Not unexpectedly, Watson's dominating Jeopardy appearance is being billed as man vs. machine, but that's not at all accurate. Watson doesn't "remember" things, or think, or even really process. He knows grammatical rules, and is programmed kinda' sorta' with a custom version of Google. Imagine if someone playing Jeopardy was able to type very quickly, and when a question was asked, they could type that question into a search engine. The average person could dig up the answer very quickly, and we certainly wouldn't claim that person "won" anything. That's what Watson is doing. In fact, it's not even impressive enough to be voice-recognition. He receives the question via text.
That's not to say that Watson isn't incredible. He is. Essentially, the advances made in Watson could be put into newer and better search engines, letting users find answers to actual questions. Much like Ask Jeeves used to advertise, back in the day. Currently, finding difficult answers in search engines requires skill, silly as it sounds, in searching. You have to know how to use key terms, how to combine them, which tangents to follow, etc. Watson could significantly reduce that requirement.
Showing posts with label artificial intelligence. Show all posts
Showing posts with label artificial intelligence. Show all posts
Thursday, February 17, 2011
Wednesday, January 26, 2011
Artificial Ridiculousness
I'm watching Nova Science Now on living forever, and there's a section on artificial intelligence. In one scene, a researcher talks about how the avatar isn't gesticulating, and they need to start thinking about making the avatar communicate with its hands.
To me, this illustrates the reason why the traditional approach to artificial intelligence will never work. We cannot build an intelligence from the top down. We can't copy the gross aspects of life, i.e. behaviors, we need to figure out the basic building blocks that generate that behavior.
This comes back to the computational theory of mind. We do have to program certain basic assumptions about the world, but the fewer, the better.
I think that the basic programming is going to be based on a binary-like yes/no language. Light is good, darkness is bad. Warmth good, cold bad, etc.
To me, this illustrates the reason why the traditional approach to artificial intelligence will never work. We cannot build an intelligence from the top down. We can't copy the gross aspects of life, i.e. behaviors, we need to figure out the basic building blocks that generate that behavior.
This comes back to the computational theory of mind. We do have to program certain basic assumptions about the world, but the fewer, the better.
I think that the basic programming is going to be based on a binary-like yes/no language. Light is good, darkness is bad. Warmth good, cold bad, etc.
Tuesday, July 21, 2009
iRobot Revisited.
I think about AI a lot. Read about. Write about. Masturbate to it (sexbots... Mmmm yeah). I still think that building AI is going to be impossible.
I look at it like a tree. Yes, we understand the basic model of a tree. We know where parts go in relation to one another, how the systems work, etc. But still, we can't build a tree from the top down. We can't engineer branches here, and leaves there. Because even though all trees are pretty similar, they are also wholly unique. Defined by the manifold variables to which they are exposed as they grow.
Instead, we start with a seed and let it grow. We know the primary, fundamental variable required to grow a tree, the seed, but we cannot construct a tree artificially. I think the various learning robots being created, where they respond to and learn from their environments, are the first steps in the creation of non-biological intelligence.
I use that term because I don't consider them artificial at that point. They've grown of their own action. Yes, we created the stuff at the beginnning, but that's like calling a GE ear of corn artificial corn. It just doesn't make sense. It's non-biological, but it has grown naturally, in the same way our own intelligence grew.
I look at it like a tree. Yes, we understand the basic model of a tree. We know where parts go in relation to one another, how the systems work, etc. But still, we can't build a tree from the top down. We can't engineer branches here, and leaves there. Because even though all trees are pretty similar, they are also wholly unique. Defined by the manifold variables to which they are exposed as they grow.
Instead, we start with a seed and let it grow. We know the primary, fundamental variable required to grow a tree, the seed, but we cannot construct a tree artificially. I think the various learning robots being created, where they respond to and learn from their environments, are the first steps in the creation of non-biological intelligence.
I use that term because I don't consider them artificial at that point. They've grown of their own action. Yes, we created the stuff at the beginnning, but that's like calling a GE ear of corn artificial corn. It just doesn't make sense. It's non-biological, but it has grown naturally, in the same way our own intelligence grew.
Friday, February 22, 2008
I Vacuum.
I got a pair of cleaning robots for Christmas: The iRobot Roomba 580 and the Scooba 380. They're the high-end variants of each model and after the last two months of pretty solid use, I must say, I'm pretty disappointed. I wasn't expecting the moon or anything, but the adventures I've had really don't justify the price.
The Roomba, for example, coming in at over $500, seems to have the dexterity and software complexity of robots I build back in high school with a TI-83. I'm not exaggerating. The vacuum turns on with this triumphant NES theme song, hums to life, and rockets away in search of a place to get stuck. It starts humping everything in sight, bashing into things, and wedging itself into nooks corners like some crazed hamster. It's suction power is such that any dirt that lies more than a femtometer beneath the surface of the carpet is well outside of the vacuume range. Hard floors work well, and large bits on the carpet surface are easily captured, but pet hair? HA!
You might as well try to clean up pet hair with hope. Now, assuming you have a room with almost nothing in it, and the dirt is fresh, the Roomba will get it. But if your house is the slightest bit lived-in, the vacuum will be utterly confounded by anything and everything you have. Bookbags, pets, shoes, magazines, carpets, chairs, bags, children, clothing, newspapers, wires, cords, and just about anything else you can think of will royally fuck with this robot's world.
Because of this, you cannot use the Roomba's most useful feather, namely, setting the timer to have it automatically clean at some set time. Because, if you should ever try this, you will be greeted every morning by your Roomba stuck in the ceiling fan. You would need to keep your house impeccably clean, just to make sure that the Roomba was able to clean it. And even then, it may decide that it just doesn't like your drapes and try to eat them. Or it will decide that it needs to be cleaned. Or it will just beep a lot and refuse to do anything because it has a cat stuck in its brush.
The Roomba 580, $500, does the same job as the basic Roomba 400, namely, a bad one. I expected very little from the basic Roomba which came free with the Scooba, and I got it. It does a decent job at cleaning up little messes. But for $500, I expected a lot more. I expected an LCD display to tell me things. I expected a vacuum which is able to suck more than just air. I expected a robot with enough sensors to adequately navigate an environment. I feel pretty confident that I could do a better job with a Lego Mindstorms kit, car battery, Hoover canister vac, and some RC car parts.
As it stands, I still vacuum once a day with the clunky Dyson. The Roomba, in all its technical glory, vacuums the tile floor in the kitchen. Which is then followed by the bigger disappointment, the Scooba. Also ringing in at $500, the Scooba does a fantastic job of getting everything very wet and then running out of battery. I won't deny that the water it sucks up is dirty, but the floor never looks any cleaner. It takes the dirty floor and makes the dirt shiny. And do you remember, when you were a kid, those battery powered cars (mine was a fire truck) that just drove in a straight line until it hit something, backed up, and went in another direction? They usually had that set of rotating wheels on the bottom. Well, the Scooba is one of those with a wet-vac attached. For $500. For $500 I could buy a Mexican to spray detergent everywhere and then suck it off the floor. It has no sensors, no advanced programming designed to develop a map of its environment, no interface with which to program it. Nothing. It just crashes around and gets stuck under the dish washer.
All of this seems created for people that have fake messes. You know those Swiffer mops and brooms. Yeah, for those messes. They're made for messes that people don't actually make. Like that puddle of grape juice on the otherwise glazen floor. Or those small plastic pellets that people in vacuum commercials always seem to be spilling. And they're always cleaned up by the mother who seems to be a state of bemused resignation that her children, husband, and pets are all retarded slobs. But, through the aid of these amazing products, she keeps the house so clean you could perform open heart surgery on the couch! These are not products for people who have actual messes. Mud tracked into the house? Get down on your knees and scrub. Pet hair? Get out the Hoover. An otherwise coruscating house and you want to pretend that you need stuff cleaned? IRobot has the answer!
The Roomba, for example, coming in at over $500, seems to have the dexterity and software complexity of robots I build back in high school with a TI-83. I'm not exaggerating. The vacuum turns on with this triumphant NES theme song, hums to life, and rockets away in search of a place to get stuck. It starts humping everything in sight, bashing into things, and wedging itself into nooks corners like some crazed hamster. It's suction power is such that any dirt that lies more than a femtometer beneath the surface of the carpet is well outside of the vacuume range. Hard floors work well, and large bits on the carpet surface are easily captured, but pet hair? HA!
You might as well try to clean up pet hair with hope. Now, assuming you have a room with almost nothing in it, and the dirt is fresh, the Roomba will get it. But if your house is the slightest bit lived-in, the vacuum will be utterly confounded by anything and everything you have. Bookbags, pets, shoes, magazines, carpets, chairs, bags, children, clothing, newspapers, wires, cords, and just about anything else you can think of will royally fuck with this robot's world.
Because of this, you cannot use the Roomba's most useful feather, namely, setting the timer to have it automatically clean at some set time. Because, if you should ever try this, you will be greeted every morning by your Roomba stuck in the ceiling fan. You would need to keep your house impeccably clean, just to make sure that the Roomba was able to clean it. And even then, it may decide that it just doesn't like your drapes and try to eat them. Or it will decide that it needs to be cleaned. Or it will just beep a lot and refuse to do anything because it has a cat stuck in its brush.
The Roomba 580, $500, does the same job as the basic Roomba 400, namely, a bad one. I expected very little from the basic Roomba which came free with the Scooba, and I got it. It does a decent job at cleaning up little messes. But for $500, I expected a lot more. I expected an LCD display to tell me things. I expected a vacuum which is able to suck more than just air. I expected a robot with enough sensors to adequately navigate an environment. I feel pretty confident that I could do a better job with a Lego Mindstorms kit, car battery, Hoover canister vac, and some RC car parts.
As it stands, I still vacuum once a day with the clunky Dyson. The Roomba, in all its technical glory, vacuums the tile floor in the kitchen. Which is then followed by the bigger disappointment, the Scooba. Also ringing in at $500, the Scooba does a fantastic job of getting everything very wet and then running out of battery. I won't deny that the water it sucks up is dirty, but the floor never looks any cleaner. It takes the dirty floor and makes the dirt shiny. And do you remember, when you were a kid, those battery powered cars (mine was a fire truck) that just drove in a straight line until it hit something, backed up, and went in another direction? They usually had that set of rotating wheels on the bottom. Well, the Scooba is one of those with a wet-vac attached. For $500. For $500 I could buy a Mexican to spray detergent everywhere and then suck it off the floor. It has no sensors, no advanced programming designed to develop a map of its environment, no interface with which to program it. Nothing. It just crashes around and gets stuck under the dish washer.
All of this seems created for people that have fake messes. You know those Swiffer mops and brooms. Yeah, for those messes. They're made for messes that people don't actually make. Like that puddle of grape juice on the otherwise glazen floor. Or those small plastic pellets that people in vacuum commercials always seem to be spilling. And they're always cleaned up by the mother who seems to be a state of bemused resignation that her children, husband, and pets are all retarded slobs. But, through the aid of these amazing products, she keeps the house so clean you could perform open heart surgery on the couch! These are not products for people who have actual messes. Mud tracked into the house? Get down on your knees and scrub. Pet hair? Get out the Hoover. An otherwise coruscating house and you want to pretend that you need stuff cleaned? IRobot has the answer!
Sunday, February 10, 2008
AIn't.
I see a fundamental problem with artificial intelligence. I haven't read any other heavy criticisms of the idea that AI is possible, save for an issue of Skeptic Magazine. Still, after a great deal of thought and research, and including my own expertise in psychology and neurology, I seriously doubt we will ever achieve AI. At least not AI in the classic, sci-fi sense.
For that kind of AI, we need to be able to construct self-awareness from the outside in. Basically, we study the brain, we study how it works, learn the language of thought as we would a computer language, and than construct a program that is self-aware. I really doubt this is possible. because of the way in which intelligence arises. It doesn't arise from the outside in. A creature never came along (Unless you're a Catholic) and just imbued something with self-awareness. The revelation of self awareness comes from within, when the entity evolves into it.
I don't think there is any "language" of self awareness since no language is required. As neuroscience shows time and time again, the human brain can function with different layouts, sizes, and structure to identical ends. It's because each computer programs itself as it grows. It's because of that that I doubt we will ever be able to program something ourself.
What we could do is create a network of computers with similar basic programming to the primordial brain, and just watch and hope that some sort of consciousness arises within the system. I think this is a possibility. Granted, we don't have the technology to do it, but it's a possibility. Unfortunately, that doesn't fulfill the God desire: the desire to create actively, not passively. It's not AI per se.
In fact, it may be an artificial environment, a computer one, but the development of the intelligence would be very natural. One that follows the rules of the system laid down but otherwise meanders about its own path. And, again, in the end we would be unable to read the code that was created and "see" the intelligence since the code is only readable by the code. Any pattern works. Any code. Any compiler, as it were. It doesn't matter. The syntax and whatnot come from itself. One aware orgamism can be 99% different from another aware organism. The basic structure of our brains may be the same, but for all we know, every brain, in every person EVERWHERE "speaks" another programming language, since the logic to the whole mess only needs to make sense to itself.
I think the movie Bicentennial Man is a good representative of how AI may arise. We may create a system that is intended to be as close to human as possible, and during its development, that one-in-a-million developmental direction takes place and a robot becomes self-aware. But what kind of awareness would it be? How do we program a system of neurochemical awards? Could we create a robot that can get addicted to drugs? Our own consciousness stems from these basic concepts. We cannot create the gross elements of our behavior and build down. It is impossible.
Would our robot have the capacity to be happy? Or sad? Would it have ambition? These are a complex mixture of various elements of the brain that build up to form "I" and no computer could recreate that. For example, I want food. How could we program a computer to want food. We could program 'acquire food,' but there's a huge difference between a mere command and the word "want." How can we program want. It's something that doesn't exist from the outside-in.
That means the programming to seek out food must come from a basic level. This basic level is apparent with us, our ancient microbial ancestors that had some basic chemical process that made them travel towards food survived. This resulted in a more complex mechanism of having the means to travel towards food, still a basic chemical process, which went on and on. We "want" because it's a fundamental part of what we as human animals are. How are we to solve that? We would need to first solve the problem of programming fundamental parts. And, again, those parts cannot be created from the outside in.
As such, I don't think AI is an endeavor that even needs to be made. In fact, I think a quest for true AI is a waste. The Turing Test had it right. We should focus on AI only so far as it relates to interactions with humans for a purpose. For example, Rosie the Maid from The Jetsons doesn't need to be a cantankerous robot from the depths of Brooklyn, she only needs to be able to interact with her environment and achieve a simple set of tasks. That's achievable, hell, that's downright easy. But true AI, forget it.
For that kind of AI, we need to be able to construct self-awareness from the outside in. Basically, we study the brain, we study how it works, learn the language of thought as we would a computer language, and than construct a program that is self-aware. I really doubt this is possible. because of the way in which intelligence arises. It doesn't arise from the outside in. A creature never came along (Unless you're a Catholic) and just imbued something with self-awareness. The revelation of self awareness comes from within, when the entity evolves into it.
I don't think there is any "language" of self awareness since no language is required. As neuroscience shows time and time again, the human brain can function with different layouts, sizes, and structure to identical ends. It's because each computer programs itself as it grows. It's because of that that I doubt we will ever be able to program something ourself.
What we could do is create a network of computers with similar basic programming to the primordial brain, and just watch and hope that some sort of consciousness arises within the system. I think this is a possibility. Granted, we don't have the technology to do it, but it's a possibility. Unfortunately, that doesn't fulfill the God desire: the desire to create actively, not passively. It's not AI per se.
In fact, it may be an artificial environment, a computer one, but the development of the intelligence would be very natural. One that follows the rules of the system laid down but otherwise meanders about its own path. And, again, in the end we would be unable to read the code that was created and "see" the intelligence since the code is only readable by the code. Any pattern works. Any code. Any compiler, as it were. It doesn't matter. The syntax and whatnot come from itself. One aware orgamism can be 99% different from another aware organism. The basic structure of our brains may be the same, but for all we know, every brain, in every person EVERWHERE "speaks" another programming language, since the logic to the whole mess only needs to make sense to itself.
I think the movie Bicentennial Man is a good representative of how AI may arise. We may create a system that is intended to be as close to human as possible, and during its development, that one-in-a-million developmental direction takes place and a robot becomes self-aware. But what kind of awareness would it be? How do we program a system of neurochemical awards? Could we create a robot that can get addicted to drugs? Our own consciousness stems from these basic concepts. We cannot create the gross elements of our behavior and build down. It is impossible.
Would our robot have the capacity to be happy? Or sad? Would it have ambition? These are a complex mixture of various elements of the brain that build up to form "I" and no computer could recreate that. For example, I want food. How could we program a computer to want food. We could program 'acquire food,' but there's a huge difference between a mere command and the word "want." How can we program want. It's something that doesn't exist from the outside-in.
That means the programming to seek out food must come from a basic level. This basic level is apparent with us, our ancient microbial ancestors that had some basic chemical process that made them travel towards food survived. This resulted in a more complex mechanism of having the means to travel towards food, still a basic chemical process, which went on and on. We "want" because it's a fundamental part of what we as human animals are. How are we to solve that? We would need to first solve the problem of programming fundamental parts. And, again, those parts cannot be created from the outside in.
As such, I don't think AI is an endeavor that even needs to be made. In fact, I think a quest for true AI is a waste. The Turing Test had it right. We should focus on AI only so far as it relates to interactions with humans for a purpose. For example, Rosie the Maid from The Jetsons doesn't need to be a cantankerous robot from the depths of Brooklyn, she only needs to be able to interact with her environment and achieve a simple set of tasks. That's achievable, hell, that's downright easy. But true AI, forget it.
Subscribe to:
Posts (Atom)