I came across the page above while I was thinking about implementing an evolutionary neural network on Mindstorms myself. Basically, it's a really obvious idea, and simple evolutionary algorithms are so easy to implement. But then, the sensing capabilities of mindstorms kits are so limited, that there's really not much potential for the software to evolve into something really interesting. To do something really interesting with so simple sensors, you'll have to work on the morphology, which is notoriously tricky to evolvel
Very true. As a Swede living in the UK for the moment I have a hard time convincing lightheaded idealists and ordinary people alike that Sweden is very far from the paradise on earth they all too often envision it to be.
Even the almighty Swedish labour unions nowadays admit that the true unemployment figure is around 20%, far from the 5% the official statistics would have it at. Very few of my highly qualified friends back home have been able to find any sort of job at all upon leaving university.
Crime statistics do not make for uplifting reading either, and ethnic tensions are on the rise (partly because of the astronomic unemployment figures among immigrant - there is a part of Malmö where 90% do not have a (legal) job). Looking at those who do work, the disposable income per capita compares unfavorably to almost any other western nation.
This is not because of the current macroeconomic plunge. In fact, Swedish export industries are currently doing well. Yet, unemployment and related problems are, if anything, on the rise.
These facts and figures are currently hotly debated in domestic media, though opinions on what to do about the mess obviously differ sharply. In my view the probable root cause is a combination of the worlds highest taxes, the worlds most powerful labour unions and the worlds probably most generous sick leave benefits.
I'm not sure software is neccesarily so important when compared to the hardware of these vehicles. I'm currently TA'ing a course in bio-inspired robotics, where students have to build robots to solve a variety of tasks. It turns out most tasks are solvable with the same few lines of code, conceptually similar to the wiring of the first few Braitenberg vehicles, only the robots are slighlty modified and sensors repositioned.
Hardware, especially sensors and their positioning, matter far more than people think.
The system you are thinking of would of course not be a hindrance to anyone wishing to publis his research, if it was implemented. But it still seems like plan economy in at least one important respect: who decides which journals are being available for free? All "academic" journals? I think the borderlines here are very fuzzy (do, for exampe, the IEE Review, New Scientist, and THES all count?) and it is better left to those active in the field to make those decisions in a decenteralized fashion. And it would still be a bit unfair to those scientists who did nothing to deserve that their country refuse to pay the bills.
Not to mention that the system would be quite unnecessary. As the examples of JAIR and JMLR show, it is perfectly sensible to run an academic journal online-only with more or less no operating expenses at all.
As for the unemployed genius in Zimbabwe, s/he probably reads papers he gets directly from researchers' homepages. That's how I get the absolute majority of the papers I read. I am of the opinion that a researcher that doesn't make his writings available on his webpage doesn't care very much about being referenced.
One is the cost per article of publishing research. This would probably be the same regardless of where you are coming from. But a researcher costs much less to maintain (salary, office space etc.) in a poor country than in a rich country - and this must be so in order for the poor countries to have any competitiveness. So the net result is that the number of articles any one researcher would be allowed to publish would be smaller in a poor country than in a rich country. Which is fundamentally at odds with the whole idea of peer-review systems, that whether an article is published or not is dependent on the quality of the research and the presentation of it, and nothing else.
I am here assuming that governments are introducing some sort of systems for endorsing articles they are willing to pay for. A justifiable assumption, as it is unlikely that many governments are willing to pay an unpredictable amount of money for something they have no control over.
Which leads us to other problems. Such as what to do if you are not employed by a university, but unemployed or employed by a private corporation. (I published an article while I was unemployed and working from home, and this was essential in getting me my PhD studentship.) Or if your employer doesn't like the research you do. Or if your country doesn't like the research you do (Iran might be unwilling to pay for publishing research in evolutionary biology or gender studies, and the US might be unwilling to pay for applied marxism or similar.) Or if you are collaborating with the "wrong" people in th wrong countries.
And what if your country refuses to take part at all in international systems like this?
In short, how would this system work if you're an unemployed genius working in Zimbabwe?
In my views, the proposed system has all the problems of a plan economic system, and not even a semblance of fairness.
In the field of AI we have at least two highly respected journals which do not have paper editions (even though libraries can buy bound collections of papers on a yearly basis) and which make their content available for free to everyone: Journal of Artificial Intelligence Research and Journal of Machine Learning Research
This works because an academic journal does not really have any expenses for peer-review. Academics review for free as part of their job - it gives status to review for a prestigious journal. If you don't have any costs for editing and printing a paper edition, you suddenly have almost no operating expenses at all. Cost of bandwith is negligible. A typical research paper in pdf format is a 100k download, so any one of us could operate one of those servers from our home. Furthermore, the cost of bandwith is continually decreasing.
In sum, I don't understand what the IEEE is whining about. Let those who want a journal on paper pay for the paper, and let the rest of us have it for free!
Another problem is surely that some countries are richer than others. A US scientist would be able to publish so much more than an Indian scientist of the same calibre.
And yet another is that of the country's government deciding which of its scientists gets to publish.
It could have been more than a thin wrapper for specific Win32 calls. It could have been an API implementable on more than one platform, for example, with.NET calls supervenient on the native calls on more than one OS.
By the way, the word you are looking for is probably "modular" rather than "functional". Functional programming is what you do with LISP.
Rodney Brooks (in his pop-sci book, which seems to have come out under several different names, amongst them "Flesh and machines") has an interesting answer to your worry: we will integrate our bodies with the technology. In that way, AI will never take over the world, because we will be one with the technology, and a cyborg incorporating both flesh and silicon will always have an advantage over a pure silicon system.
I pretty much agree.
But let us hypothetically grant you the (huge) benefit of the doubt. How would we do to discontinue or at least severely restrict the research that would potentially be dangerous, and only that research? The problem is that "AI" is a very vague umbrella term nowadays. Back in the fifties, it was actually possible for someone to be an expert in AI - nowadays, most of the research done in computer science departments (and then quite some in other departments, such as electrical engineering) is stuff that has at some point been called artificial intelligence. In fact, things as diverse as object-oriented programming, railway scheduling systems and much modern image processing has sprung out of research that was originally aimed at replicating the mechanisms of the mind. I like telling my mother that I "work in AI", but to anyone who knows anything about the field that is as informative as saying that I sit in front of computers all day.
So, would you be ready to pull the plug on most CS research? Or do you have any specific suggestions?
From what I can understand of the papers downloadable from the guys website hes primarily working in physical (humanoid) robotics and evolutionary computation. Getting from there to evolving aspects of robot behaviour on the physical robots is not very far fetched - if you then want to call these behaviour-aspects "emotions" is entirely up to you...
However, similar things have been done before, by the eminent people at Brandeis:
http://www.demo.cs.brandeis.edu/pr/ee/
I would say it's on the wrong track. It seems to have at least one foot on the ground at all times, and thus displays a "stable" and not a "dynamic" gait. Also, notice how it seems to walk with its knees bent all the time.
For an in-depth look at more natural-like (and potentially more useful) approaches to walking, see http://www.cogs.susx.ac.uk/users/inmanh/HART2004_H arvey_etal.pdf
For some amazing movies of evolved dynamic walking, see http://www.droidlogic.com/
Slashdot Mirror
And I actually agree with Bill Gates. I don't know which one of us is dirtiest, but it seems like a good idea to take that shower anyway.
Several years ago as well:
n dex.html
http://www.kaffedepartementet.nu/LegoMindstorms/i
I came across the page above while I was thinking about implementing an evolutionary neural network on Mindstorms myself. Basically, it's a really obvious idea, and simple evolutionary algorithms are so easy to implement. But then, the sensing capabilities of mindstorms kits are so limited, that there's really not much potential for the software to evolve into something really interesting. To do something really interesting with so simple sensors, you'll have to work on the morphology, which is notoriously tricky to evolvel
Very true. As a Swede living in the UK for the moment I have a hard time convincing lightheaded idealists and ordinary people alike that Sweden is very far from the paradise on earth they all too often envision it to be.
Even the almighty Swedish labour unions nowadays admit that the true unemployment figure is around 20%, far from the 5% the official statistics would have it at. Very few of my highly qualified friends back home have been able to find any sort of job at all upon leaving university.
Crime statistics do not make for uplifting reading either, and ethnic tensions are on the rise (partly because of the astronomic unemployment figures among immigrant - there is a part of Malmö where 90% do not have a (legal) job). Looking at those who do work, the disposable income per capita compares unfavorably to almost any other western nation.
This is not because of the current macroeconomic plunge. In fact, Swedish export industries are currently doing well. Yet, unemployment and related problems are, if anything, on the rise.
These facts and figures are currently hotly debated in domestic media, though opinions on what to do about the mess obviously differ sharply. In my view the probable root cause is a combination of the worlds highest taxes, the worlds most powerful labour unions and the worlds probably most generous sick leave benefits.
I'm not sure software is neccesarily so important when compared to the hardware of these vehicles. I'm currently TA'ing a course in bio-inspired robotics, where students have to build robots to solve a variety of tasks. It turns out most tasks are solvable with the same few lines of code, conceptually similar to the wiring of the first few Braitenberg vehicles, only the robots are slighlty modified and sensors repositioned.
Hardware, especially sensors and their positioning, matter far more than people think.
The system you are thinking of would of course not be a hindrance to anyone wishing to publis his research, if it was implemented. But it still seems like plan economy in at least one important respect: who decides which journals are being available for free? All "academic" journals? I think the borderlines here are very fuzzy (do, for exampe, the IEE Review, New Scientist, and THES all count?) and it is better left to those active in the field to make those decisions in a decenteralized fashion. And it would still be a bit unfair to those scientists who did nothing to deserve that their country refuse to pay the bills.
Not to mention that the system would be quite unnecessary. As the examples of JAIR and JMLR show, it is perfectly sensible to run an academic journal online-only with more or less no operating expenses at all.
As for the unemployed genius in Zimbabwe, s/he probably reads papers he gets directly from researchers' homepages. That's how I get the absolute majority of the papers I read. I am of the opinion that a researcher that doesn't make his writings available on his webpage doesn't care very much about being referenced.
There are actually several problems:
One is the cost per article of publishing research. This would probably be the same regardless of where you are coming from. But a researcher costs much less to maintain (salary, office space etc.) in a poor country than in a rich country - and this must be so in order for the poor countries to have any competitiveness. So the net result is that the number of articles any one researcher would be allowed to publish would be smaller in a poor country than in a rich country. Which is fundamentally at odds with the whole idea of peer-review systems, that whether an article is published or not is dependent on the quality of the research and the presentation of it, and nothing else.
I am here assuming that governments are introducing some sort of systems for endorsing articles they are willing to pay for. A justifiable assumption, as it is unlikely that many governments are willing to pay an unpredictable amount of money for something they have no control over.
Which leads us to other problems. Such as what to do if you are not employed by a university, but unemployed or employed by a private corporation. (I published an article while I was unemployed and working from home, and this was essential in getting me my PhD studentship.) Or if your employer doesn't like the research you do. Or if your country doesn't like the research you do (Iran might be unwilling to pay for publishing research in evolutionary biology or gender studies, and the US might be unwilling to pay for applied marxism or similar.) Or if you are collaborating with the "wrong" people in th wrong countries.
And what if your country refuses to take part at all in international systems like this?
In short, how would this system work if you're an unemployed genius working in Zimbabwe?
In my views, the proposed system has all the problems of a plan economic system, and not even a semblance of fairness.
In the field of AI we have at least two highly respected journals which do not have paper editions (even though libraries can buy bound collections of papers on a yearly basis) and which make their content available for free to everyone:
Journal of Artificial Intelligence Research and
Journal of Machine Learning Research
This works because an academic journal does not really have any expenses for peer-review. Academics review for free as part of their job - it gives status to review for a prestigious journal. If you don't have any costs for editing and printing a paper edition, you suddenly have almost no operating expenses at all. Cost of bandwith is negligible. A typical research paper in pdf format is a 100k download, so any one of us could operate one of those servers from our home. Furthermore, the cost of bandwith is continually decreasing.
In sum, I don't understand what the IEEE is whining about. Let those who want a journal on paper pay for the paper, and let the rest of us have it for free!
Another problem is surely that some countries are richer than others. A US scientist would be able to publish so much more than an Indian scientist of the same calibre.
And yet another is that of the country's government deciding which of its scientists gets to publish.
Sounds awfully much like plan economy to me.
It could have been more than a thin wrapper for specific Win32 calls. It could have been an API implementable on more than one platform, for example, with .NET calls supervenient on the native calls on more than one OS.
By the way, the word you are looking for is probably "modular" rather than "functional". Functional programming is what you do with LISP.
Rodney Brooks (in his pop-sci book, which seems to have come out under several different names, amongst them "Flesh and machines") has an interesting answer to your worry: we will integrate our bodies with the technology. In that way, AI will never take over the world, because we will be one with the technology, and a cyborg incorporating both flesh and silicon will always have an advantage over a pure silicon system. I pretty much agree. But let us hypothetically grant you the (huge) benefit of the doubt. How would we do to discontinue or at least severely restrict the research that would potentially be dangerous, and only that research? The problem is that "AI" is a very vague umbrella term nowadays. Back in the fifties, it was actually possible for someone to be an expert in AI - nowadays, most of the research done in computer science departments (and then quite some in other departments, such as electrical engineering) is stuff that has at some point been called artificial intelligence. In fact, things as diverse as object-oriented programming, railway scheduling systems and much modern image processing has sprung out of research that was originally aimed at replicating the mechanisms of the mind. I like telling my mother that I "work in AI", but to anyone who knows anything about the field that is as informative as saying that I sit in front of computers all day. So, would you be ready to pull the plug on most CS research? Or do you have any specific suggestions?
From what I can understand of the papers downloadable from the guys website hes primarily working in physical (humanoid) robotics and evolutionary computation. Getting from there to evolving aspects of robot behaviour on the physical robots is not very far fetched - if you then want to call these behaviour-aspects "emotions" is entirely up to you... However, similar things have been done before, by the eminent people at Brandeis: http://www.demo.cs.brandeis.edu/pr/ee/
I would say it's on the wrong track. It seems to have at least one foot on the ground at all times, and thus displays a "stable" and not a "dynamic" gait. Also, notice how it seems to walk with its knees bent all the time. For an in-depth look at more natural-like (and potentially more useful) approaches to walking, see http://www.cogs.susx.ac.uk/users/inmanh/HART2004_H arvey_etal.pdf
For some amazing movies of evolved dynamic walking, see http://www.droidlogic.com/