 |
|
|
 |
|
|
|
|
Pennywise |
| | | | | | posted 10/22/2003 18:46 |  | | | | | | | | | | |
Geuis wrote @ 10/22/2003 12:08:00 AM:
I think that most people, even those heavily involved in AI research, are missing a critical, critical point to the whole game.
There is a reason that there has been only limited successes in AI research over the last 50 years, especially in the use of neural networks.
Real brains, animal or human, do not use algorithms.
| |
Is that to say that you don't think there is some algorithm that describes the operation of our brains/minds?
| | |
| | |
Nabarun Mondal |
| | | | | | posted 10/23/2003 04:48 |  | | | | | | | | | | |
Pennywise wrote @ 10/22/2003 6:46:00 PM:
Is that to say that you don't think there is some algorithm that describes the operation of our brains/minds?
| |
To be exact, it is hardly possible that operations of our brain can be described by formal ALGORITHMS.
The operation of 1 bit additon takes several million KT energy in joule, in computers, whereas, the same operation in brain takes some kelo calory. Which shows our brain is much more energy efficient than the Computers we made.
Now, if that is true, lowering the energy of computation requires less and less irreversibility, i.e. more "randomness". So, may be that our brain acts in some random way to have a ststistically predictable outcome.
MAY BE , I am not saying that IT IS CORRECT.
Only there is a chance....
| | |
| | |
Pennywise |
| | | | | | posted 10/23/2003 23:14 | | | | | | | | | | | |
Nabarun Mondal wrote @ 10/23/2003 4:48:00 AM:
To be exact, it is hardly possible that operations of our brain can be described by formal ALGORITHMS.
| |
Perhaps it's just my understanding of what an algorithm is, but I'm thinking that there's probably a set of rules and/or steps that could be applied to describe the way we think. This set of steps is independant of energy usage, whether the underlying mechnaism is silicon or biological, etc. That is to say that there are many functionally identical ways to implement these steps.
I would go on to say that these steps are called an "algorithm". And since computers are pretty good at implementing algorithms, they could probably implment the same set of steps that is used to create our minds. If we only knew what the steps were, then we could have an algorithm for the mind.
| | |
| | |
v
[Guest] |
| | | | | | posted 10/24/2003 00:33 | | | | | | | | | | | |
Pennywise wrote @ 10/23/2003 11:14:00 PM:
Perhaps it's just my understanding of what an algorithm is, but I'm thinking that there's probably a set of rules and/or steps that could be applied to describe the way we think.
| |
Unfortunately I failed to understand the deep meaning of it. I guess that this algorithm is very limited, but I still cannot reach all the depth of it. What algorithm do you mean exactly?
| | |
| | |
Pennywise |
| | | | | | posted 10/24/2003 07:08 | | | | | | | | | | | |
v wrote @ 10/24/2003 12:33:00 AM:
Unfortunately I failed to understand the deep meaning of it. I guess that this algorithm is very limited, but I still cannot reach all the depth of it. What algorithm do you mean exactly?
| |
I'm speaking of algorithms in general.
Dictionary.com defines an algorithm as "A step-by-step problem-solving procedure". This highlights the difference between how some people think of algorithms (eg- a computer program) and what an algorithm really is (just a series of steps).
It would seem to me that anyone interested in AI believes that there are rules that govern the brain and mind -- eg, rules of what you'll remember and for how long, rules for what will catch your attention, rules for recognition, etc. If you're to argue that there are no rules, you're essentially arguing that the brain and mind are outside the scope of science. (Since science is the observation of the world and the attempt to make generalizations of it, if generalizations (or rules, or laws) can't be made then it then it cannot be science.)
To argue that there is no algorithm that describes the mind is to argue that the mind is beyond the realm of science.
| | |
| | |
Rob Hoogers |
| | | | | | posted 10/24/2003 10:28 | | | | | | | | | | Errr. Some mathematical problems cannot be solved algorithmically. That does not make them 'beyond the realm of mathematics', just beyond 'the realm of the algorithmically provable'.
"There are many non-algorithmic problems in mathematics that cannot and will not ever be solved. Real numbers of such cannot be computed. Penrose states many aspects of mathematics, Diophantine arithmetic for example, which is nonalgorithmic. But his main argument lies on Gödel's Numbers and theory on mathematical proofs. Gödel's theorem states that in any formal system that uses standard arithmetic and accepted logical statements, one can create statements that can neither be proved true nor can the negation of the statement be proved true. If one were to run this mathematical statement through a Turing Machine, no answer could be found and it would run till time infinity."
So if 'the mind' does not run algorithmically, it still exists, and obeys some rules. It does not put in into some unreachable state, it just means you're not using the right tools for the job.
Science (as does Nature) has more tools than only algorithms, in short.
from:
| | |  Penrose | | | Last edited by Rob Hoogers @ 10/24/2003 10:33:00 AM | | |
| | |
v
[Guest] |
| | | | |
Geuis |
| | | | | | posted 10/24/2003 16:02 | | | | | | | | | | | |
Geuis wrote @ 10/22/2003 12:08:00 AM:
Real brains, animal or human, do not use algorithms. It makes me very, very weary to continually see all of this discussion on top-down algorithmic ANN's.
| |
The traditional approach that is most commonly used in creating neural networks is the idea that every neuron follows the same exact rules and will respond to the same input in the same way.
If this were so, then we could say we already had solved the theory and are just waiting until the hardware gets fast/small enough to make a true AI.
Sadly, this is not true.
One of the main hurdles in understanding consciousness is a seemingly simple question, but one we have yet to fully answer.
"How do you go from the one to one interactions of billions of neurons to what we and other animals perceive as 'consciousness' or 'self-awareness'?"
Sometimes I simplify this question with a simple graphic. Draw the top and bottom of a pyramid and leave the middle blank, like the top triangle is floating over the bottom quadralateral. Then draw a big question mark in the blank spot in the middle.
In a simple form the quadralateral represents all of our neurons interacting with each other on the base, one to one, level. The top of the pyramid, the little triangle, represents what we perceive in our own heads as consciousness.
The big question, again, is "What happens in the middle?"
Perhaps its more of a question of not "What happens?" but more of "How do trillions of interconnections lead to a self-aware state?"
This comes back to my original point is that people commonly misunderstand what an algorithm is.
1: get in car.
2: turn on car.
3: back out of driveway into street.
4: press gas to accelerate.
5: stop at red light.
6: go at green light.
7: make right turn.
8: make left turn into store parking lot.
9: park car in parking spot.
10: turn off car.
11: get out of car.
12: walk into store.
13: walk to dairy section.
14: get milk out of refrigerator.
15: walk to cashier.
16: put milk on counter
17: remove money from wallet.
18: give money to cashier.
19: get change from cashier.
20: get back cashier put milk into.
21: walk to car.
22: get in car.
23: turn car on.
24: back out of parking space.
25: make right at entrance to street.
26: make left at light.
27: pull into driveway.
28: turn off car.
29: get out of car.
30: walk into house.
31: go into kitchen.
32: open refrigerator.
33: put milk into refrigerator.
34: grab cold beer.
35: grab bag of chips.
36: walk to living room.
37: sit down on couch.
38: turn on tv.
39: change channel to football game.
40: open beer.
41: open chips.
42: proceed to consume.
This long ungainly thing is an annoying demonstration of what an algorithm is. Its simply list of steps used to solve a problem. In this case, it was getting milk from the store in time to watch the football game.
You must remember that things like mathematics and algorithms are simply human invented tools to help us understand the universe around us. Too often people forget that while the universe can be described very well by the mathematics we've invented, it is still simply a description to let human beings attempt to understand events external to our own minds. An alien species, with different corporeal forms and senses than our own, may also develop their own kind of mathematics to describe the kind of universe they see.
A fish, frog, dog, or human being perceive the universe in completely different ways. Fish live in water. Frogs don't see the same way we do. Basically if you are not moving in some way, a frog cannot see you. Dogs are color-blind but at least perceive things in a binocular way like us. And we see things the way we do.
Even in our own species, their are folks who have synathesia where some of their senses are crossed. They see smells, or hear colors. And no, they are not tripping on acid. Its just the way their brains put together the information it receives from the world outside.
So, what I originally stated about algorithms is that so many people think there is some set of steps that every neuron follows in some sequential pattern that tell it to respond to input in an exact way. They then try to use this same set of steps for every neuron in the network they are simulating. This is what I refer to as a "top-down algorithm."
I hate to burst some people's bubbles, but neurons are NOT sequential. About the only things that are sequential in our bodies is that our heart beats one beat after another. And food comes in and crap comes out.
Neurons, however, are simultaneously responding to all kinds of input and are having to make decisions on if/not to trigger themselves. They have to make new connections to other neurons or let current ones die. Additionally, the neurons tune their own excitability so that they seem to evaluate the best results of their interaction in the brain. They change their excitability and only afterwords compare the strength of the input and the threshold.
I've included a link on that article at the bottom.
The point being, while there may be ways of describing the behavior of neurons, a sequential list of steps that every single neurons follows exactly is simply not true.
We have to look beyond these simplistic models and develop models of behavior that accurately reflect the true nature of neural interactions.
| | | Neurons tune their own excitability | | | Last edited by Geuis @ 10/24/2003 4:16:00 PM | | |
| | |
Sam Fentress |
| | | | | | posted 10/24/2003 20:24 | | | | | | | | | | This is all true, but I'm still confused as to whether you're saying that our current ANNs are not good enough for the task, or whether the entire idea of algorithms existing in the brain is wrong. I think that this may be because you keep mixing up whether you are talking about the big things like consciousness, or the little things like neurons. Leaving aside consciousness (noone is saying that ANNs simulate consciousness, and we already have huge threads on this topic elsewhere), you seem to be saying that a neuron is not based on any sort of algorithm. This is the same as saying that they are non-deterministic (anything that is deterministic could theoretically be described by an algorithm). Leaving out for the moment question such as free will, if we isolate a neuon in a petri-dish and control its environment perfectly, I would certtainly contend that, if we know enough about the neuron, we would be able to predict its actions. It's perfectly possible that we do not currently know enough about neurons - your article is an example of how complicated individual neurons can be. But nowhere does your article state that neurons are not bound by physical laws that we expect to eventally understand. The article doesn't say that neurons are completly random either. Thus, these two things together imply that neurons are deterministic, and can therefore be described by an algorithm. Logical induction would thus imply that if each component of the brain can be described by an algorithm, then any set of such components could be also described by such.
| | | Last edited by Sam Fentress @ 10/24/2003 8:27:00 PM | | |
| | |
Geuis |
| | | | | | posted 10/24/2003 23:28 | | | | | | | | | | | |
Sam Fentress wrote @ 10/24/2003 8:24:00 PM:
This is all true, but I'm still confused as to whether you're saying that our current ANNs are not good enough for the task, or whether the entire idea of algorithms existing in the brain is wrong. I think that this may be because you keep mixing up whether you are talking about the big things like consciousness, or the little things like neurons. Leaving aside consciousness (noone is saying that ANNs simulate consciousness, and we already have huge threads on this topic elsewhere), you seem to be saying that a neuron is not based on any sort of algorithm. This is the same as saying that they are non-deterministic (anything that is deterministic could theoretically be described by an algorithm). Leaving out for the moment question such as free will, if we isolate a neuon in a petri-dish and control its environment perfectly, I would certtainly contend that, if we know enough about the neuron, we would be able to predict its actions. It's perfectly possible that we do not currently know enough about neurons - your article is an example of how complicated individual neurons can be. But nowhere does your article state that neurons are not bound by physical laws that we expect to eventally understand. The article doesn't say that neurons are completly random either. Thus, these two things together imply that neurons are deterministic, and can therefore be described by an algorithm. Logical induction would thus imply that if each component of the brain can be described by an algorithm, then any set of such components could be also described by such.
| |
By *NO* means am I one of these people who go around touting that it is impossible to know how the brain fundamentally functions at the neural level because of some mystical BS that makes knowing the nature of conscioussness impossible.
To be unlike a politician, to clear up your confusion I shall be more specific. I do not believe most current ANN research being done are, by themselves, going to give us the fundamental theory breakthrough that lets us understand how you go from neural interactions to larger things like individual behaviors.
I'm not even really talking about consciousness at this point. It's a three tiered theorem that is currently being worked on, but in the wrong ways.
TIER 1: Neurons connect to each other, send signals back and forth, make decisions internally about those signals, and if needed, make new connections and let old ones die.
TIER 2: Groups of neurons connected together form subnets that respond to input and output as a group and the more individualistic behaviors at the Tier 1 level seem subsumed by the larger group behavior. What is the theory that accurately can explain how you go from the Tier 1 behaviors to Tier 2 group behaviors?
TIER 3: Subnets act in collusion with other subnets based on certain input/output from their groups. These subnets are what we monitor using our tools such as EEG's and PET scans. Such tools are primitive but will be much more powerful in the next 5-7 years and will let us refine our scans to a much greater resolution, getting closer to being able to directly montior on a broad scale Tier 2 subnet activities. The subnets, also commonly known as different "regions" of the brain, all act together to generate different behaviors. The theory we are lacking at this level is how do all of these different regions acting together give rise to what lies at the top of the pyramid I described earlier, our self-awareness or conscioussness.
These 3 theorems, if there really are 3, will be able to be mathematically described. These are the keys of AI that will open up the entire field into a new golden age.
This all relates to algorithms in this way: The kinds of algorithms that people always talk about are *sequential*. Step 1, step 2, step 3, etc. The kinds of maths that are needed to develop the theorems of the three Tiers of AI are going to be found in the maths that network engineers are developing to understand large-scale networks like the internet, social systems, and biological interactions. They will not be found in traditional mathematics like everyone thinks.
To illustrate a point, I'll mention a story from when I was in chemistry and physics in high-school, and even in college. I had a very hard time in physics because the way that it was taught made no sense. The teachers, the books, and the other students all insisted on using models where certain forces were isolated from the rest of the universe and any other variables that might affect the problem were ignored. It was explained a multitude of times that in the instances of studying this particular force or this particular chemical reaction, the idea was to isolate them to learn just about those interactions.
This only makes sense in the mathematical Western tradition stretching back well over 400 years now. I could never get anyone to *really* understand that nothing works in isolation. There is little point in breaking everything down into its component parts in an attempt to understand the whole event becaue the event is destroyed in the process.
I firmly believe there is an alternate math that is radically different from traditional Western mathematics that fundamentally stresses understanding the interactions of disparate elements and how those interactions combine to form the event.
In traditional physics like I was taught, when you bounce the ball on the wall, you only focus on the speed of the ball, its mass, and how hard the wall is in order to predict where the ball will go. While at an elementary level traditional mathematics work for this situation, even leading up to things like orbital mechanics, they are very limited.
To me, you have to figure into your calculation not just the speed, mass, and hardness of the wall, but also the atmospheric pressure, the microscopic surfaces of the ball and the wall, how the individual atoms interact, how the photons of light and cosmic particles in the area affect the interaction of the ball and the wall.
Most people think that all of these additional variables are un-needed in the calculation. Those people are still trapped in the 400 year old paradigm of Western mathematics. What is needed is a new math that can be used not only in this kind of problem, but in really complex things where traditional maths begin to fail. Such areas are large scale networks, biological systems, simulating protein folding, ecologies, evolution, learning, conscioussness, cosmology, etc, etc.
There is a problem involving medicine that cannot be solved by Western mathematical/scientific traditions. If anyone gets DirectTV, watch WorldLink TV ch. 375. I forget the exact name of the show, but its about a researcher who spent a few years in South America researching traditional medicines used by shamans who have learned and passed down their medical knowledge for generations. He took some of these medicines back to the U.S. where researches attempted to isolate the active compounds from the medicine to determine what the chemicals were and how they might be synthesized.
For those not familiar with this process, its fairly simple but time consuming. The researchers divide the sample material into different parts and then test those individual parts to see which still responds positively to tests. They then take that part and sub-divide it again and test it. This process continues until they have isolated all extraneous chemicals and have narrowed it down to the active chemical in the sample.
This process works fairly well, for it has given us a lot of good medicines over the years. The problem with the South American samples is that at a certain point when they divided the samples basically all the way down to their individual chemicals, the medicine stopped giving a response. The reason why is that the shamans understand that it is not a singular compound that gives the result, it is the interactions of different compounds acting together that give the desired result in the patient.
To this day, Western scientists have yet to explain how this type of medicine works.
With a methodology based on understanding the interactions between disparate elements, we might have a chance of finding new medicines at a faster rate that are even more effective and with less side-effects. One reason so many medicines now have so many side effects is that researchers don't know how to test medicines for bodily reactions as well as they might. A combination of compounds might mitigate some side-effects while making a bigger impact on the disease.
I would recommend a book by James Bailey, "After Thought: the computer challenge to human intelligence." In the book he talks about many of the points I've raised.
Sam, in another topic on the Learning Engine I mentioned that perhaps language may follow scale-free network architecture. I've done a little more testing, and I'm finding more and more it appears language does. I don't know what this means, as it probably doesn't mean anything by itself. But in combination with some other ideas or methodologies, the knowledge that language is structured in the same way the Internet and our brains are could reveal something very useful.
I'll be more than happy to elaborate more on anything I've said here if I'm still confusing to anyone.
| | |
| | |
Pennywise |
| | | | | | posted 10/25/2003 08:07 | | | | | | | | | | | |
Rob Hoogers wrote @ 10/24/2003 10:28:00 AM:
[1] That does not make them 'beyond the realm of mathematics', just beyond 'the realm of the algorithmically provable'.
[2] So if 'the mind' does not run algorithmically, it still exists, and obeys some rules. It does not put in into some unreachable state, it just means you're not using the right tools for the job.
[3] Science (as does Nature) has more tools than only algorithms, in short.
| |
[1] Who said anything about "algorithmically provable"? I was simply talking about algorithms. And not even "mathematical algorithms," with which they may be confused.
[2] So, let me try to understand your arguement. You're saying that there are rules that underlie the mind, but they don't form an algorithm? Is this true of all natural systems that obey rules, or is the mind a special exception? Physics and chemistry have rules (indeed, laws) -- can these rules be applied in the form of an algorithm?
[3] Although it could be argued that the only tool of science is the scientific method, which is an algorithm, I won't go there.
I don't really see an algorithm as a tool of science so much as a result. Science, as I described ea
Be careful with Penrose. I do believe he's trying to prove out that Artificial Intelligence is impossible. Do you subscribe to those theories?
For those of you unfamiliar with Penrose, I've found these to be good resources:
http://online.itp.ucsb.edu/online/plecture/penrose/
http://www.friesian.com/penrose.htm
http://www.tribunes.com/tribune/art99/ham3.htm
| | |
| | |
Rob Hoogers |
| | | | | | posted 10/25/2003 11:13 | | | | | | | | | | | |
Pennywise wrote @ 10/25/2003 8:07:00 AM:
[1] Who said anything about "algorithmically provable"? I was simply talking about algorithms. And not even "mathematical algorithms," with which they may be confused.
[2] So, let me try to understand your arguement. You're saying that there are rules that underlie the mind, but they don't form an algorithm? Is this true of all natural systems that obey rules, or is the mind a special exception? Physics and chemistry have rules (indeed, laws) -- can these rules be applied in the form of an algorithm?
[3] Although it could be argued that the only tool of science is the scientific method, which is an algorithm, I won't go there.
I don't really see an algorithm as a tool of science so much as a result. Science, as I described ea
Be careful with Penrose. I do believe he's trying to prove out that Artificial Intelligence is impossible. Do you subscribe to those theories?
For those of you unfamiliar with Penrose, I've found these to be good resources:
http://online.itp.ucsb.edu/online/plecture/penrose/
http://www.friesian.com/penrose.htm
http://www.tribunes.com/tribune/art99/ham3.htm
| |
[1] 'Provable' and 'solvable' are very close cousins. Some real-life problems are not even 'solvable' algorithmically, then.
[2] There is always an algorithm you can use in the knowledge that is does not describe perfectly what is going on. To predict unknowns via that same algorithm is what proves impossible. They generally end up in infinities, making it utterly unusable. (QM and gravitons, etc.)
[3] I see the end bit got mangled there, so I'll only refer to [2] which at least concurs with the 'result' bit.
About Roger Penrose: he's brilliant, and I'm a great fan. He can also be wrong at times, though. Like everyone else.
As to the impossibilty of AI according to Penrose:
- with current tech-levels I'd say he's probably right, which would basically mean stick to some form of organic until we sussed out more. We have encountered quite a few problems (emotional drive, survival traits, social behaviour) that don't look as if they're solvable without biological imperatives underlying such an attempt at creating consciousness. Maybe a consciousness *has* to be aware of death, in short. Even if that awareness is pushed away from everyday existence as deeply as possible, as is the case with us modern humans.
But thanks for the warning anyway. Although I think that Godel bashing link a few posts back is ultimately more misleading, in the sense that it undermines without anyone really having the math to follow it. Penrose doesn't, although he's umpteen times more qualified to do so. He's one of our best mathematicians ever, period.
| | |
| | |
Pennywise |
| | | | | | posted 10/25/2003 19:27 | | | | | | | | | | | |
Rob Hoogers wrote @ 10/25/2003 11:13:00 AM:
[1] 'Provable' and 'solvable' are very close cousins. Some real-life problems are not even 'solvable' algorithmically, then.
[2] There is always an algorithm you can use in the knowledge that is does not describe perfectly what is going on. To predict unknowns via that same algorithm is what proves impossible. They generally end up in infinities, making it utterly unusable. (QM and gravitons, etc.)
[3] I see the end bit got mangled there
[4] with current tech-levels I'd say he's probably right, which would basically mean stick to some form of organic until we sussed out more.
[5] We have encountered quite a few problems (emotional drive, survival traits, social behaviour) that don't look as if they're solvable without biological imperatives underlying such an attempt at creating consciousness. Maybe a consciousness *has* to be aware of death, in short. Even if that awareness is pushed away from everyday existence as deeply as possible, as is the case with us modern humans.
| |
[1] Oh? Like what? And what do humans do when they run into these problems?
Even then, as Geuis's 10/22 posting points out, an algorithm may not be in search of a solution in terms of math. Well, unless you can come up with an equation that represents fetching milk...
[2] So, you're saying that both Physics and AI may be able to be descibed by an algorithm, but neither can describe the unknowns of their own systems?
[3] Damn cut and paste. Where I was essentially going was that science creates these generalizations (or rules) through observing nature. After these rules have been acheived, they can be applied to problems in an algorithmic fashion.
[4] So then, just to be sure, you don't believe that AI can be performed on computers as they exist today? But you do believe that AI can be performed through biological material?
[5] And thus the joys of talking about consciousness. In listening to the UCSB lecture from Penrose, it's interesting to see that his arguments are philosophically epistemological and his complaints about computing are not against computers per say (although he would attempt to make it seem that way) but rather against the current algorithms which they run.
I also found this slide interesting:
http://online.itp.ucsb.edu/online/plecture/penrose/oh/04.html
I'm not sure how many people would believe that "intelligence" requires "understanding" -- especially when Penrose refuses to define both understanding and awareness, but instead (even to his own admission) talks about things he knows nothing about. The pretext of the entire conversation is that he's going to show the relationship between these things, but I find it hard to swallow that he's comparing and contrasting undefined quantities.
| | |
| | |
Rob Hoogers |
| | | | | | posted 10/26/2003 11:36 | | | | | | | | | | [1] An algorithm for fetching milk: ants can, and do (on aphids).
[2] Every theory in physics and AI is an approximation of reality. Scientific rigor insists on such theories to produce repeatable results, i.e having a high algorithmic content. As soon as inconsistencies turn up, the algorithm fails horribly, since it was only an approximation (a result, or more exactly an intermediate result).
[3] see 2. Your inclination as always is to think we've sussed out the essentials already and you want to be able to project yourself cleanly and scientifically towards an endsolution, preferably within your lifetime.
That almost never turns out to be the way science worked in the past. It is a much more messy affair.
[4] Either via straight ACI's (guess...), or at least by incorporating bees, funguses, and other biological media in direct computations, yes.
[5] I would disagree with him there. Our computers are incapable, per se.
| | |
| | |
Nabarun Mondal |
| | | | | | posted 10/27/2003 12:05 | | | | | | | | | | | |
Rob Hoogers wrote @ 10/26/2003 11:36:00 AM:
[2] Every theory in physics and AI is an approximation of reality. Scientific rigor insists on such theories to produce repeatable results, i.e having a high algorithmic content....
| |
That is why AI is considered *unsuccessful*. ;-)
Too much predictability!
Even a bee, if someone tries to kill it, behaves in an unpredictable manner !
Possibly that is the biggest setback of the current AI research.
Real world is not predictable beyond certain limit, and so the simulation should also be unpredictable... to simulate correctly.
| | |
| | |
Geuis |
| | | | | | posted 10/27/2003 14:22 | | | | | | | | | | | |
Nabarun Mondal wrote @ 10/27/2003 12:05:00 PM:
That is why AI is considered *unsuccessful*. ;-)
Too much predictability!
Even a bee, if someone tries to kill it, behaves in an unpredictable manner !
Possibly that is the biggest setback of the current AI research.
Real world is not predictable beyond certain limit, and so the simulation should also be unpredictable... to simulate correctly.
| |
The world is infinitly predictable. The limiting factor is your ability to know and simulate variables in a given system.
Weather forcasting used to be walking outside to see if its raining. Now we are able to predict a hurricane's direction a week from landfall. We are increasingly able to create better and more accurate long-term global climate simulations.
These abilities come from a) a massive increase in available computing power(i.e. the ability to simulate variables) and b) increased knowledge in weather theory and c) better, more accurate historical climate data taken from sources such as ice cores, bog(marsh) material, and studying growth rings on trees from old growth forces.
If you try to kill a bee, its response isn't unpredicatable. It'll try to sting you!
| | |
| | |
Nabarun Mondal |
| | | | | | posted 10/28/2003 04:50 | | | | | | | | | | | |
Geuis wrote @ 10/27/2003 2:22:00 PM:
If you try to kill a bee, its response isn't unpredicatable. It'll try to sting you!
| |
That is for the first or second time... and after that it will try to avoid you...and then to catch it what *algorithm* will you take???
| | |
| | |
Geuis |
| | | | | | posted 10/28/2003 14:50 | | | | | | | | | | | |
Nabarun Mondal wrote @ 10/28/2003 4:50:00 AM:
That is for the first or second time... and after that it will try to avoid you...and then to catch it what *algorithm* will you take???
| |
I'm sorry Nabarun, I don't quite understand what you mean:
"and then to catch it what *algorithm* will you take???"
I hate to get stuck on the subject of the bee, since its a hive based insect and its social patterns are genetic, not really learned. Bees have great ability to learn about the world around them and also to learn from each other, but this is in the very strict sense of the location of food. Social behavior is inborn genetically, and recent studies indicate that a worker's behavior changes predictably as it ages.
North American honey bees live about 6 weeks. In that time, their behaviors change from first being caretakers in the hive nurseries, to a few weeks later being the drones that go outside the nest to find food. After a few more weeks of this, they die.
| | |
| | |
Ribald
[Guest] |
| | | | | | posted 10/28/2003 16:28 | | | | | | | | | | I would suggest that there are two definitions of "algorithm" being bandied about here.
The term to a mathematician is much closer to what is called "formal methods" in computer science. In computer science an algorithm is indeed as described merely a procedure, not a formula.
I suggest reading Wolfram's work, as that it illuminates that a system can produce results that can not be linearly predicted. That doesn't mean that it isn't deterministic or algorithmic.
Penrose completely failed to separate the two concepts, which is why Godel's theorem is naively used as "proof" that the brain is not A UTM. The emphasis on "A" because if you simply add another UTM to create a system most of the Godel based arguments fall apart.
We do not die of infinite loops because we have hierarchal structures in the mind, and our brain uses depletable resources while processing. When something is too hard we decide it is impossible. Place one UTM in the position of monitoring another and you break the "endless loop" problem.
The monitoring UTM will not be able to know with 100% certainly that the first UTM is never going to finish, but neither does our brains. We are frequently mistaken in both directions.
| | |
| | |
Nabarun Mondal |
| | | | | | posted 11/3/2003 06:11 | | | | | | | | | | | |
Ribald wrote @ 10/28/2003 4:28:00 PM:
I would suggest that there are two definitions of "algorithm" being bandied about here.
We do not die of infinite loops because we have hierarchal structures in the mind, and our brain uses depletable resources while processing. When something is too hard we decide it is impossible. Place one UTM in the position of monitoring another and you break the "endless loop" problem.
The monitoring UTM will not be able to know with 100% certainly that the first UTM is never going to finish, but neither does our brains.
| |
Exactly! Our brain does it Randomly, well almost randomly! The monitoring UTM randomly selects the exiting time. And our brain also does the same.Then comes again the question of Randomness in Intelligent beings...
| | |
|
|
|
|
|
