GAs and simulated annealing are hill-climbing searches. The fact that they might be run from multiple starting points is neither here nor there. Their parallelism and stochasticity doesn't detract from the fact that they simply evaluate the objective function at certain points and then tend to move in the direction of best performance, i.e. they hill-climb.
Non-hill-climbing optimisation methods (such as, say, treating the evaluations as samples from a Gaussian process and using regression to estimate a distribution model over the objective function from which new points are sampled) are fundamentally different.
I'm still not sure what you mean when you say neural nets are a "broad-front search" as they are not a search algorithm but a representation.
I'll nitpick a bit: genetic algorithms are hill-climbing searches. Also, I'm not sure what you mean by "broad-front search algorithm" in terms of neural networks as backprop is gradient ascent (thus hill-climbing). Unless you meant something other than backprop?
I guess you meant to say gradient ascent rather than hill-climbing but your argument is quite right: the performance of any search algorithm ultimately depends on the nature of the objective surface (something the GA people seem to casually ignore).
It's interesting how GA seem to be a very contagious meme in terms of popular "AI" decades past any serious research into them. Guess it's the biological reference that gets the geeks' juices flowing...
"... I'm just claiming that science isn't any better than carpentry..."
The difference between science and carpentry is that science is about explaining the rules governing the universe and carpentry is about cutting up a plant and sticking its pieces together.
That the former should have no higher interest to laypeople than the latter is an exceptionally ugly thought.
Your observation is that individuals derive utility from "compassionate" behaviour just as concretely as they derive utility from watching a movie, and that the amount of utility they derive from such actions differs between individuals (and, you say, races).
How does this disprove the concept of a rational (utility maximising) individual?
"Communication between the PDA and the onboard CerfBoard 405EP is achieved through cellular communications... An 802.11b connection is used for near field communications and software updates."
OK, I'll make this simple so that the logic is completely transparent:
In Java, program structure is defined by the programmer inserting appropriate "{"s and "}"s. Indentation can be implicitly determined by the editor based on this structure.
In Python, program structure is defined by the programmer inserting appropriate indentation. (If you wanted to), "{"s and "}"s could be implicitly determined by the editor based on this structure.
Your argument is that, with Python, an editor cannot magically retab the document as it can with Java. An equivalent statement would be to deride Java based on the fact that an editor cannot magically insert (redundant) "{"'s and "}"'s based on the indentation as it could in Python.
You ask "where do you stop tabbing?" after inserting an if statement. The answer, of course, is that in all languages, creating a new block is changing the structure. In Java, you do this by inserting your braces. In Python, you do this by changing the indentation. Your question is really: "how does an editor determine this implicit presentation based on the structure given by the programmer?". In Python they are the same thing so the answer is trivial.
To answer your question: I've exclusively used emacs as my Python editor for several years and have never touched any emacs settings w.r.t. Python.
Your sole point is that a "miss tabbed document" (sic) breaks the code. Could you explain where this altered tabbing comes from? Would you similarly deride a language for not being impervious to random removal of characters?
I've written 10's of thousands of lines of Python code and have never had (that I can recall) a single problem with indentation, blocking or scope. I must admit, I had exactly the same knee-jerk reaction to Python as you before actually trying it -- it took all of one day of using it to wash away any doubts. This is after years of coding C++ and Java, BTW.
Side note: if you're terminating lines with a semicolon in Python then I doubt you "use it alot" (sic).
Wow. This sounds awesome. You could run all of your applications on this personal server so that the device with your user interface can be really lightweight... I mean, do we really need a huge hard-drive in every terminal? It just makes sense. You could even let others use your server to run their applications. I can see it now. You'd probably want to run Java on it, and then you could upgrade the low-speed wireless link so that you could connect multiple of these servers together via high-speed wired links. It's almost like the network IS the computer.
Wrong: if you're announcing something that doesn't exist as being the first of its kind, then it means there aren't any others, thus implying that it must be the largest.
Personally, I'd go further and say that statistical methods are the only way of making informed judgement. In my view statistics is the problem of reasoning under explicit assumptions in the presence of uncertainty. Surely, that's exactly what AI is trying to achieve? If there's no uncertainty then you're just playing chess. If there's no assumptions then you're just inventing heuristics (read "hacks").
I think the difference between the two camps is that one (the computer scientists) focuses on the computational intractability of decision making, while the other (the mathematicians/statisticians) focuses on the uncertainty inherent in decision making. I do believe they've got a lot to offer each other, in particular statistics is finally seriously taking into account tractability.
My personal view is that all the careful positioning of the smoke and mirrors that the traditional AI'ers is doing isn't going to help understanding or even emulating intelligence in the long run.
I'd be happy if the term "AI" was tossed overboard and we all sailed off into the sunset of enlightenment under the power of Bayesian statistics and mathematical maturity...
Are you truly impressed by a Chess program or expert system? Honestly, in your heart, do you think "Wow-- that computer has brute-force searched 6 moves ahead to compute every possible board position and then selected the one that minimises the maximum possible captures the opponent can make -- that's really intelligent"? Do you gush over the fact that an expert system has queried its database of 50,000 rules pre-programmed by domain experts to diagnose a medical condition?
I'm sorry, but to me that isn't AI. Call it what you will (I call it software engineering) but it's so far removed from the real adaptive intelligence we observe everyday in the billions of organisms we share this planet with that it's not funny. You can't just look at the outputs to classify intelligence (artificial or not). You have to look at the inputs -- and in all these cases, the assumptions of these systems are hand-coded by real intelligences (humans).
You can wave your multiple definitions of AI around but at the end of the day we all know that this isn't anything special...
Artificial intelligence does not mean "simulating intelligence artificially."
If it did, then an actor with an IQ of 50 playing the role of Albert Einstein in a movie has intelligence because he's simulating intelligence. If we then replace the actor with a mechanical fish which sings the lines, this would now be artificial intelligence as the fish is "simulating intelligence artificially."
I'm a masters student in computer science at Victoria University in Wellington, NZ and went to this seminar. I'm as big a fan of LOTR as the next guy. However, I have this pet gripe. I agree that LOTR is an impressive feat of computer graphics but I'm annoyed by this talk of how MASSIVE is "AI on steroids."
There is NO AI in MASSIVE. Surely if AI means anything, it means the ability to optimise behaviour, or learn from data, or at least demonstrate adaptation of some sort. There is no adaptation in MASSIVE. Each agent is consulting a list of rules of what to do in a given situation and then executing the specified motion-captured animation. Not only is the motion not generated by the agent, but the rules are just hand-coded by humans. They're not even evolving these "brains."
The reason that it looks impressive is because instead of using identical, dumb, particle-like agents the agents have pre-programmed decision trees that generate their actions. Great work -- good programming job, but nothing that any hacker couldn't come up with. Show me a single agent in MASSIVE learning to walk or lifting a weapon or producing any movement that wasn't pre-scripted and I'll be impressed.
In my opinion the cool thing here is the remarkably ability of complex systems to generate interesting global phenomena from locally interacting agents.
Can someone who knows better please prove me wrong? I'd love to believe this was something more than a trumped up screensaver...
Physics does constrain the speed of computation. Read some David Deutsch (such as The Fabric of Reality) for the low-down.
Re:Must publically held mean no morals?
on
Google vs. Evil
·
· Score: 1
You seem intelligent enough to agree that, if the market is privy to all of the information available to management when they make a decision in the best interests of the company, then this short/long-term bias doesn't exist.
Now, any market inefficiency is due to lack of public information. Why would a management with performance incentives be circumspect about divulging information that would be interpreted by market participants as being in the best interests of the company in the long-term?
As for the "non sequitor" (sic), I fail to see why it is preferable for resources to be allocated by the management of companies rather than elected representatives when it comes to projects purely of social/environmental concern. If managers care so much about being socially concious, why don't they invest their own money and not that of the shareholders?
Re:Must publically held mean no morals?
on
Google vs. Evil
·
· Score: 1
You say that tying the remuneration of management to share price is somehow a bad thing. I don't understand. The share market determines current price based on estimates of the value of the future performance of a company and its decisions. Unless the management withholds information from the market (which they wouldn't do if they truly believe that what they're doing is in the company's long-term interests) then, in your example, the CEO and board is incented to retain the new employees.
As for democracy, you obviously believe that peoples' opinions should be weighted by the size of their wallets...
Slashdot Mirror
I always thought it stemmed from p-ness envy.
GAs and simulated annealing are hill-climbing searches. The fact that they might be run from multiple starting points is neither here nor there. Their parallelism and stochasticity doesn't detract from the fact that they simply evaluate the objective function at certain points and then tend to move in the direction of best performance, i.e. they hill-climb.
Non-hill-climbing optimisation methods (such as, say, treating the evaluations as samples from a Gaussian process and using regression to estimate a distribution model over the objective function from which new points are sampled) are fundamentally different.
I'm still not sure what you mean when you say neural nets are a "broad-front search" as they are not a search algorithm but a representation.
I'll nitpick a bit: genetic algorithms are hill-climbing searches. Also, I'm not sure what you mean by "broad-front search algorithm" in terms of neural networks as backprop is gradient ascent (thus hill-climbing). Unless you meant something other than backprop?
I guess you meant to say gradient ascent rather than hill-climbing but your argument is quite right: the performance of any search algorithm ultimately depends on the nature of the objective surface (something the GA people seem to casually ignore).
It's interesting how GA seem to be a very contagious meme in terms of popular "AI" decades past any serious research into them. Guess it's the biological reference that gets the geeks' juices flowing...
"... I'm just claiming that science isn't any better than carpentry ..."
The difference between science and carpentry is that science is about explaining the rules governing the universe and carpentry is about cutting up a plant and sticking its pieces together.
That the former should have no higher interest to laypeople than the latter is an exceptionally ugly thought.
>> My ISP (internet express in regional NSW, australia) ...
> Which ISP, pray tell?
Not too quick on the uptake, are you?
"Just like every mammal has breasts."
You haven't met my girlfriend, have you?
Your observation is that individuals derive utility from "compassionate" behaviour just as concretely as they derive utility from watching a movie, and that the amount of utility they derive from such actions differs between individuals (and, you say, races).
How does this disprove the concept of a rational (utility maximising) individual?
Who said it was?
RTFA.
"Communication between the PDA and the onboard CerfBoard 405EP is achieved through cellular communications... An 802.11b connection is used for near field communications and software updates."
Don't fret. It's not an English invention but actually another (along with powered flight) New Zealand invention.
You're grammar sucks.
my first attempt at a first post...
OK, I'll make this simple so that the logic is completely transparent:
In Java, program structure is defined by the programmer inserting appropriate "{"s and "}"s. Indentation can be implicitly determined by the editor based on this structure.
In Python, program structure is defined by the programmer inserting appropriate indentation. (If you wanted to), "{"s and "}"s could be implicitly determined by the editor based on this structure.
Your argument is that, with Python, an editor cannot magically retab the document as it can with Java. An equivalent statement would be to deride Java based on the fact that an editor cannot magically insert (redundant) "{"'s and "}"'s based on the indentation as it could in Python.
You ask "where do you stop tabbing?" after inserting an if statement. The answer, of course, is that in all languages, creating a new block is changing the structure. In Java, you do this by inserting your braces. In Python, you do this by changing the indentation. Your question is really: "how does an editor determine this implicit presentation based on the structure given by the programmer?". In Python they are the same thing so the answer is trivial.
To answer your question: I've exclusively used emacs as my Python editor for several years and have never touched any emacs settings w.r.t. Python.
Your sole point is that a "miss tabbed document" (sic) breaks the code. Could you explain where this altered tabbing comes from? Would you similarly deride a language for not being impervious to random removal of characters?
I've written 10's of thousands of lines of Python code and have never had (that I can recall) a single problem with indentation, blocking or scope. I must admit, I had exactly the same knee-jerk reaction to Python as you before actually trying it -- it took all of one day of using it to wash away any doubts. This is after years of coding C++ and Java, BTW.
Side note: if you're terminating lines with a semicolon in Python then I doubt you "use it alot" (sic).
Where?
Wow. This sounds awesome. You could run all of your applications on this personal server so that the device with your user interface can be really lightweight... I mean, do we really need a huge hard-drive in every terminal? It just makes sense. You could even let others use your server to run their applications. I can see it now. You'd probably want to run Java on it, and then you could upgrade the low-speed wireless link so that you could connect multiple of these servers together via high-speed wired links. It's almost like the network IS the computer.
Why didn't someone think of this before?
Wrong: if you're announcing something that doesn't exist as being the first of its kind, then it means there aren't any others, thus implying that it must be the largest.
Bingo. Thanks for that, Paul.
Personally, I'd go further and say that statistical methods are the only way of making informed judgement. In my view statistics is the problem of reasoning under explicit assumptions in the presence of uncertainty. Surely, that's exactly what AI is trying to achieve? If there's no uncertainty then you're just playing chess. If there's no assumptions then you're just inventing heuristics (read "hacks").
I think the difference between the two camps is that one (the computer scientists) focuses on the computational intractability of decision making, while the other (the mathematicians/statisticians) focuses on the uncertainty inherent in decision making. I do believe they've got a lot to offer each other, in particular statistics is finally seriously taking into account tractability.
My personal view is that all the careful positioning of the smoke and mirrors that the traditional AI'ers is doing isn't going to help understanding or even emulating intelligence in the long run.
I'd be happy if the term "AI" was tossed overboard and we all sailed off into the sunset of enlightenment under the power of Bayesian statistics and mathematical maturity...
Are you truly impressed by a Chess program or expert system? Honestly, in your heart, do you think " Wow -- that computer has brute-force searched 6 moves ahead to compute every possible board position and then selected the one that minimises the maximum possible captures the opponent can make -- that's really intelligent"? Do you gush over the fact that an expert system has queried its database of 50,000 rules pre-programmed by domain experts to diagnose a medical condition?
I'm sorry, but to me that isn't AI. Call it what you will (I call it software engineering) but it's so far removed from the real adaptive intelligence we observe everyday in the billions of organisms we share this planet with that it's not funny. You can't just look at the outputs to classify intelligence (artificial or not). You have to look at the inputs -- and in all these cases, the assumptions of these systems are hand-coded by real intelligences (humans).
You can wave your multiple definitions of AI around but at the end of the day we all know that this isn't anything special...
Artificial intelligence does not mean "simulating intelligence artificially."
If it did, then an actor with an IQ of 50 playing the role of Albert Einstein in a movie has intelligence because he's simulating intelligence. If we then replace the actor with a mechanical fish which sings the lines, this would now be artificial intelligence as the fish is "simulating intelligence artificially."
I'm a masters student in computer science at Victoria University in Wellington, NZ and went to this seminar. I'm as big a fan of LOTR as the next guy. However, I have this pet gripe. I agree that LOTR is an impressive feat of computer graphics but I'm annoyed by this talk of how MASSIVE is "AI on steroids."
There is NO AI in MASSIVE. Surely if AI means anything, it means the ability to optimise behaviour, or learn from data, or at least demonstrate adaptation of some sort. There is no adaptation in MASSIVE. Each agent is consulting a list of rules of what to do in a given situation and then executing the specified motion-captured animation. Not only is the motion not generated by the agent, but the rules are just hand-coded by humans. They're not even evolving these "brains."
The reason that it looks impressive is because instead of using identical, dumb, particle-like agents the agents have pre-programmed decision trees that generate their actions. Great work -- good programming job, but nothing that any hacker couldn't come up with. Show me a single agent in MASSIVE learning to walk or lifting a weapon or producing any movement that wasn't pre-scripted and I'll be impressed.
In my opinion the cool thing here is the remarkably ability of complex systems to generate interesting global phenomena from locally interacting agents.
Can someone who knows better please prove me wrong? I'd love to believe this was something more than a trumped up screensaver...
1. Largest population => biggest market.
2. Biggest market => strongest economy.
I don't see how either of those are valid -- please explain.
Oh, and could you make special mention of globalisation and present day China in your answer.
Physics does constrain the speed of computation. Read some David Deutsch (such as The Fabric of Reality) for the low-down.
You seem intelligent enough to agree that, if the market is privy to all of the information available to management when they make a decision in the best interests of the company, then this short/long-term bias doesn't exist.
Now, any market inefficiency is due to lack of public information. Why would a management with performance incentives be circumspect about divulging information that would be interpreted by market participants as being in the best interests of the company in the long-term?
As for the "non sequitor" (sic), I fail to see why it is preferable for resources to be allocated by the management of companies rather than elected representatives when it comes to projects purely of social/environmental concern. If managers care so much about being socially concious, why don't they invest their own money and not that of the shareholders?
You say that tying the remuneration of management to share price is somehow a bad thing. I don't understand. The share market determines current price based on estimates of the value of the future performance of a company and its decisions. Unless the management withholds information from the market (which they wouldn't do if they truly believe that what they're doing is in the company's long-term interests) then, in your example, the CEO and board is incented to retain the new employees.
As for democracy, you obviously believe that peoples' opinions should be weighted by the size of their wallets...