i've been reading a bit lately in the philosophy of science, although i am certainly not a philospher by profession. parsimony -- choose the "simpler" of two theories when both account for the same empirical data -- is a rule of thumb, a heuristic, that is violated in at least two ways.
first, as the nobel laureate physicist steven weinberg eloquently argues in his "dreams of a final theory," theoreticians are often guided more by their aesthetic senses than parsimony, and choose complex but beautiful theories over simpler, but sterile ones. check out this book if you are at all interested in the art of scientific theorizing -- it's really quite underrated.
second, there seem to be no good arguments that choosing simple theories over complex ones actually leads to the selection of truer theories over their more-false alternatives. the philosopher james mcallister's "beauty and revolution in science" covers this ground.
a long, long time ago -- the early 1980s if i recall correctly -- there was a computer called the 'dimension' into which you could plug up to four processor boards. one had a z80 or 8080 and made it behave like a cp/m system, one had am 8088 i believe and endowed with the powers of an ibm xt, etc. it seemed like the swiss army knife of computers, which was enticing back then because there were ten or more incompatible microcomputers around, ranging from the $100s to the $1000s. the dimension purported to allow you to avoid the perils of locking in on the wrong platform. (anyone remember the exidy sorcerer?) i never saw one live -- just in the pages of byte magazine. don't know what happened to them, either.
the uberplatform is like an urban myth that surfaces every five years, always in different forms but with the same ultimate end.
There are two famous books by the phenomenologist philosopher Hubert Dreyfus on the folly of Artificial Intelligence.
"What Computers Can't Do: A Critiqe of Artificial Reason"
"What Computers Still Can't Do: A Critiqe of Artificial Reason"
AI folks hate these books for many reasons, but especially because Dreyfus is a technical doofus. He consistently misunderstands what computation is, how computers are programmed, etc. (Sometimes with comical results -- there's a great story in Levy's "Hackers" about Dreyfus claiming (in the 1960s) that no computer would ever play decent chess and then being soundly defeated by a primitive chess-playing program shortly thereafter.)
It's pretty clear that the title of Harel's book ("What Computers Really Can't Do") plays on the titles of Dreyfus's books, reasoning soundly about the formal limits of computation rather than insinuating rhetorically about what computation cannot be based on a particular philosophical (phenomenological) critique.
I haven't read the book, but I was given a copy of Harel's earlier book, "Algorithmics: The Spirit of Computing" by my prof as a freshman CS major, and I can say he (1) really knows theoretical computer science and (2) is an excellent writer. This is the book I recommend to people with plenty of knowledge about the practice of computer programming or network administration who desire to know more about the theoretical foundations of computer science.
I have no doubt that Harel's new book is both technically sound and pedagogically accessible. I'm gonna buy it as soon as I can!
Slashdot Mirror
i've been reading a bit lately in the philosophy of science, although i am certainly not a philospher by profession. parsimony -- choose the "simpler" of two theories when both account for the same empirical data -- is a rule of thumb, a heuristic, that is violated in at least two ways.
first, as the nobel laureate physicist steven weinberg eloquently argues in his "dreams of a final theory," theoreticians are often guided more by their aesthetic senses than parsimony, and choose complex but beautiful theories over simpler, but sterile ones. check out this book if you are at all interested in the art of scientific theorizing -- it's really quite underrated.
second, there seem to be no good arguments that choosing simple theories over complex ones actually leads to the selection of truer theories over their more-false alternatives. the philosopher james mcallister's "beauty and revolution in science" covers this ground.
a long, long time ago -- the early 1980s if i recall correctly -- there was a computer called the 'dimension' into which you could plug up to four processor boards. one had a z80 or 8080 and made it behave like a cp/m system, one had am 8088 i believe and endowed with the powers of an ibm xt, etc. it seemed like the swiss army knife of computers, which was enticing back then because there were ten or more incompatible microcomputers around, ranging from the $100s to the $1000s. the dimension purported to allow you to avoid the perils of locking in on the wrong platform. (anyone remember the exidy sorcerer?) i never saw one live -- just in the pages of byte magazine. don't know what happened to them, either.
the uberplatform is like an urban myth that surfaces every five years, always in different forms but with the same ultimate end.
There are two famous books by the phenomenologist philosopher Hubert Dreyfus on the folly of Artificial Intelligence.
"What Computers Can't Do: A Critiqe of Artificial Reason"
"What Computers Still Can't Do: A Critiqe of Artificial Reason"
AI folks hate these books for many reasons, but especially because Dreyfus is a technical doofus. He consistently misunderstands what computation is, how computers are programmed, etc. (Sometimes with comical results -- there's a great story in Levy's "Hackers" about Dreyfus claiming (in the 1960s) that no computer would ever play decent chess and then being soundly defeated by a primitive chess-playing program shortly thereafter.)
It's pretty clear that the title of Harel's book ("What Computers Really Can't Do") plays on the titles of Dreyfus's books, reasoning soundly about the formal limits of computation rather than insinuating rhetorically about what computation cannot be based on a particular philosophical (phenomenological) critique.
I haven't read the book, but I was given a copy of Harel's earlier book, "Algorithmics: The Spirit of Computing" by my prof as a freshman CS major, and I can say he (1) really knows theoretical computer science and (2) is an excellent writer. This is the book I recommend to people with plenty of knowledge about the practice of computer programming or network administration who desire to know more about the theoretical foundations of computer science.
I have no doubt that Harel's new book is both technically sound and pedagogically accessible. I'm gonna buy it as soon as I can!