I really did mean consider turning the engine of as a possibility before aiming for something to crash into. i.e. after going into neutral, trying the brakes etc.
But before ringing the police;)
Also, you could turn the engine off and on again. Wouldn't have helped me, but then my car didn't even know what a computer was. A quick reboot of the car's computers might have sorted it.
Seeing as you're sitting calmly at a computer and you still didn't consider turning the damn engine off as a possibility before aiming for something to crash into, it kind of begs the question as to why you think people panicking in a runaway car at 100 mph would be more cogent than they were.
I once had my car start to accelerate without my foot on the pedal (due to a build up of grease around the wire next to the engine), where I tried clearing the problem by hitting the accelerator a few times (which made it much worse), before going out of gear (hitting about 7000 revs and heading towards damaging the engine), then finally resolving it by turning the engine off and coasting to a stop without power steering. Believe me, you panic like f*ck and if I hadn't been on a quiet, straight country road I could have been in a heap of trouble.
Still, having enough time to call the police, I agree they weren't the smartest tools in the box.
We could begin to have a dialogue by placing a little trust in one another to quote things we know rather than things we've heard or swear we heard during some conversation some time ago. For that I'll retract my kitten statement. It was from a conversation long ago, I didn't read whatever paper was quoted then and I certainly can't find it now so I'll drop it and hope you can forgive the irrelevancy of my bringing it up in the first place.
Secondly, I'm not employed in the scientific establishment and I'm hoping this doesn't alter your perception of my arguments. I have a meagre degree in artificial intelligence and a masters in artificial life, so my background is reasonable academically. I am also currently employed as an AI programmer in the games industry, so my professional background is also reasonably well oriented.
The problem with trust on experimental evidence is one where only your own subjective viewpoint upon reading the papers might make you believe they have some worth.
So, look here for Ezequiel's paper http://www.cogs.susx.ac.uk/users/ezequiel/homeo.ps
It heavily suggests that neuron level internal, local stability (with local Hebbian-like adaptation) begets external, behavioural stability. This is a first point in suggesting that how transistor level mechanisms work has a fundamental effect on large scale (behavioural) processes. I'm not trying to suggest that perspective of behaviour isn't useful at any level between neuronal and large scale social, but that neuronal level experimentation is still useful in gross understanding of the process of intelligence.
I'm not condoning this animal experiment (although I'm not an anti-vivisectionist by any means) but I think I had some more interesting points than this one unsubstantiated experimental quote;-)
There are two lines of reasoning here that I'd like to comment on. First of all, the way he says the brain "is terrible at math, prone to errors, susceptible to distraction, and it requires half its uptime for food, sleep, and maintenance."
The fact that it requires upkeep is a fair one, you can't argue that it would be better if we required no sleep, food or maintenance and I'm sure if you were running the perfect operating system on perfect hardware then computers would require no power, no hardware replacements, no OS reboots no reintstalls, no disk defragmentation, no windows update and no process explorer (a handy little tool for killing any process on a PC dead without windows taking 30 minutes. It also tells you which processes are using which files and which registry keys). The truth is we don't have a perfect digital computer unless you're looking at systems that are built for very specific tasks and reduce their overall functionality (possibly stopping them being universal Turing machines).
The reason computers are good at maths and people aren't is difference in the domains we're dealing with. The computer's side, incorporates perfect world knowledge, in a digital system with direct abstraction between the problem domain and the system analysing that problem domain. Humans, on the other hand, have to deal with any number of problem domains that are fuzzy, unpredictable (often _just_ on the stochastic side of being random), subjectively viewed and highly complex. The reason we're so good at visual analysis of a situation compared to a computer is that visual analysis (or aural analysis for bats or chemo-analysis for dogs or tactile analysis for moles) is the essential partitioning of the world into happily grouped subsets upon which we can carry out the operations commonly considered intelligence.
The definition of an object is its functionality. Moving from any given set of stimuli to an analytical understanding of a situation requires taxonomical processing that is the chief role of a large portion of our brain (based on our interaction with those objects, defining their functionality for us). The whole role of logical or "higher" brain function is to cast an operator over a situation to produce an answer (also known as a new behaviour). This is essentially an easy problem and well understood/answered (to a given degree) by academics in every AI department in the land, using planning or expert systems or some such logical system designed to deal with such abstractions). The fact that a brain, that is designed for coping with fuzzy situation, is bad at well defined, concisely put, precise situations, is no surprise and a fair point.
The second point I'd like to raise is that of the sheer adaptive power of the human brain. There is no piece of digital computer software that can be given any input and, over time, come to adapt to minimise certain stimuli (pain, hunger, cold) and maximise others (pleasure, fullness, warmth and comfort). In essence the human brain is a massively complex homeostat, that internally adapts based on external stimuli and internal activity levels (see W Ross Ashby for a 1950 cyberneticists view of such systems, Humberto Maturana and Fransisco Varela for a Chilean biologists view of such things, Ezequiel Di Paolo's paper "Homeostatic adaptation to inversion of the visual field and other sensorimotor disruptions" for simulated empirical evidence or look up Spike-time dependant plasticity to see how Hebbian learning actually works in the brain).
It's been shown that it's possible to unplug the visual cortex and the aural cortex and switch them around (in kittens I believe) without undue effect on behaviour. The brain can adapt to whatever situation it is placed in from an initial position of no knowledge whatsoever. You can argue that this is software rather than hardware and that you could simulate whatever brain function there is on a computer with enough power but when Wallace states "My longstanding opinion is that neural networks are the wrong level of abstraction for understanding intelligence, human or machine.", he's ignoring the hardware altogether in suggestion that neurons and axons and ganglion cells are not important at all and that there's some higher level abstraction that is of more use (to further quote "But to me that does not rule out the possibility of reducing the mind to a mathematical description, which is more or less independent of the underlying brain architecture"). I don't believe this to be true or, at least, we would have to completely understand the interactions of this shitty computer before we understood its functionality so we could build our perfect mathematical reconstruction and implement it on a Turing machine (that only required the occasional disk-defrag, windows update and system tools to deal with blue screens of death).
Mike
"All of us strongly believe that machines are the next step in evolution," said Dunietz. "The distinction between real flesh and blood, old-fashioned and the new kind, will start to blur."
The distinction between machines that manipulate symbols and those that have some understanding of their environment is experience. Dunietz highlights the problems perfectly by remarking on the difference between flesh and blood and the "new kind". We exist in the world, we learn from experience, from a level of interaction that no learning machine undergoes. When your definition of the word 'tree' is your every experience-based interaction of actual _trees_ how can a machine hope to compare to that in terms of understanding without being embodied and situated in the same world as us.
There are millions of facts, not explicit, not definable with predicate calculus or any logical system, that you can describe with merely one second of being near a tree, that one thousand gender-non-specific hours inputting data will fail to give a Hal-like system. The only way to gain human intelligence is to be a human. The only way to truly communicate is across a consensual domain defined by similarity of experience. Everything else is just so much meaningless symbolic manipulation.
Slashdot Mirror
I really did mean consider turning the engine of as a possibility before aiming for something to crash into. i.e. after going into neutral, trying the brakes etc. But before ringing the police ;)
Also, you could turn the engine off and on again. Wouldn't have helped me, but then my car didn't even know what a computer was. A quick reboot of the car's computers might have sorted it.
Seeing as you're sitting calmly at a computer and you still didn't consider turning the damn engine off as a possibility before aiming for something to crash into, it kind of begs the question as to why you think people panicking in a runaway car at 100 mph would be more cogent than they were. I once had my car start to accelerate without my foot on the pedal (due to a build up of grease around the wire next to the engine), where I tried clearing the problem by hitting the accelerator a few times (which made it much worse), before going out of gear (hitting about 7000 revs and heading towards damaging the engine), then finally resolving it by turning the engine off and coasting to a stop without power steering. Believe me, you panic like f*ck and if I hadn't been on a quiet, straight country road I could have been in a heap of trouble. Still, having enough time to call the police, I agree they weren't the smartest tools in the box.
We could begin to have a dialogue by placing a little trust in one another to quote things we know rather than things we've heard or swear we heard during some conversation some time ago. For that I'll retract my kitten statement. It was from a conversation long ago, I didn't read whatever paper was quoted then and I certainly can't find it now so I'll drop it and hope you can forgive the irrelevancy of my bringing it up in the first place. Secondly, I'm not employed in the scientific establishment and I'm hoping this doesn't alter your perception of my arguments. I have a meagre degree in artificial intelligence and a masters in artificial life, so my background is reasonable academically. I am also currently employed as an AI programmer in the games industry, so my professional background is also reasonably well oriented. The problem with trust on experimental evidence is one where only your own subjective viewpoint upon reading the papers might make you believe they have some worth. So, look here for Ezequiel's paper http://www.cogs.susx.ac.uk/users/ezequiel/homeo.ps
It heavily suggests that neuron level internal, local stability (with local Hebbian-like adaptation) begets external, behavioural stability. This is a first point in suggesting that how transistor level mechanisms work has a fundamental effect on large scale (behavioural) processes. I'm not trying to suggest that perspective of behaviour isn't useful at any level between neuronal and large scale social, but that neuronal level experimentation is still useful in gross understanding of the process of intelligence.
I'm not condoning this animal experiment (although I'm not an anti-vivisectionist by any means) but I think I had some more interesting points than this one unsubstantiated experimental quote ;-)
There are two lines of reasoning here that I'd like to comment on. First of all, the way he says the brain "is terrible at math, prone to errors, susceptible to distraction, and it requires half its uptime for food, sleep, and maintenance." The fact that it requires upkeep is a fair one, you can't argue that it would be better if we required no sleep, food or maintenance and I'm sure if you were running the perfect operating system on perfect hardware then computers would require no power, no hardware replacements, no OS reboots no reintstalls, no disk defragmentation, no windows update and no process explorer (a handy little tool for killing any process on a PC dead without windows taking 30 minutes. It also tells you which processes are using which files and which registry keys). The truth is we don't have a perfect digital computer unless you're looking at systems that are built for very specific tasks and reduce their overall functionality (possibly stopping them being universal Turing machines). The reason computers are good at maths and people aren't is difference in the domains we're dealing with. The computer's side, incorporates perfect world knowledge, in a digital system with direct abstraction between the problem domain and the system analysing that problem domain. Humans, on the other hand, have to deal with any number of problem domains that are fuzzy, unpredictable (often _just_ on the stochastic side of being random), subjectively viewed and highly complex. The reason we're so good at visual analysis of a situation compared to a computer is that visual analysis (or aural analysis for bats or chemo-analysis for dogs or tactile analysis for moles) is the essential partitioning of the world into happily grouped subsets upon which we can carry out the operations commonly considered intelligence. The definition of an object is its functionality. Moving from any given set of stimuli to an analytical understanding of a situation requires taxonomical processing that is the chief role of a large portion of our brain (based on our interaction with those objects, defining their functionality for us). The whole role of logical or "higher" brain function is to cast an operator over a situation to produce an answer (also known as a new behaviour). This is essentially an easy problem and well understood/answered (to a given degree) by academics in every AI department in the land, using planning or expert systems or some such logical system designed to deal with such abstractions). The fact that a brain, that is designed for coping with fuzzy situation, is bad at well defined, concisely put, precise situations, is no surprise and a fair point. The second point I'd like to raise is that of the sheer adaptive power of the human brain. There is no piece of digital computer software that can be given any input and, over time, come to adapt to minimise certain stimuli (pain, hunger, cold) and maximise others (pleasure, fullness, warmth and comfort). In essence the human brain is a massively complex homeostat, that internally adapts based on external stimuli and internal activity levels (see W Ross Ashby for a 1950 cyberneticists view of such systems, Humberto Maturana and Fransisco Varela for a Chilean biologists view of such things, Ezequiel Di Paolo's paper "Homeostatic adaptation to inversion of the visual field and other sensorimotor disruptions" for simulated empirical evidence or look up Spike-time dependant plasticity to see how Hebbian learning actually works in the brain). It's been shown that it's possible to unplug the visual cortex and the aural cortex and switch them around (in kittens I believe) without undue effect on behaviour. The brain can adapt to whatever situation it is placed in from an initial position of no knowledge whatsoever. You can argue that this is software rather than hardware and that you could simulate whatever brain function there is on a computer with enough power but when Wallace states "My longstanding opinion is that neural networks are the wrong level of abstraction for understanding intelligence, human or machine.", he's ignoring the hardware altogether in suggestion that neurons and axons and ganglion cells are not important at all and that there's some higher level abstraction that is of more use (to further quote "But to me that does not rule out the possibility of reducing the mind to a mathematical description, which is more or less independent of the underlying brain architecture"). I don't believe this to be true or, at least, we would have to completely understand the interactions of this shitty computer before we understood its functionality so we could build our perfect mathematical reconstruction and implement it on a Turing machine (that only required the occasional disk-defrag, windows update and system tools to deal with blue screens of death). Mike
"All of us strongly believe that machines are the next step in evolution," said Dunietz. "The distinction between real flesh and blood, old-fashioned and the new kind, will start to blur." The distinction between machines that manipulate symbols and those that have some understanding of their environment is experience. Dunietz highlights the problems perfectly by remarking on the difference between flesh and blood and the "new kind". We exist in the world, we learn from experience, from a level of interaction that no learning machine undergoes. When your definition of the word 'tree' is your every experience-based interaction of actual _trees_ how can a machine hope to compare to that in terms of understanding without being embodied and situated in the same world as us. There are millions of facts, not explicit, not definable with predicate calculus or any logical system, that you can describe with merely one second of being near a tree, that one thousand gender-non-specific hours inputting data will fail to give a Hal-like system. The only way to gain human intelligence is to be a human. The only way to truly communicate is across a consensual domain defined by similarity of experience. Everything else is just so much meaningless symbolic manipulation.