Printable View
Asked my mate this question in reference to ALICE:
Q:Why is it AI if it is just a load of If and Thens?
A: What is intelligence? Why are you sure your brain doesn't just do a whole lot of if's and then's?
There's a thought!
Anyone answer that for me?
i suppose your brain does do alot of If's and thens. But you cant really compare the way your brain does it to the way a processor does. At least not a NORMAL processor. :)
There are a couple of languages for AI, but cannot remember what they are called. At Uni we used one to program an evaluation thing for army snipers.
Basically it revolves around having a list of attributes for a thing.
eg
Fruit(Apple,Orange,Pear,Grape)
Colour(Red,Yellow,Green,Black)
You then program the computer to answer questions,
e.g Is a mandarin a Fruit.
The program reflects this against the list, but never answers in total negative,
e.g "I don't think it is a fruit, should it be a fruit"
If the operator answers yes, the computer will modify it's opwn programming structure
Fruit(Apple,Orange,Pear,Grape,Mandarin)
Hence if the question is asked again it can answer yes. This is why ppl think the machine is learning.
AI is still at start up stage, but it's getting better.
Whether or not a computer will have true intelligence is a moot point really.
Umm check this jethro www.a-i.com ask BOTSON for Jason hutchens!
But isnt your example just like: if its a fruit then yes!
Your brain would do the same: If orange is a fruit then yes!
Beacause you know it's a fruit. If the program knows it's a fruit than it will tell you!
The only difference bieng that we no defintely a car is not a fruit a program will ask you if it should be!
But that is the same with us We could answer with no a orange is not a fruit and then be proved wrong by someone else until we come to realise that it is a fruit!
The program just skips this process!
Am i wrong?
Basically no,
Was giving a simple example. Generally you have a whole slew of these table things from which to formulate an answer. So you could ask the computer "Is a Mandarin a Fruit"
and it starts asking "What colour is a Mandarin" etc etc etc.
Basically the code constaintly re-invents itself, seemingly to get smarter. And this folks ain't a sign of intelligence.
I agree with it not bieng intelligent!:)
Beacuase intelligence requires more than just bieng able to learn. It needs the ability to manipulate ones environment, problem solving skills etc.
But still the main question/statement bieng "Human Brains just does a whole lot of if and thens!"
??????????????????????
Yes had an email from my Uni project leader. The program is still going, and growing. Apparently they are getting some weird results, that look like the program is thinking for itself, but it's not, the code has topped 1 gig:eek:
That site i mentioned first is actually making a AI that learns!
No *****
That's the idea behind AI. However l still believe it's more the program is constantly regenerating itself and adding to or improving the logic threads. Mind you, isn't that the definition of developing intelligence:confused:Quote:
Originally posted by Beacon
That site i mentioned first is actually making a AI that learns!
No *****
Cannot find the site at the moment, but it's about these Japanese sci dudes who claim that the next advance in hardware will come through using a bio soup. They are claiming that current technology is being hampered by the speed with which chips can process electrons. Weird, and wild.
Go to that site at the top it'll explain everything for ya.
I dont quite understand it fully. But it was in New Scientist and crap!
www.a-i.com
It's go to learn somehow. By re-correcting itself than it will i would imagine!
I've never really heard a satisfactory definition of intellegence let alone artificial intelligence.
For one thing, I don't think intelligence has anything to do with logic. I believe that logic arises from intelligence and not the other way round.
So, if intelligence is not based on logic, what is it based on?
You can't give a supposed AI a small databse of fruit & then give it an opportunity to make its database bigger - this isn't learning.
Remember when you first saw a bus, or a computer, or when you first discovered that a tomato is a fruit, not a vegetable. You (eventually) adapted to it. I believe there are people in this forum who have never seen snow - but know it exists. "AI" can't do this. Learning is not simply increasing the data sources at ones disposal, not just increasing vocabulary. Learning even involves learning how to learn differently.
Humans tend to classify stuff. A computer can't do this unless certain aspects are rigorously fixed. (colour for example).
(oh no, I'm really rambling now...)
Take the colour 244,18,6. What colour is it? Probably red? But a computer would say "no, 255,0,0 is red." It may be programmed to say that a range of colours is red, but it would never learn this for itself.
(Dammnit! shut up Behemoth!)
Ok. Sorry. Bye.
I think you underestimate the potential of AI. There have been signifaicant advances in artificial intelligence, knowledge based systems and so forth.
Behemoth - what you are talking about is classification and prototypes. Classification needn't be as rigid as you suggest. To address your colour example, the colour isn't actually pure red, athough if you, as a human with good eyesight and background knowledge, were asked to say what colour it was you would (almost certainly) say it was red. The computer can in this case easily recognise how close to pure red it is and use some sort of fuzzy logic to express the 'red-ness' of it.
I take issue with the assertion that adding to a database isn't learning. Well actually that's true, but adding to a knowledge base (which is a layer of abstraction higher)? I would consider that to be learning. Knowledge representation is actually not tremendously complex. Problem solving can be done to some extent, with the correct knowledge base.
The difference between a knowledge base and a database is this: a database contains data and facts, but a knowledge base also contains rules and meta rules. This means that an AI can not only know a fact, but it can infer new facts from it using the rules it knows. Just how different is that to any cognate being? When you add sensory arrays which allow the AI to perceive its environment (thus adding new data) and use robotics to allow the AI to interact with that environment directly and adjust its senses, that is really an artificial intelligence, at a very simple level. Current AIs are at approximately the same level as insects.
Did you see the Christmas lectures in December? I am guessing they only show them in Britain. To anyone who doesn't know what I'm on about, the Royal Society (a very old scientific institution) puts on a series of lectures no scietific subjects over the course of a week or so which are televised. This year it was robotics. Well anyway, they showed some small robots which could explore their surroundings. They started off with just a neural net (I think they had something like 47 neurons each), some senses and a motor and wheels to travel around with. The first robot spent a couple of minutes learning how to move around, then started to wander around an area with obstacles, mapping the region out. A second identical robot was then placed in the area, and the first robot transmitted its knowledge to the second by radio. It was really quite cool :) The second robot then knew everything the first did.
Anyway what I am saying is that learning in AI is already happening and has been for some time as far as I am aware. Admittedly it's not that great yet, but it's developing rapidly and the prospects are good.
We used to use a language called Lisp at college for AI. It was much like the earlier example given by Jethro, but I have to agree with Harry that this is just a knowledge base, not true AI.
FOr example my daughter knows what a cat is. She see a picture of a Tiger and says "Meow" which is what she says for a cat. She hasn't yet learned that Tigers don't Meow. She will eventually and she'll store that piece of knowledge away. Next time she sees a picture of a Tiger she'll know that it Roars. She's just building a database of knowledge.
One of the tings that makes her intelligent (as far as a child can be) is that she can solve problems that are similair to previous experiences.
e.g. As she becomes older she can read two newspaper reports, one saying that Tony Blair is the best Prime-Minister ever, and another saying that he's the worst. She can weight the two articles up and decide how much of either to believe based on previous experience with the papers or how the article is written. She can then supplement this information with more data from other sources and finally come up with a coherent argument as to what she thinks of Tony Blair.
One of the advantages humans have is that we are saturated with information daily by all five senses and still it takes us about 7-9 years to be relatively smart. A poor computer only has human input to learn from.
Maybe this will change, and it should be feasible to build a system that starts reading all the news articles on the web and therefore forms an opinion about various world leaders. But this opinion will be based on the writers perception. The computer will still lack the perspective of a human.
For example: A newspaper writes "Our Glorious leader today ordered the extermination of all foreigners in the country."
To a computer this is a very positive sounding account of the leader.
A human would see through the propaganda and be revulsed by the article.
Just my thoughts,
SD
To a computer which didn't already have an extensive knowledge base involving a system of morals and a knowledge of how to recognise biased reporting, yes that would pose a problem, but the source of that problem is, as you say SD, that the computer hasn't had the wealth of life experience that most humans have. They are very domain-specific currently because it is impractical to provide an AI with knowledge, in detail, of a range of subjects.
On the subject of reading, natural language processing is just one of many challenging sub-fields of AI. Every natural language in the world shares the same universal grammar, with a few additions specific to the individual grammar. I've had a few lectures on NLP, it's amazing (to me at least) what can be done already, in theory at least.
I noticed my previous post has thus far been neglected. Either because people believe it is not worthy of a response or that they believe that a definition of intelligence is self evident.
However intelligent a computer seems to become, how intelligent is it really? If you built a computer that was as good at humans at pattern matching, learning and adapting, is it intelligent? Is it still just processing rules or does it really "know" what it is doing? How can we ever really know if a computer program really "knows" what it is doing?
If a computer can never really "know" what it is doing, are there always going to be limits that prevent it from doing everything a human mind can do?
We may be able to build useful tools/toys that use AI technology but until we address the problems of defining intelligence and what gives rise to consiousness, we may never build true AI.
Sorry Simon, didn't mean to ignore you.
In answer I don't know of a definition of intelligence. I know some-ones going to jump in with the Turing test, but that is a very specific test.
I think the real question is where do we set the benchmark.
Is an ant inteligent because it can perform certain tasks. A bird, A Dog, A Dolphin, An Ape, A Human, or must it be able to outsmart us before we will admit that it's inteligent?
I know I'm posing more questions and not answering any, but that's because I don't have the answers. Does anyone out there?
SD
The question is not answerable because the goal posts keep moving. Many definitions of AI have come, been achieved, and gone. The fact is noone is really in agreement as to what consitutes any kind of intelligence. In order to answer the question you need to understand the brain and nervous system properly, otherwise what can you compare the computer to? We can merely emulate, not model. The field of Cognitive psychology provides some insight into what may be the way the brain is working, but the fact is we don't understand ourselves enough to apply some definition to a computer. What if we defined intelligence and then discovered we weren't intelligent ourselves? That would be a little silly.
I'd just like to add something to my statements about the extensive knowledge base and system of morals problem - there is another question of what are the correct morals?
I just felt the need to add that.
I know those are difficult questions to answer but I don't believe we need a complete understanding of the human brain to be able to formulate an answer.
Many definitions of (artificial) intelligence have failed merely because they were wrong (or incomplete). That does not mean that the answer is unattainable.
I certainly would strongly disagree with the "Turing test" being any serious measure of "AI". It may be an interesting way to appraise AI programs but certainly isn't a good measure.
Many scientists are still trying to understand exactly what it is that gives rise to consiosness in the human brain and that is still undecided. Some people say that consiousness is not a result of mechanistic processes (and therefore not a result of brain functions) and if this is the case, it would probably be the nail in the coffin for "true" AI.
I would just say that if we developed a definition of intelligence that excluded humans, we would have missed the point. I think we can be sure of our intelligence as we "know" we "know". Being self consious proves (to ourselves at least) that we have some degree of intelligence.
Besides this, what about the mathematical objections to AI? Has anybody ever come up with a serious way of overcomming Godel's objections?
I want to know why I spontaneous gibber to myself...Is this fuctionality available in AI.
Actually, what started as a silly response strikes a thought. Is AI spontaneous, and would AI talk to itself. We can teach ourselves things without outside input by clarifying or redefining things we already knew. ..
Bah I've done it again. I'll get my coat...
I would ask, is the human mind spontaneous?Quote:
Is AI spontaneous...
For free will to exist it must be but there is evidence to suggest that every descision we make (or think we make) has already been made by our subconsious up to 9 seconds before we "realise" it consiously. Indeed, the mechanistic view of the universe is that every event is a direct consequance of it's preceeding events and that nothing happens spontaneously (i.e. without cause).
We may think we're spontaneous but that is because we are often suprised by what suddenly "pops" into our heads. :)
Behemoth, that's like the asserting new facts from existing facts thing I mentioned.
Simon, I don't know about Godel's objections... care to clarify?
I agree with you that the answer is not necessarily unattainable, in fact I think it will be attainable eventually, I just don't think we can say for sure right now. Yes, if we defined intelligence such that we weren't intelligent we would have missed the point, and that was exactly the contradiction I was pointing out. If we define intelligence now without understanding ourselves what makes us like we are, then what if we discovered more about ourselves and realised that actually, by the definition of intelligence we have decided on without all the facts, we have defined ourselves to be unintelligent? This is exactly my point.
You really cannot equate conciousness with intelligence. They are not the same thing, although they are linked. A lack of conciousness does not, in my opinion, preclude intelligence. Is a woodlouse concious? A virus? What about an ant? What about a colony of ants?
We can formulate a tentative definition of intelligence, subject to change as we learn more about ourselves, but I think that's really just about it. I don't think we necessarily need to define intelligence so clearly in order to work towards it. Whether we achieve it or not is then simply a matter of definition.
I don't think the argument about free will can go anywhere. Believe me, I've tried. There have been many arguments about it in this very forum, many of which got quite heated. It all centres around the underlying randomness (or not) of quantum processes and what you consider to be free will. Personally I don't see that it matters either way whether we have free will or not, since we're never going to be able to predict anything through it, and it's just inconsequential other than some people are uncomfortable with the idea of their lives being predefined.
I was not trying to spark a debate about 'free will' even though I believe I missed the previous discussions in this forum. If anyone wants to debate the notion of free-will with me then I'd be more than happy but I raised it purely to question the need for 'spontinaety' as part of the definition of intelligence.
I'm unsure as to whether Godel himself ever objected to the possibility of "true" AI but his incompleteness theorems have been interpreted by many as a valid, mathematical objection to the possibility of computers ever being able to achieve "human" intelligence.
The gist of the incompleteness theorem's are that any formal system can construct a proof (based on the assumptions of that system) that contradict the assumptions of that system.
Basically, this statement is true but the system cannot prove it without contradicting itself. Human beings make the intuitive leap to grasp the truth of the statement that computerised systems cannot. Computers cannot "step outside" of themselves to "know" what they cannot prove.
I think an example of an unproveable but "knowable" concept is the understanding of numerosity.
I suppose it comes down to are we really being spontaneous when we do something or are we trying to achieve something. (An adrenaline rush etc. that makes us feel better). What I'm getting out, is, if I spontaneously buy a computer game I'm doing it to satisfy a need (or suspected need), the fact that I haven't really thought it through is the only difference between a spontaneous reaction and a premeditated one.
I'm unlikely to spontaneously stand up in the office and shout "Flurpl Lipp Lipp ping pong doey". That might be spontaneous, but it would not benefit me in any way. In fact, if someone did it you might think they were mad (a bug in the program).
So if a computer doesn't have needs (happiness etc.) then it has no desire to do things spontaneously. Sure, we could program it to randomly do things, but that would just be erratic/mad behaviour. So maybe as long as a computer doesn't have any desires it will appear to be quite happy to just sit there waiting for the next external stimulous to react to.
If we programmed it to say, make sure that it's batteries don't run down (assuming it's a Californian computer and doesn't have a steady supply of electricity ;)), it might start doing things "spontaneously" if left alone. For example, it knows it's batteries are running down within the next 30 minutes, it also knows that if it calls out for help (screen, speech synthisis) someone might come and change the batteries. So it will start to cry out for help (much like a baby). But this is still not spontaneous, it's trying to achieve a goal.
SD
What the heck is numerosity?
I didn't mean to say you were trying to spark some conflict, I just meant 'argument' as 'assertion' or 'discussion', you know what I mean? Maybe it was the wrong word.
I don't see what you mean when you say 'true' AI, it doesn't really convey any meaning to me. Are you perhaps meaning intelligence of the same calibre as human intelligence?
I wasn't aware that I gave the impression of being offended. I wasn't.Quote:
I didn't mean to say you were trying to spark some conflict, I just meant 'argument' as 'assertion' or 'discussion', you know what I mean? Maybe it was the wrong word.
Numerosity (may be a spelling mistake) is an understanding of the concept of numbers. If you realise that two oranges and two pears are both occurences of the number 2, then you have grasped numerosity.Quote:
What the heck is numerosity?
I'm not sure I understand the meaing of "True AI" any more than I understand the meaning of "Intelligence". I suppose I mean artificial intelligence that is "trully" intelligent rather than only being called intelligent.Quote:
I don't see what you mean when you say 'true' AI, it doesn't really convey any meaning to me. Are you perhaps meaning intelligence of the same calibre as human intelligence?
I was refering to a different definition of spontinaety than you were, i think. When I use the word, I don't just mean "sudden" but also "without cause".Quote:
I suppose it comes down to are we really being spontaneous when we do something or are we trying to achieve something.
Thats one of the things I was tryinvg to get at. Do we ever really do anything without cause. I can't think of a none mad spontaneous act that does not have some payoff for the individual.Quote:
Originally posted by simonm
I was refering to a different definition of spontinaety than you were, i think. When I use the word, I don't just mean "sudden" but also "without cause".
Any examples anyone?
SD
The original answer l gave was a simplistic explanation of how you could AI. Naturally you don't just have a fruit rule, you have thousands of them.Quote:
Originally posted by Behemoth
I want to know why I spontaneous gibber to myself...Is this fuctionality available in AI.
Actually, what started as a silly response strikes a thought. Is AI spontaneous, and would AI talk to itself. We can teach ourselves things without outside input by clarifying or redefining things we already knew. ..
Bah I've done it again. I'll get my coat...
The project l was mentioning was used for pysch evaluations, and took a team of coders approx five years to code. The program was then considered to be in the just born stage. It has since grown itself through question and answers to be over a gig in size. Now, l am not claiming it has intelligence, but the project manager is getting real confussed about what it is coming up with. Answers and questions which were never intended by the original code.
AI is always going to be restricted by lack of senses. How do you get a PC to smell for example.
Simon
Sorry did you say something:rolleyes:
Surfdemon
Well there was that time l just got up and cleaned the lounge room.
"AI" look at Data from star trek, perfect example of somehting we don't know enough about. besides AI is just a man in the box.
computers are basically stupid until we tell them to do something.
Intelligence = somthing someone is doing better than someone else.
:D :D :cool: :D :D
But your right Scout, a lot of systems claimed to have AI are just glorified calculators.
Are we more than complex calculators though? That's not a rhetorical question, I don't know the answer for sure.
The problem of environmental perception (including senses such as smell, but can also include senses humans lack) is not insurmountable. I remember seeing machines that could analyse scents years ago on TV, they have come a long way since then. So yes, it's always going to be a problem in that it will require a solution if an AI with senses is to be built, but it's by no means insurmountable.
I'd tend to agree with SD on the point about spontanaity, I don't think any sane human being would do something without cause. They might not know why they're doing it because the reasoning is taking place in the subconcious, but there are still reasons. Well that's my opinion anyway.
The brain works by taking input, eg. sound, smell, sight etc. and then processing it through a series of links. Therefore what you may think is a spontanious thought actually may come from something you've just smelt and has passed through a series of subconscious links. That is possible to do with computers.
That's not really excluding the possibility of spontaneity. The brain is essentially just 'a series of links', yes, but that's the same as any kind of structure; mechanical, logical, anything.
What you are suggesting is that AI be achieved through replication of the links between neurons in the brain. There are millions (billions?) of nueurons in the brain all working in parallel, each with input from and output to many other neurons. To replicate such a structure would be impractical, for the foreseeable future anyway. Parallel systems can be simulated by a serial system (such as most computers are) but they are extremely slow.
Just you wait till quantum computers are out...
:D
Harry: You asked if we are just complex calculators i would prefer to look at our brain as a Orchestra with a conductor also with various sections as an orchestra does. eg woodwind! etc.
Undertand what i mean?
By the way the guy's at www.a-i.com are doing it is they are teaching it like you would a baby how to talk instead of saying here you go this is all the morals and all the words etc.
INTELLIGENCE:
Here's the reply from The Chief Sceintist at AI i put to him a statement of what intelligence is!
No, there's no right answer. I would say intelligence is in the eye
of the beholder. If the behaviour of something seems to be intelligent
then does it really matter how that behaviour comes about? If your
first impression of a computer program is that it is intelligent and
you later find out that it is nothing more than millions of if...then
statements would you change your mind?
The measure of intelligence we use here is the famous Turing Test. If
you can talk to a computer program and if it responds in a way that
makes it indistinguishable from a human being then we say it is
intelligent. It doesn't matter how it does it!
Cheers mate,
That sounds like a sensible take on the issue of measuring intelligence. I think the Turing test does require some level of intelligence, although it is a very domain-specific problem so the requirements are limited. The kind of thing it sounds like the a-i.com people are doing, with the AI building its own knowledge base through being taught, is what I would expect from a more general AI project. I think similar techniques are used for teaching a computer that can pass the Turing test - a lot of time is spent just conversing with the AI and it builds up its vocabulary. That ALICE program that's linked in a thread that's around here somewhere seems to work on this principle, although it's a long way from passing the Turing test at the moment.
Yes i agree Harry the guys at a-i get a lot of talk in New Scientist etc.
I also agree that alice is a long way off passing the turing test though it is pretty cool!
Magahal is better though!
Apparently the a-i guys have already fooled 8 out of 10 "Independent Experts":eek:
This is not what I meanth. I think you're confusing spontineatiy with Auturism.Quote:
Do we ever really do anything without cause. I can't think of a none mad spontaneous act that does not have some payoff for the individual.
Ok, you're deferring the possibility of spontaneous thought to the subconsious rather than the consious. That still leaves the question on whether notions can arise spontaneously in the subconsious or are these notions just the results of proir subconsious computation?Quote:
I'd tend to agree with SD on the point about spontanaity, I don't think any sane human being would do something without cause. They might not know why they're doing it because the reasoning is taking place in the subconcious, but there are still reasons. Well that's my opinion anyway.
It's a serious point because there needs to be spontineatity in the mind to allow for creativity (let alone free will).
I personnally think that to have intelligence requires the mind in question to know something that cannot be attained by reason or logic. It must be able to make intuitive leaps to grasp concepts that are otherwise out of it's reach. If this could be demonstrated, it would certainly "out think" any program that just happened to fool someone into passing the Turing test.Quote:
I think the Turing test does require some level of intelligence, although it is a very domain-specific problem so the requirements are limited.