It's particularly relevant in areas of CS research with significant philosophical implications, like AI. In some cases knowing relevant philosophical problems can point out likely technical problems and potential approaches to solving them.
For example, machine learning repeatedly bangs its head against the age-old philosophical problem of induction, and in my view (as an AI academic), the people who know about that and the relevant literature are more likely to make non-naive technical contributions.
Reinforcement learning (a specific branch dealing with learning how to act in an environment) bangs its head against issues like the relationship between something we might call "the real world", the data from your senses, and how to infer between them. Specific technical proposals have largely recapitulated some of the philosophical debate: for example, there was a semi-recent and somewhat influential proposal to replace a priori "states", which represent a view of the "real" states in an environment, with phenomenological state, constructed on the fly from sequences of sensor values clustered based on their ability to predict future sensor values (Predictive State Representations, or PSRs). This is essentially recapitulating the empiricists' "sense-data" view of the early 20th century, which they proposed as a replacement for metaphysical ontologies of the world.
Or, slightly more specifically, it depends on what parts of symbolic logic you focus on. Given a specific system of symbolic logic, working out its technical implications is yes, essentially mathematics or theoretical computer science. Using it to implement automated reasoners is artificial intelligence (a branch of computer science).
Designing logics can go either way, though. You could do it purely as a technical matter: you want a logic with a particular property, so you design one that has that property. Most logics are designed from a more philosophical perspective, though: logic basically as a way of formalizing statements and ways of reasoning about statements. From Aristotle through the middle ages people had catalogued valid and fallacious methods of reasoning; a system of logic encompasses a formalization of such a system. It also has ontological implications, depending on what you decide to make representable in the logic, and what you view as the implications of doing so. For example, W.V.O. Quine's works on logic, while they contain technical results as well, are mainly philosophical in nature. Bertrand Russell's research program in logic, while it contained a lot of technical results, was also primarily philosophical in nature.
You said that if you look at a spectrum as an image you can spot degradation in a lossy encoder. That's entirely irrelevant, since mp3 is a perceptual encoder; it deliberately loses information that, according to its predictive model, listeners cannot actually hear. To show that it fails to do this, you need to conduct a blind listening test with the encoded and un-encoded version, and actually reliably (i.e. significantly more than 50% of the time) pick out the encoded one.
The stock closed over $30 every day for approximately two weeks. During that time, approximately 700 million buy/sell transactions took place. Unless you were the Yahoo founder (who didn't actually want to sell), there was such incredibly high volume (topping 80 million shares changing hands on some individual days) that you could've sold any realistically sized position.
YHOO was trading for around $30 on the open market at the time of that offer. If shareholders wanted to sell out, they could've just sold their shares through any broker on the exchange, and gotten $30. Instead they wanted to hang on and get the extra few bucks of merger arbitrage; and they lost their bet, since the merger didn't go through.
I don't really get blaming Yahoo's board here--- anyone who wanted to sell out at that price could have, without needing Yahoo's management's permission to do so.
Mandatory conscription into forced work for the State is also a breeding ground for ideologies that are antagonistic to freedom.
The idea that individuals can legitimately be forced to spend their time in service of the State against their will has historically been promoted by national-patriotic conservatives, usually for military conscription but occasionally for other reasons, and been anathema to liberals.
As Thomas Jefferson aptly noted: "In Virginia a draft was ever the most unpopular and impracticable thing that could ever be attempted. Our people, even under the monarchical government, had learned to consider it as the last of all oppressions."
According to the FAQ anyway (I haven't parsed the license text fully), the Nov 1 cutoff is only for externally originating GFDL content that was imported into Wikipedia, like the FOLDOC entries that Wikipedia imported years ago. Any material after that date that *originates* on Wikipedia can still be relicensed.
Whoever decided on trying to shove a round Wikipedia into a square GFDL initially made a big mistake. I'm suspecting they just didn't even read the GFDL or make any cursory effort to try to map the concepts in the GFDL, which are very obviously designed for software manuals and only for software manuals, onto what they were trying to build. Rather, probably did more of a "hmm, I need a free license for text? I hear that's what the GFDL is, let's use it". Then a few months later people started inventing rationales about which part is the "Title Page" and which is the "History section" once we were already stuck. At least when I started editing around 2003 nobody had any clue what the GFDL meant as applied to Wikipedia, and hadn't even made that much effort yet at clarifying reuse conditions.
It probably would've been a better decision to write a one-paragraph, Wikipedia-specific license on the back of a napkin, doing some sort of generic copyleft with an option of collective attribution to "got this from the Wikipedia article 'articlename'".
It really is a practicality problem for "meaningful reuse of its content". If you have to staple the entire text of the GFDL to a short article that you hope to print on a flier, you effectively can't reuse that article on a flier. What's more, no reuser can be confident that they're even doing it legally, even if they're willing to take heroic measures. The FSF will not say what the GFDL means as applied to Wikipedia. What is the History section? What is a derived version? What is the Title Page? When it says maintain the history section, does that mean you have to include on every redistributed copy the entire wiki edit history? If so, that surely limits the practicality of reuse.
In any case, I've been one of the people pushing for reuse (though I have no particular position in the WMF, and have not been actively involved), and I'm a strong proponent of the FSF. I like the GPL and LGPL lots, and recommend them as best-choice free-software licenses. But that doesn't mean everything they touch is golden.
As for your allegation that people supporting this change "hate the viral nature" of the GFDL, that wouldn't make any sense--- this relicensing is to another viral license. This isn't a relicensing to just any Creative Commons license you want to pick, but specifically to the Creative Commons Attribution Share-Alike (cc-by-sa); the "share-alike" part is a viral copyleft clause.
This is basically a special-case clause to let Wikipedia get out of the GFDL and relicense itself to cc-by-sa, because the GFDL turns out to be highly impractical for Wikipedia and especially for any meaningful reuse of its content.
The date clause is designed to prevent someone from using this as a way to relicense all GFDL content that has ever been created, by laundering it through Wikipedia. Since you didn't know about this license until too late, you can't now go take a GFDL software manual, paste it into Wikipedia, and say this allows you to relicense it. Since people who wrote manuals years ago were not expecting to have their work relicensed in this way, the FSF felt compelled to avoid that outcome.
Basically, Wikipedia was GFDL'd because the GFDL existed at the time. Since then, cc-by-sa has gotten a lot more momentum everywhere else, so it would be nice to move to it so content can be reused between Wikipedia and the many cc-by-sa books, websites, etc. that come out frequently.
The other reason is that the GFDL was designed for software manuals, so some of its technical requirements are highly impractical. You must reprint the entire GFDL text, which is several pages long, with any reuse. Fine if you're reprinting a book of 5,000 Wikipedia articles. But if you just want to print one on a flier, do you have to attach a pamphlet containing the GFDL text to every copy of the flier? And where the hell would you fit the list of all the article's authors, in the "History" section the GFDL requires you to maintain? Cc-by-sa has generally much more reasonable reuse requirements for all of this.
There's lots of speculation about telecomm lobbyists and whatnot, but I don't really think the degree of lockstep on this issue can be explained simply by AT&T dollars. Moreover, if they just wanted to shield AT&T, they could've allowed investigations to go through, but capped damages for anything that AT&T could show the government had ordered them to do to a very nominal figure, or even agreed to have the government cover the damages.
They pretty clearly though, in both parties, didn't want this investigation happening at all. My guess is that that's because, if it were fully investigated with subpoena power and whatnot, we'd find out that both a number of administration members broke the law, but also that a number of Congress members from both parties either broke the law or knew about law-breaking.
"Tracing basic implications" is hardly the only thing computers do in mathematics; there is plenty of work on the "flash of insight" part, which computers have done successfully on a number of occasions. In particular, there's a long body of work on conjecture-generating systems, which don't try to prove things, but look for conjectures that: 1) would be interesting if true; and 2) seem that they could at least plausibly be true. Generating conjectures is historically a large part of the creativity in mathematics, and in some areas, computers are getting good enough at it that professional mathematicians use conjecture-generating software to get ideas for interesting problems to work on or useful lemmas to prove on the way to another problem.
This survey provides a useful overview of some of the work.
I was responding to the claim that it's somehow been "proven" that computers can't do mathematics as well as mathematicians, presumably because of the negative results about formal systems in Goedel's incompleteness theorem. I was pointing out that human mathematicians manipulating formal systems (which is what mathematics consists of) are subject to the same negative results, so this doesn't really prove anything one way or another.
Now whether computers in practice can do interesting mathematics is of course an open problem.
Insofar as a rigorous math proof produced by a human is operating in a formal system (showing that its conclusions logically follow from its premises), it's subject to the same limitations.
Unless you're arguing that there are correct proofs that depend inherently on hand-waving that could not be made more rigorous, which I think few mathematicians would accept.
I guess I don't see a huge benefit to having internet connectivity at every possible moment. I do see a benefit to being able to get online when I'm out easily, but this is solved by living in a city with free-wifi coffeeshops on every other street corner. I guess I don't see the huge advantage to being able to stand on the sidewalk on the internet a block away from the coffee shop instead of just going and sitting down.
As for on the move, depends on how you're moving. =] One of the buses I take most frequently has free wifi on the bus now. Main culprits still missing are commuter rail it seems.
I personally don't spend most of my time roaming the entire earth, so the relevant figure is what proportion of my local urban area is covered by wifi. Now even that could use a lot of work, but I'd estimate offhand that I'm able to get into a free network in about 40% of locations I've tried it at.
Microsoft made a mainly stock offer, of slightly over one share of MSFT per share of YHOO. At the time, MSFT was trading at about $28-29 per share, so this valued Yahoo at ~$31-32 per share. However, MSFT is now at $22, so had Yahoo's investors taken that offer, they would be sitting on something worth ~$23-24.
Of course, they could've sold the Microsoft stock immediately upon completion of the deal for ~$31-32. But if they had wanted to do that, they could've just sold their Yahoo stock in the first place, deal or no deal, because it was trading at about $30 itself at the time.
You could rewrite your post to say "films" or "video" also. People go to YouTube to be entertained by silly videos about cats, not to be fed loads of political horsecrap. People see films to be entertained by Hollywood special effects and love stories, not to get some sort of political message. Oh except those aren't true. People do go to YouTube to be entertained mostly, but that doesn't mean they (or other people) can't also go to YouTube to watch political videos. That a medium s primarily used for entertainment doesn't mean that nobody can ever use it for anything else.
Gamers as some sort of social group may not be particularly special, but the use of game-like setups for rhetorical purposes has been greatly underexplored. Like most media, it will probably eventually differentiate into different areas, like how you can currently watch films designed for entertainment, films designed to inform, films designed to persuade, and many other kinds of films.
In fact, the author of this article has a book-length treatment of the subject (Persuasive Games, MIT Press, 2007).
Many academics do want their writings available for free, if for no other reason than because the average academic-press book (textbooks excluded) makes the author no royalties anyway. The reason they bother to publish in the first place is that "I got a book published with MIT Press" is prestigious, whereas "I gave my book away free on my website as a PDF" is not.
The first tenant at the famed Stanford Research Park was Varian, and the government was at the time Varian's only customer. Many of the other spin-offs were organized around government-funded research labs, many also at/near Stanford, the most famous of which was probably Engelbart's lab (which invented the mouse).
The ideal is that games are partly used as a lure to trick more 18-year-olds into finding a degree in computer science interesting---rather than a class on asm programming on the SPARC or something, you teach them similar concepts with a class that makes them program asm on the Gameboy Advance or Atari 2600, making the low-level architecture/asm class seem more interesting. Of course, programs vary in how exactly they integrate games into the curriculum.
It depends on how you count "most". Science can move forward as long as the really important things that everyone relies on are at least close to correct. Most of the actual results could still be wrong, or seriously off, and you could still get your job done.
In fact, regardless of what they might say in public, a lot of experimental scientists seem to operate under the assumption that most of the relevant published research is wrong, or at least far enough off to be useless. When you want to synthesize some molecule, you don't look up all the published data on mechanisms, piece it together using some first-principles logic, and then assume it's going to work. Usually, it doesn't. Sometimes this is because you messed up, but often it was because the stuff in the literature isn't entirely correct. The very basics are correct: our understanding of how chemical reactions work is not likely to be seriously off. But a lot of the details, like how two particular substances will react in the presence of a third in a particular set of circumstances, need to be taken with a bit of skepticism, especially if you're trying to rely on the claimed implications of an experiment (i.e. the general scientific conclusions it's supposed to imply), rather than literally replicating the exact experiment under the exact same conditions. You only reduce the skepticism a bit if lots and lots of papers have been published on the subject, so even if "most" are wrong, 10 saying the same thing still gives you a good chance the result is right.
Slashdot Mirror
It's particularly relevant in areas of CS research with significant philosophical implications, like AI. In some cases knowing relevant philosophical problems can point out likely technical problems and potential approaches to solving them.
For example, machine learning repeatedly bangs its head against the age-old philosophical problem of induction, and in my view (as an AI academic), the people who know about that and the relevant literature are more likely to make non-naive technical contributions.
Reinforcement learning (a specific branch dealing with learning how to act in an environment) bangs its head against issues like the relationship between something we might call "the real world", the data from your senses, and how to infer between them. Specific technical proposals have largely recapitulated some of the philosophical debate: for example, there was a semi-recent and somewhat influential proposal to replace a priori "states", which represent a view of the "real" states in an environment, with phenomenological state, constructed on the fly from sequences of sensor values clustered based on their ability to predict future sensor values (Predictive State Representations, or PSRs). This is essentially recapitulating the empiricists' "sense-data" view of the early 20th century, which they proposed as a replacement for metaphysical ontologies of the world.
Or, slightly more specifically, it depends on what parts of symbolic logic you focus on. Given a specific system of symbolic logic, working out its technical implications is yes, essentially mathematics or theoretical computer science. Using it to implement automated reasoners is artificial intelligence (a branch of computer science).
Designing logics can go either way, though. You could do it purely as a technical matter: you want a logic with a particular property, so you design one that has that property. Most logics are designed from a more philosophical perspective, though: logic basically as a way of formalizing statements and ways of reasoning about statements. From Aristotle through the middle ages people had catalogued valid and fallacious methods of reasoning; a system of logic encompasses a formalization of such a system. It also has ontological implications, depending on what you decide to make representable in the logic, and what you view as the implications of doing so. For example, W.V.O. Quine's works on logic, while they contain technical results as well, are mainly philosophical in nature. Bertrand Russell's research program in logic, while it contained a lot of technical results, was also primarily philosophical in nature.
You said that if you look at a spectrum as an image you can spot degradation in a lossy encoder. That's entirely irrelevant, since mp3 is a perceptual encoder; it deliberately loses information that, according to its predictive model, listeners cannot actually hear. To show that it fails to do this, you need to conduct a blind listening test with the encoded and un-encoded version, and actually reliably (i.e. significantly more than 50% of the time) pick out the encoded one.
The stock closed over $30 every day for approximately two weeks. During that time, approximately 700 million buy/sell transactions took place. Unless you were the Yahoo founder (who didn't actually want to sell), there was such incredibly high volume (topping 80 million shares changing hands on some individual days) that you could've sold any realistically sized position.
YHOO was trading for around $30 on the open market at the time of that offer. If shareholders wanted to sell out, they could've just sold their shares through any broker on the exchange, and gotten $30. Instead they wanted to hang on and get the extra few bucks of merger arbitrage; and they lost their bet, since the merger didn't go through.
I don't really get blaming Yahoo's board here--- anyone who wanted to sell out at that price could have, without needing Yahoo's management's permission to do so.
Mandatory conscription into forced work for the State is also a breeding ground for ideologies that are antagonistic to freedom.
The idea that individuals can legitimately be forced to spend their time in service of the State against their will has historically been promoted by national-patriotic conservatives, usually for military conscription but occasionally for other reasons, and been anathema to liberals.
As Thomas Jefferson aptly noted: "In Virginia a draft was ever the most unpopular and impracticable thing that could ever be attempted. Our people, even under the monarchical government, had learned to consider it as the last of all oppressions."
According to the FAQ anyway (I haven't parsed the license text fully), the Nov 1 cutoff is only for externally originating GFDL content that was imported into Wikipedia, like the FOLDOC entries that Wikipedia imported years ago. Any material after that date that *originates* on Wikipedia can still be relicensed.
Whoever decided on trying to shove a round Wikipedia into a square GFDL initially made a big mistake. I'm suspecting they just didn't even read the GFDL or make any cursory effort to try to map the concepts in the GFDL, which are very obviously designed for software manuals and only for software manuals, onto what they were trying to build. Rather, probably did more of a "hmm, I need a free license for text? I hear that's what the GFDL is, let's use it". Then a few months later people started inventing rationales about which part is the "Title Page" and which is the "History section" once we were already stuck. At least when I started editing around 2003 nobody had any clue what the GFDL meant as applied to Wikipedia, and hadn't even made that much effort yet at clarifying reuse conditions.
It probably would've been a better decision to write a one-paragraph, Wikipedia-specific license on the back of a napkin, doing some sort of generic copyleft with an option of collective attribution to "got this from the Wikipedia article 'articlename'".
It really is a practicality problem for "meaningful reuse of its content". If you have to staple the entire text of the GFDL to a short article that you hope to print on a flier, you effectively can't reuse that article on a flier. What's more, no reuser can be confident that they're even doing it legally, even if they're willing to take heroic measures. The FSF will not say what the GFDL means as applied to Wikipedia. What is the History section? What is a derived version? What is the Title Page? When it says maintain the history section, does that mean you have to include on every redistributed copy the entire wiki edit history? If so, that surely limits the practicality of reuse.
In any case, I've been one of the people pushing for reuse (though I have no particular position in the WMF, and have not been actively involved), and I'm a strong proponent of the FSF. I like the GPL and LGPL lots, and recommend them as best-choice free-software licenses. But that doesn't mean everything they touch is golden.
As for your allegation that people supporting this change "hate the viral nature" of the GFDL, that wouldn't make any sense--- this relicensing is to another viral license. This isn't a relicensing to just any Creative Commons license you want to pick, but specifically to the Creative Commons Attribution Share-Alike (cc-by-sa); the "share-alike" part is a viral copyleft clause.
This is basically a special-case clause to let Wikipedia get out of the GFDL and relicense itself to cc-by-sa, because the GFDL turns out to be highly impractical for Wikipedia and especially for any meaningful reuse of its content.
The date clause is designed to prevent someone from using this as a way to relicense all GFDL content that has ever been created, by laundering it through Wikipedia. Since you didn't know about this license until too late, you can't now go take a GFDL software manual, paste it into Wikipedia, and say this allows you to relicense it. Since people who wrote manuals years ago were not expecting to have their work relicensed in this way, the FSF felt compelled to avoid that outcome.
Basically, Wikipedia was GFDL'd because the GFDL existed at the time. Since then, cc-by-sa has gotten a lot more momentum everywhere else, so it would be nice to move to it so content can be reused between Wikipedia and the many cc-by-sa books, websites, etc. that come out frequently.
The other reason is that the GFDL was designed for software manuals, so some of its technical requirements are highly impractical. You must reprint the entire GFDL text, which is several pages long, with any reuse. Fine if you're reprinting a book of 5,000 Wikipedia articles. But if you just want to print one on a flier, do you have to attach a pamphlet containing the GFDL text to every copy of the flier? And where the hell would you fit the list of all the article's authors, in the "History" section the GFDL requires you to maintain? Cc-by-sa has generally much more reasonable reuse requirements for all of this.
There's lots of speculation about telecomm lobbyists and whatnot, but I don't really think the degree of lockstep on this issue can be explained simply by AT&T dollars. Moreover, if they just wanted to shield AT&T, they could've allowed investigations to go through, but capped damages for anything that AT&T could show the government had ordered them to do to a very nominal figure, or even agreed to have the government cover the damages.
They pretty clearly though, in both parties, didn't want this investigation happening at all. My guess is that that's because, if it were fully investigated with subpoena power and whatnot, we'd find out that both a number of administration members broke the law, but also that a number of Congress members from both parties either broke the law or knew about law-breaking.
"Tracing basic implications" is hardly the only thing computers do in mathematics; there is plenty of work on the "flash of insight" part, which computers have done successfully on a number of occasions. In particular, there's a long body of work on conjecture-generating systems, which don't try to prove things, but look for conjectures that: 1) would be interesting if true; and 2) seem that they could at least plausibly be true. Generating conjectures is historically a large part of the creativity in mathematics, and in some areas, computers are getting good enough at it that professional mathematicians use conjecture-generating software to get ideas for interesting problems to work on or useful lemmas to prove on the way to another problem.
This survey provides a useful overview of some of the work.
I was responding to the claim that it's somehow been "proven" that computers can't do mathematics as well as mathematicians, presumably because of the negative results about formal systems in Goedel's incompleteness theorem. I was pointing out that human mathematicians manipulating formal systems (which is what mathematics consists of) are subject to the same negative results, so this doesn't really prove anything one way or another.
Now whether computers in practice can do interesting mathematics is of course an open problem.
Insofar as a rigorous math proof produced by a human is operating in a formal system (showing that its conclusions logically follow from its premises), it's subject to the same limitations.
Unless you're arguing that there are correct proofs that depend inherently on hand-waving that could not be made more rigorous, which I think few mathematicians would accept.
I guess I don't see a huge benefit to having internet connectivity at every possible moment. I do see a benefit to being able to get online when I'm out easily, but this is solved by living in a city with free-wifi coffeeshops on every other street corner. I guess I don't see the huge advantage to being able to stand on the sidewalk on the internet a block away from the coffee shop instead of just going and sitting down.
As for on the move, depends on how you're moving. =] One of the buses I take most frequently has free wifi on the bus now. Main culprits still missing are commuter rail it seems.
Adding an unlimited-internet data plan for a Treo costs $15/mo with Sprint.
I personally don't spend most of my time roaming the entire earth, so the relevant figure is what proportion of my local urban area is covered by wifi. Now even that could use a lot of work, but I'd estimate offhand that I'm able to get into a free network in about 40% of locations I've tried it at.
Microsoft made a mainly stock offer, of slightly over one share of MSFT per share of YHOO. At the time, MSFT was trading at about $28-29 per share, so this valued Yahoo at ~$31-32 per share. However, MSFT is now at $22, so had Yahoo's investors taken that offer, they would be sitting on something worth ~$23-24.
Of course, they could've sold the Microsoft stock immediately upon completion of the deal for ~$31-32. But if they had wanted to do that, they could've just sold their Yahoo stock in the first place, deal or no deal, because it was trading at about $30 itself at the time.
You could rewrite your post to say "films" or "video" also. People go to YouTube to be entertained by silly videos about cats, not to be fed loads of political horsecrap. People see films to be entertained by Hollywood special effects and love stories, not to get some sort of political message. Oh except those aren't true. People do go to YouTube to be entertained mostly, but that doesn't mean they (or other people) can't also go to YouTube to watch political videos. That a medium s primarily used for entertainment doesn't mean that nobody can ever use it for anything else.
Gamers as some sort of social group may not be particularly special, but the use of game-like setups for rhetorical purposes has been greatly underexplored. Like most media, it will probably eventually differentiate into different areas, like how you can currently watch films designed for entertainment, films designed to inform, films designed to persuade, and many other kinds of films.
In fact, the author of this article has a book-length treatment of the subject (Persuasive Games, MIT Press, 2007).
Many academics do want their writings available for free, if for no other reason than because the average academic-press book (textbooks excluded) makes the author no royalties anyway. The reason they bother to publish in the first place is that "I got a book published with MIT Press" is prestigious, whereas "I gave my book away free on my website as a PDF" is not.
The first tenant at the famed Stanford Research Park was Varian, and the government was at the time Varian's only customer. Many of the other spin-offs were organized around government-funded research labs, many also at/near Stanford, the most famous of which was probably Engelbart's lab (which invented the mouse).
The ideal is that games are partly used as a lure to trick more 18-year-olds into finding a degree in computer science interesting---rather than a class on asm programming on the SPARC or something, you teach them similar concepts with a class that makes them program asm on the Gameboy Advance or Atari 2600, making the low-level architecture/asm class seem more interesting. Of course, programs vary in how exactly they integrate games into the curriculum.
It depends on how you count "most". Science can move forward as long as the really important things that everyone relies on are at least close to correct. Most of the actual results could still be wrong, or seriously off, and you could still get your job done.
In fact, regardless of what they might say in public, a lot of experimental scientists seem to operate under the assumption that most of the relevant published research is wrong, or at least far enough off to be useless. When you want to synthesize some molecule, you don't look up all the published data on mechanisms, piece it together using some first-principles logic, and then assume it's going to work. Usually, it doesn't. Sometimes this is because you messed up, but often it was because the stuff in the literature isn't entirely correct. The very basics are correct: our understanding of how chemical reactions work is not likely to be seriously off. But a lot of the details, like how two particular substances will react in the presence of a third in a particular set of circumstances, need to be taken with a bit of skepticism, especially if you're trying to rely on the claimed implications of an experiment (i.e. the general scientific conclusions it's supposed to imply), rather than literally replicating the exact experiment under the exact same conditions. You only reduce the skepticism a bit if lots and lots of papers have been published on the subject, so even if "most" are wrong, 10 saying the same thing still gives you a good chance the result is right.