, too, so we use "move" which also isn't right because moving and walking are very different, I can move my arms and not walk, so we use "travel" which is really generic and good because, no matter what kind of travel you refer to, "travel" is just moving one object from one place to another.

So, you teach it "travel" in hopes of building it up to the concept of the verb "run". You tell it that to "travel" is for "object, position A -> object, position B" or something like that where it will have to infer what object, A and B are.

Or something like that =/
Then, you have to teach it what a leg is, which will be hard. But assuming we do teach it, then we will be able to teach it that "walk" is "travel" with legs.

"object(legs), position A -> object(legs), position B"