(Caveat: this comment is based on reading the project overview - please correct me if I missed something)
I have also pondered such a technology-based solution to censorship. Although the idea is attractive, a huge obstacle to implementation, which as far as I can tell Freenet does not address, is the spam problem. If such a system allowing anonymous posting ever becomes widely used (which is the goal, after all), I would predict that spammers would quickly clog it to the point of making it unusable (cite Usenet as precedent).
This problem rules out most straightforward implementations. My idea to solve this is not fully developed, but would be a network that anyone could access, but that only those who "payed" could post on. Payment would consist of operating a server, and via an untracable digital cash scheme, operation of the server would give you the right to post (minus system overhead).
For example, if you run a server that can store 20 MB of content from the network and the overhead tax is 50%, you would get digital cash that let you post 10 MB of material on the network. This would prevent spam, since the spammers couldn't get a free ride: they would have to pony up storage space.
The big problem (obviously) is devising an authentication system that creates the digital cash that is both 1) secure and 2) does not constitute a point of attack for a goverment trying to bring down the system.
I'd be interested in hearing whatever ideas people might have as to how to accomplish this.
According to the request for comments, electronically submitted comments will be posted on a web page. I have just checked the page and so far no comments have been put up, but in the future it might be a good source for ideas.
If you placed your electronics inside a Faraday cage, they would be invulnerable to EMP. A Faraday cage is basically a box (or a room if you're thinking big) which has an electrical conductor completely surrounding it.
Even with medieval technlogy this would not be too hard to build, and once you had the orbits of the Wanderers figured out you would know when the box / room had to be closed up. Figuring the orbits would be easy as there is no way the satellites could carry enough reaction mass to still be manuevering millenia later.
I agree that it is easiest to presume sentience in entities like ourselves. We all use this as a heuristic in everyday life: you don't subject people you meet to the Turing test before deciding they are sentient. Instead we assume because an entity looks human it has an inner life similar to our own. I also agree that it is more difficult to decide on the sentience of entities explicitly created as mimics. The experiences with Elisa, in which people ascribed complex mental characteristics to a very simple program, showed that naive individuals can be fooled.
However, just because it is harder to make a decision about the sentience of entities that are more different from you doesn't mean it is impossible. Otherwise, you would seem to be taking the position that humans (and perhaps a few other mammals) are the only entities in the entire universe that can be sentient, just because they happen to closely resemble you.
The example of wetness of a simulated ocean is a bit of a red herring. I agree that no matter how well modelled, a simulated ocean can never be wet to an operator outside the machine, nor could it ever drown her or taste salty. But the quality of thinking is different from the quality of wetness. Consider a legal brief or a computer program. For these objects it is the pattern, not the physical medium that the pattern is embeded in that is important. It does not make any sense to talk about a 'simulated' legal brief or a 'simulated' computer program - each object with the specified pattern is in some sense as good as the original, although in practice one physical embodiement may be more practical to use in a specific situation.
In the same way, the important thing about sentience is its pattern, not its physical embodiment. Even neurobiologists have only the vaugest grasp of the physical embodiement underlying our sentience, but that doesn't stop us from thinking and experiencing.
I think that the Turing test remains an excellent minimum test for sentience. I can imaging sentient entities that would not pass the Turing test, but if the test is long enough and with a good enough examiner, I don't see any way that a non-sentient entity could pass it. I would argue that the fact that no computer program has come even close to passing a Turning test is a sign of the test's strength.
It seems highly implausible to me that we will be able to create sentient, human-level intelligence any time before the hardware power of computers at least matches the hardware power of the human brain (estimated at about 1000 million MIPS, see http://www.transhumanist.com/volum e1/moravec.htm ). If Moore's law continues to hold, this amount of computer power will cost about $1000 in the 2020-2030 time frame (see figure 2 in the cited article).
You argue that because we don't know the exact basis of consciousness, we have no ability to decide whether an entity is conscious or not, and thus can't know whether it is deserving of rights or not. While an interesting philosophical position, that stance is simply not practical for living in the uncertain, messy world that we all inhabit. People make decisions on the basis of uncertain information and imperfect understandings all the time. Some of those decisions have really important consequences, like whether people get the right diagnosis and treatment and live, or die when they have an illness (doctors); whether people get deprived of many of their rights and go to jail (judges, juries); and even what legal rights you have in the first place (legislators, the electorate at large).
It would be nice if we had ways of removing that uncertainty from life, but we don't. So I see no reason why a similar level of uncertainty presents any barrier to the practical decision about whether an AI is conscious or not.
You leave out the point that you can only verify the subjective experience of one person - yourself. How do you know that other people have subjective experiences or are sentient? Only by observing their behavior (includes communicative behavior).
That is the same way we will decide whether AIs have a subjective experience (leaving out the possiblity of uploading - in which case you will have direct evidence for the subjective experience of a single 'AI' - yourself): by observing their objective behavior and talking with them. We will have no direct way of verifying their subjective experience, but that does not seem to be a handicap (most of the time) in dealing with other people, where again we have no direct way of verifying their subjective experience.
Another way to see this point is to paraphrase the first part of your last sentence: "It's not hard for me to imagine a world in which a bunch of people act as if they had some sort sort of a culture and a sentient life, but in fact are just a bunch of overcomplicated tree shrews..."
Slashdot Mirror
(Caveat: this comment is based on reading the project overview - please correct me if I missed something)
I have also pondered such a technology-based solution to censorship. Although the idea is attractive, a huge obstacle to implementation, which as far as I can tell Freenet does not address, is the spam problem. If such a system allowing anonymous posting ever becomes widely used (which is the goal, after all), I would predict that spammers would quickly clog it to the point of making it unusable (cite Usenet as precedent).
This problem rules out most straightforward implementations. My idea to solve this is not fully developed, but would be a network that anyone could access, but that only those who "payed" could post on. Payment would consist of operating a server, and via an untracable digital cash scheme, operation of the server would give you the right to post (minus system overhead).
For example, if you run a server that can store 20 MB of content from the network and the overhead tax is 50%, you would get digital cash that let you post 10 MB of material on the network. This would prevent spam, since the spammers couldn't get a free ride: they would have to pony up storage space.
The big problem (obviously) is devising an authentication system that creates the digital cash that is both 1) secure and 2) does not constitute a point of attack for a goverment trying to bring down the system.
I'd be interested in hearing whatever ideas people might have as to how to accomplish this.
According to the request for comments, electronically submitted comments will be posted on a web page. I have just checked the page and so far no comments have been put up, but in the future it might be a good source for ideas.
If you placed your electronics inside a Faraday cage, they would be invulnerable to EMP. A Faraday cage is basically a box (or a room if you're thinking big) which has an electrical conductor completely surrounding it.
Even with medieval technlogy this would not be too hard to build, and once you had the orbits of the Wanderers figured out you would know when the box / room had to be closed up. Figuring the orbits would be easy as there is no way the satellites could carry enough reaction mass to still be manuevering millenia later.
You raise several points:
I agree that it is easiest to presume sentience in entities like ourselves. We all use this as a heuristic in everyday life: you don't subject people you meet to the Turing test before deciding they are sentient. Instead we assume because an entity looks human it has an inner life similar to our own. I also agree that it is more difficult to decide on the sentience of entities explicitly created as mimics. The experiences with Elisa, in which people ascribed complex mental characteristics to a very simple program, showed that naive individuals can be fooled.
However, just because it is harder to make a decision about the sentience of entities that are more different from you doesn't mean it is impossible. Otherwise, you would seem to be taking the position that humans (and perhaps a few other mammals) are the only entities in the entire universe that can be sentient, just because they happen to closely resemble you.
The example of wetness of a simulated ocean is a bit of a red herring. I agree that no matter how well modelled, a simulated ocean can never be wet to an operator outside the machine, nor could it ever drown her or taste salty. But the quality of thinking is different from the quality of wetness. Consider a legal brief or a computer program. For these objects it is the pattern, not the physical medium that the pattern is embeded in that is important. It does not make any sense to talk about a 'simulated' legal brief or a 'simulated' computer program - each object with the specified pattern is in some sense as good as the original, although in practice one physical embodiement may be more practical to use in a specific situation.
In the same way, the important thing about sentience is its pattern, not its physical embodiment. Even neurobiologists have only the vaugest grasp of the physical embodiement underlying our sentience, but that doesn't stop us from thinking and experiencing.
I think that the Turing test remains an excellent minimum test for sentience. I can imaging sentient entities that would not pass the Turing test, but if the test is long enough and with a good enough examiner, I don't see any way that a non-sentient entity could pass it. I would argue that the fact that no computer program has come even close to passing a Turning test is a sign of the test's strength.
It seems highly implausible to me that we will be able to create sentient, human-level intelligence any time before the hardware power of computers at least matches the hardware power of the human brain (estimated at about 1000 million MIPS, see http://www.transhumanist.com/volum e1/moravec.htm ). If Moore's law continues to hold, this amount of computer power will cost about $1000 in the 2020-2030 time frame (see figure 2 in the cited article).
You argue that because we don't know the exact basis of consciousness, we have no ability to decide whether an entity is conscious or not, and thus can't know whether it is deserving of rights or not. While an interesting philosophical position, that stance is simply not practical for living in the uncertain, messy world that we all inhabit. People make decisions on the basis of uncertain information and imperfect understandings all the time. Some of those decisions have really important consequences, like whether people get the right diagnosis and treatment and live, or die when they have an illness (doctors); whether people get deprived of many of their rights and go to jail (judges, juries); and even what legal rights you have in the first place (legislators, the electorate at large).
It would be nice if we had ways of removing that uncertainty from life, but we don't. So I see no reason why a similar level of uncertainty presents any barrier to the practical decision about whether an AI is conscious or not.
You leave out the point that you can only verify the subjective experience of one person - yourself. How do you know that other people have subjective experiences or are sentient? Only by observing their behavior (includes communicative behavior).
That is the same way we will decide whether AIs have a subjective experience (leaving out the possiblity of uploading - in which case you will have direct evidence for the subjective experience of a single 'AI' - yourself): by observing their objective behavior and talking with them. We will have no direct way of verifying their subjective experience, but that does not seem to be a handicap (most of the time) in dealing with other people, where again we have no direct way of verifying their subjective experience.
Another way to see this point is to paraphrase the first part of your last sentence: "It's not hard for me to imagine a world in which a bunch of people act as if they had some sort sort of a culture and a sentient life, but in fact are just a bunch of overcomplicated tree shrews..."