Eventually, at least the learning component will converge; returns will diminish for feeding it more data. This is particularly true given the independence assumption inherent in their classifier (but would also hold on stronger learners). I suspect that this will happen to the reader component as well. If it were as simple as applying Naive Bayes to classify on a corpus of text connected to a knowledge base (which is probably just a set of posteriors left from previous training sessions), Cyc would have already passed the Turing test.
Machine learning is the logical place to take a combined knowledge of programming and statistics. It's a much rarer skill and commands a much higher salary, plus you're doing the closest thing we currently have to predicting the future for a living - and you generally still get to code plenty.
In other words, statistical knowledge can be a significant career advantage in addition to enhancing development and debugging.
If this were completely true, science wouldn't exist in the first place. Yes, we do have animal natures to deal with. But as rational beings, it is within our power to override these instincts in the name of abstract principles which we consider nobler. In a way, it's what makes us human.
Many people may diminish my hope for humanity, many scientists among them (academia's very ugly when you get a close enough look at it), but a handful are enough to restore it.
The way I see it, I'm going to be shown ads either way, so they may as well be relevant to my interests. That said, I don't see any particularly compelling reason to switch to Google's DNS, and beyond knowing every DNS lookup that users make, I can't see any particularly compelling reason why Google rolled this out in the first place. Fewer cache misses are nice, but that isn't really game-changing.
Useful, but real-world tagging is much more specific than "person", "animal", or "inanimate". The number of classes required in the classification task is thus far greater and one would expect the accuracy to be proportionally lower.
OTOH, it could be a great preprocessing step for further manual analysis, or a step in a hierarchical clustering algorithm. Or maybe 3 classes suffice for certain specific situations.
Then on top of that, the world of scholarship doesn't stand still - new things about historical topics are routinely discovered.
Remember, you are not allowed to post "original research". I for one would prefer that the Wiki contained the most cutting edge information, but I suppose they don't want something posted that may later turn out to be premature.
Though a bit uglier than usual, that type of behavior is fairly prevalent in the scientific community. Articles in "unpopular" topics have always tended to get sidelined (reviewers can reject papers simply on the premise that they're not on a topic a journal would wish to publish, for example), and it's easy to see how this can progress to choosing a side in a scientific debate.
Though a model more in-line with arXiv might mitigate this, I think it represents a fundamental flaw in the currently-used system of peer review: it's essentially a binary threshold. Either your paper is accepted and you have a voice in the scientific community, or it's rejected and you have none. Something along the lines of a Slashdot or Digg-style moderation may work much better: other researchers can mod you down all they wish and send it to the last page of a query, but they can't actually make your work disappear. And since the ranking is relative to other relevant papers, unpopular topics and positions would not be penalized relative to each other using such a system.
You could even generate confidence intervals for the rankings based on the number of reviews. Right now a decision on a paper is based on 2 or 3 reviews at most, and it would be difficult for a more open system not to exceed this.
The convenience of being able to navigate to a URL without having to type it is a really limited example. How about writing music with it? Being able to notate exactly what's playing in your head without needing to manually write a single note down? Weeks worth of work reduced to a few minutes! Or art: Can't draw? Just visualize!
Anything you can think about but can't actually do would be fair game.
Even with those sorts of apps, I still wouldn't get an implant unless my skull was being opened up for some other reason already. It's certainly not a fair tradeoff against something as simple as web browsing, as the summary suggests. I'm all for the braincaps. That's where BCI technology's headed anyway. And those have the distinct advantage of being removable as well...
This isn't really strong AI in the sense that you're thinking of it:
The latest feat, being presented at a supercomputing conference in Portland, Ore., doesn't mean the computer thinks like a cat, or that it is the progenitor of a race of robo-cats.
The simulation, which runs 100 times slower than an actual cat's brain, is more about watching how thoughts are formed in the brain and how the roughly 1 billion neurons and 10 trillion synapses in a cat's brain work together.
To me, this translates into "we've made a big unspecialized neural network and we're watching the weights update as we try to classify corporate logos with it". While building something on this scale is quite a feat, this is not really modeling a cat's cortex... unless you happen to be including specialized structures and modeling those parts of the brain differently. Does this thing have a hippocampus, for instance?
I believe that the ultimate test of an AI system is functional: can it solve mental challenges that cats can solve (on its own, without being instructed in them in advance)? If so, it's at least as intelligent as a cat. If not, it isn't.
This is probably why it's being presented at a supercomputing conference and not at something like AAAI.
"No matter how far into the future we may look, nuclear fusion as an energy source is even less probable than large-scale breeder reactors, for the accumulated knowledge on this subject is already sufficient to say that commercial fusion power will never become a reality."
I have a feeling this will go up there with "it's impossible to build a heavier-than-air flying machine" and "there's a world market for about 6 computers".
That's New York City (and for good reason). In NJ we measure everything by exits on the highways:)
But getting back to the original discussion, on a galactic scale, distance and time are scaled versions of the same thing anyway, since everything is traveling at a fixed maximum speed. Something 1 ly away will have a minimum latency of 1 year (2 for the round trip) unless we figure out how to send information FTL. And when you're dealing with distances of thousands of ly, you really wouldn't care about a few seconds here or there due to other overhead.
Sort of. There is a meaningful statistical correlation behind subscores in IQ tests, and despite the existence and appeal of the multiple intelligences model, the data still tends to support the general intelligence factor model.
And while it's true that no one is good (or even experienced) at literally everything, people who learn one subject earlier and/or faster have far more time to devote to learning additional proficiencies later on. This is also disregarding the role of overlapping thoughts and ideas, which may greatly hasten learning a new subject. In other words, the more you know, the easier it becomes to know even more (as an added bonus, broad knowledge has also been found to aid creativity, presumably because you have more analogies to draw upon and apply in different places). And the faster you learn, the quicker you'll know a lot (assuming you have the dedication). One of my ideas has served as a magnet for such polymathic types - I've easily spoken to hundreds by now - and the vast majority of them are in fact gifted.
My impression from all of this is that you could have particular proficiency in a single subject that doesn't reflect on an IQ test - but general (or at least multifocal) proficiency exists and does seem to associate much more strongly with IQ.
None of this has anything to do with making good choices in life, of course. That's not usually a question of advanced reasoning ability (though it may have some relation with executive function). Plenty of people across the IQ spectrum consistently make bad choices. Like going to grad school:)
That's just the thing - we might not see it until it's too late. One strong gamma ray burst could just wipe us out and we wouldn't even know it was coming until the moment it arrived. It's not about responding to an imminent threat, but being prepared to recover from one that may occur at a later time. Right now all of our data is on one hard drive. We need to make a backup.
It's really illegal for us. Just not for them. This is precisely contrary to the example that the government should be setting. Those who make and enforce the laws should also be their exemplars, not their exceptions.
Slashdot Mirror
Eventually, at least the learning component will converge; returns will diminish for feeding it more data. This is particularly true given the independence assumption inherent in their classifier (but would also hold on stronger learners). I suspect that this will happen to the reader component as well. If it were as simple as applying Naive Bayes to classify on a corpus of text connected to a knowledge base (which is probably just a set of posteriors left from previous training sessions), Cyc would have already passed the Turing test.
But not too much!
Machine learning is the logical place to take a combined knowledge of programming and statistics. It's a much rarer skill and commands a much higher salary, plus you're doing the closest thing we currently have to predicting the future for a living - and you generally still get to code plenty.
In other words, statistical knowledge can be a significant career advantage in addition to enhancing development and debugging.
The sheer scale of all of that data would swamp them.
Doing precisely this is one of the more distasteful parts of most scientists' jobs.
Absolutely not! You'll be stuck in an office, writing code :)
If this were completely true, science wouldn't exist in the first place. Yes, we do have animal natures to deal with. But as rational beings, it is within our power to override these instincts in the name of abstract principles which we consider nobler. In a way, it's what makes us human.
Many people may diminish my hope for humanity, many scientists among them (academia's very ugly when you get a close enough look at it), but a handful are enough to restore it.
My mind has been mystified.
The way I see it, I'm going to be shown ads either way, so they may as well be relevant to my interests. That said, I don't see any particularly compelling reason to switch to Google's DNS, and beyond knowing every DNS lookup that users make, I can't see any particularly compelling reason why Google rolled this out in the first place. Fewer cache misses are nice, but that isn't really game-changing.
3: Open up a branch office of your multinational corporation and employ lots of locals for nearly nothing by the standards of the developed world?
Useful, but real-world tagging is much more specific than "person", "animal", or "inanimate". The number of classes required in the classification task is thus far greater and one would expect the accuracy to be proportionally lower. OTOH, it could be a great preprocessing step for further manual analysis, or a step in a hierarchical clustering algorithm. Or maybe 3 classes suffice for certain specific situations.
Quickly, write a Wikipedia article about it before it's too late!
Remember, you are not allowed to post "original research". I for one would prefer that the Wiki contained the most cutting edge information, but I suppose they don't want something posted that may later turn out to be premature.
Though a bit uglier than usual, that type of behavior is fairly prevalent in the scientific community. Articles in "unpopular" topics have always tended to get sidelined (reviewers can reject papers simply on the premise that they're not on a topic a journal would wish to publish, for example), and it's easy to see how this can progress to choosing a side in a scientific debate.
Though a model more in-line with arXiv might mitigate this, I think it represents a fundamental flaw in the currently-used system of peer review: it's essentially a binary threshold. Either your paper is accepted and you have a voice in the scientific community, or it's rejected and you have none. Something along the lines of a Slashdot or Digg-style moderation may work much better: other researchers can mod you down all they wish and send it to the last page of a query, but they can't actually make your work disappear. And since the ranking is relative to other relevant papers, unpopular topics and positions would not be penalized relative to each other using such a system.
You could even generate confidence intervals for the rankings based on the number of reviews. Right now a decision on a paper is based on 2 or 3 reviews at most, and it would be difficult for a more open system not to exceed this.
In Soviet Russia, brain implants will control YOU.
The convenience of being able to navigate to a URL without having to type it is a really limited example. How about writing music with it? Being able to notate exactly what's playing in your head without needing to manually write a single note down? Weeks worth of work reduced to a few minutes! Or art: Can't draw? Just visualize!
Anything you can think about but can't actually do would be fair game.
Even with those sorts of apps, I still wouldn't get an implant unless my skull was being opened up for some other reason already. It's certainly not a fair tradeoff against something as simple as web browsing, as the summary suggests. I'm all for the braincaps. That's where BCI technology's headed anyway. And those have the distinct advantage of being removable as well...
To me, this translates into "we've made a big unspecialized neural network and we're watching the weights update as we try to classify corporate logos with it". While building something on this scale is quite a feat, this is not really modeling a cat's cortex... unless you happen to be including specialized structures and modeling those parts of the brain differently. Does this thing have a hippocampus, for instance?
I believe that the ultimate test of an AI system is functional: can it solve mental challenges that cats can solve (on its own, without being instructed in them in advance)? If so, it's at least as intelligent as a cat. If not, it isn't.
This is probably why it's being presented at a supercomputing conference and not at something like AAAI.
Sounds like Windows ME.
"No matter how far into the future we may look, nuclear fusion as an energy source is even less probable than large-scale breeder reactors, for the accumulated knowledge on this subject is already sufficient to say that commercial fusion power will never become a reality."
I have a feeling this will go up there with "it's impossible to build a heavier-than-air flying machine" and "there's a world market for about 6 computers".
That's New York City (and for good reason). In NJ we measure everything by exits on the highways :)
But getting back to the original discussion, on a galactic scale, distance and time are scaled versions of the same thing anyway, since everything is traveling at a fixed maximum speed. Something 1 ly away will have a minimum latency of 1 year (2 for the round trip) unless we figure out how to send information FTL. And when you're dealing with distances of thousands of ly, you really wouldn't care about a few seconds here or there due to other overhead.
Why would you want to talk to Marvin?
Sort of. There is a meaningful statistical correlation behind subscores in IQ tests, and despite the existence and appeal of the multiple intelligences model, the data still tends to support the general intelligence factor model.
And while it's true that no one is good (or even experienced) at literally everything, people who learn one subject earlier and/or faster have far more time to devote to learning additional proficiencies later on. This is also disregarding the role of overlapping thoughts and ideas, which may greatly hasten learning a new subject. In other words, the more you know, the easier it becomes to know even more (as an added bonus, broad knowledge has also been found to aid creativity, presumably because you have more analogies to draw upon and apply in different places). And the faster you learn, the quicker you'll know a lot (assuming you have the dedication). One of my ideas has served as a magnet for such polymathic types - I've easily spoken to hundreds by now - and the vast majority of them are in fact gifted.
My impression from all of this is that you could have particular proficiency in a single subject that doesn't reflect on an IQ test - but general (or at least multifocal) proficiency exists and does seem to associate much more strongly with IQ.
None of this has anything to do with making good choices in life, of course. That's not usually a question of advanced reasoning ability (though it may have some relation with executive function). Plenty of people across the IQ spectrum consistently make bad choices. Like going to grad school :)
I've always thought this phenomenon was at the heart of the Flynn effect.
That's just the thing - we might not see it until it's too late. One strong gamma ray burst could just wipe us out and we wouldn't even know it was coming until the moment it arrived. It's not about responding to an imminent threat, but being prepared to recover from one that may occur at a later time. Right now all of our data is on one hard drive. We need to make a backup.
It's really illegal for us. Just not for them. This is precisely contrary to the example that the government should be setting. Those who make and enforce the laws should also be their exemplars, not their exceptions.