h a l f b a k e r yPoof of concept
add, search, annotate, link, view, overview, recent, by name, random
news, help, about, links, report a problem
browse anonymously,
or get an account
and write.
register,
|
|
|
Please log in.
Before you can vote, you need to register.
Please log in or create an account.
|
No doubt everyone has heard of automated scanner-bots that search the Net for audio/video/etc. files, looking for certain patterns inside... usually for repressive purposes. Think of a format resembling (for example) MP3, with the additional feature of being set up like the body of a polymorphic virus.
The file would contain a Decryptor block at its start, which would in essence be an Interpreted Program which the player software would use to decode the file. After it is played, the file (if stored on a writeable disk) would be replaced with a functionally identical copy without a single byte kept identical to the previous iteration, due to the decoder block and the body of the file having been randomly re-encoded... in short, every use of the file would produce a file that is functionally identical to the previous generation, but NEVER digitally identical. The procedure could be somewhat computationally intensive, but this is hardly a problem for files that must be played in real time, like MP3 (most modern computers already wast most of their CPU cycles when playing an audio file at the usual 44.1KHz.) Thus, censorbots could no longer scan thousands of files, as each one would have to be decoded (i.e. the decoderblock program interpreted, the audiostream produced, and only *then* could it be searched.) **** Files such as audio could also be altered to produce inaudible (or generally not detectable by listeners) changes... this would not be scannable at all without sophisticated AI.
[link]
|
| |
Being quite naive, I had not heard of these scanner bots in relation to multimedia files. I like anything that throws a wrench in corporate hegemony but at the same time, what about the honest musician/author trying to make a living from her work? |
|
| |
Why hasn't the idea caught on that the concept of for-profit creativity is about to die, for better or worse? One day it will no longer be possible for a central authority to control or even monitor citizen-to-citizen distribution of information, be it a political phrase or a terabyte data stream. |
|
| |
If there is a way for users to find what they're looking for, why can't censorbots take precisely the same route? |
|
| |
Technology capable of telling that two files sound similar to a human isn't all that mysterious these days, and doesn't involve artificial intelligence at all. |
|
| |
This came up with the Napster court case where it was demonstrated that Napster's blocking software did not prevent humans finding copyright files. This caused Napster to be ordered to block files more effectively as people knew they could it better. |
|
| |
...plus, if these scanners are still quite dumb put your files into a zip file or other container. |
|
| |
There might be something interesting here. |
|
| |
The Turing Test employs a human to recognize a human. Outside of extremely specialized cases, nothing but a human has passed the Turing Test. |
|
| |
Can we employ a computer to administer something like the Turing Test? This effectively insures that access to the data is possible for any reasonably capable human being, but inaccessible to robots. That doesn't block censors, but it would block automated censorbots and make censorship expensive. |
|
| |
I could easily devise a test only another human could pass (using natural language is a good start, for example). |
|
| |
Could a computer be programmed to generate these tests on demand, such that it's impossible for a computer to pass any of these tests, even if the original algorithm is well-known? |
|
| |
I'm sure someone else has posed this question, but I wouldn't know how to find it. |
|
| |
egnor: This sounds like the challenge of bear-proof bins in American parks. The problem with making bear-proof bins was that there was a considerable overlap in the intelligence of the smartest bear and the dumbest park visitor. |
|
| |
If most people can understand any given system then it is probable that an AI can be taught it as well. |
|
| |
Won't a system that can generate and judge a Turing test (or something like one) have to be more or less capable of passing one itself? It will just be holding up the other end of the conversation. |
|
| |
Aristotle: Fine, eliminate stupid people, if that helps. |
|
| |
Monkfish: Not necessarily. The test could simply be an English sentence; the answer would be a decryption key. Any compatible decryption program is sufficient to act as judge. |
|
| |
Generation is another matter, admittedly, but please note that I'm not talking about the conversational Turing Test per se, just some generalized way to detect human-level intelligence. |
|
| |
Machine-generated Sokoban maps, anyone? Or would they by necessity be machine-solveable? |
|
| |
No, that's true. But a single sentence isn't much of a test -- a search through IRC and Usenet archives could come up with a suitable reply to many -- and a machine that could pose the riddle could also probably answer it. Some way is needed for the electronic sphinx to find a one-way (to it) mapping from a random answer to a generated question which passes through human-only ways of relating things. This is probably the tricky part. |
|
| |
Would anything that requires a simple answer be resistant to brute-force attempts, anyway? A one-word answer would probably be one of fewer than 100,000 English words (it would have to be in most human users' vocabularies) and multiple-attempt limits and copy-protection don't seem possible, at least with this idea. Longer phrases would be difficult to elicit consistently from humans. (Are there many generalized ways of consistently detecting human-level intelligence apart from the Turing test?) |
|
| |
As is probably obvious, though, I know very little about these things. It does sound like a very interesting question, and I realize that that's the important part. |
|
| |
egnor: try submitting a link to Google sometime. In order to get to the submit form, you have to go through a page where Google verifies that you're human, and not a spambot. It does this by presenting you with an image of several letters and numbers in random typefaces, plus noise. It looks like an Nth-generation xerox of a ransom note. If you type in the letters & numbers correctly, you can submit a few links to Google. |
|
| |
I assume that a sufficiently motivated person could write a clever OCR program to do this automatically, but I'm guessing no one has done that yet. |
|
| |
Another reason I don't think this would work is we are apparently talking about publically accessable files that anyone can access other than bots interested in their content. Sending encrypted files to a specific person with private keys involved is a lot easier. |
|
| |
I think ultimately this would boil down to a contest between resources devoted to programming bots and the lengths that people are prepared to incovenience themselves to counter this programming. Google's link submission system, for example, is OK for spambots but I'm sure that academics could crack it. |
|
| |
... a peer-to-peer human verification service that employs the conversational Turing Test ... disguised as a chat room ... or a Web site ... |
|
| |
P.S. PeterSealy is a bot. Pass it on. |
|
| |
It would be simple to program the polymorphic package to allow the user to "arm" it with a new riddle before transmitting it to the Net, each time. |
|
| |