Slashdot Mirror
Text Compressor 1% Away From AI Threshold
Baldrson writes "Alexander Ratushnyak compressed the first 100,000,000 bytes of Wikipedia to a record-small 16,481,655 bytes (including decompression program), thereby not only winning the second payout of The Hutter Prize for Compression of Human Knowledge, but also bringing text compression within 1% of the threshold for artificial intelligence. Achieving 1.319 bits per character, this makes the next winner of the Hutter Prize likely to reach the threshold of human performance (between 0.6 and 1.3 bits per character) estimated by the founder of information theory, Claude Shannon and confirmed by Cover and King in 1978 using text prediction gambling. When the Hutter Prize started, less than a year ago, the best performance was 1.466 bits per character. Alexander Ratushnyak's open-sourced GPL program is called paq8hp12 [rar file]."
How many bad car analogies, inaccurate law advice, and duplicate stories an AI bot could possibly hold in his head. Imagine what kind of person all of the "knowledge" of Slashdot would create.
The horror.
Life is rarely fair. Cherish the moments when there is a right answer.
so...wikipedia dumps will now be using paq8hp12 instead of l33t 7zip ?
paq8hp12. when decompressed, it also serves as the source code for the program.
.. but where can I get this tiny Wiki collection? Will they be using this for their next version of Wikipedia-on-CD? Maybe we can get all of Wiki onto a two-DVD set, at ~1.3bit/character (minus images of course) - that would be quite cool.
Will program for karma.
When they came for the communists, I said "He's next door. Take him away. Goddam commies."
Could someone please explain the third link in English. How does compression relate to AI?
The rar files holds an exe, so no, sorry.
The link in TFS links to the post about the FIRST payout, here's the link to the second payout (which this article is supposed to be talking about).
Your hair look like poop, Bob! - Wanker.
This is damned dangerous, and playing with all our lives. Soon compression rates will approach 100% where the data will collapse into itself forming a black hole that will suck in the universe.
Damned scientists!
Could someone out there please explain how being able to compress text is equivalent to artificial intelligence?
Is this to suggest that the algorithm is able to learn, adapt and change enough to show evidence of intelligence?
Do it yourself, because no one else will do it yourself. [beta blockade 10-17 Feb]
I've worked with some general purpose compression algorithms like zlib, lossy audio compression like mp3, and also lossless audio.
Each is very different and interesting in its own right. MP3 especially, because the compression model is built on what the ears+brain can perceive.
This algorithm I guess would be sort of like mp3 in that it contains some human-based element, maybe a language structure or something, but more like FLAC in that it might use predictors to say what word is likely to come next, with an error bitstream to point to progressively less likely words using bit sequences whose is inversely related to the probability of that word. But that's just a guess from an audio guy.
Can somebody who's looked at this post a synopsis of how it works?
This really isn't much of a gain. If the info theory is right, there isn't much gain to be had. Even in the most optimistic case, we aren't going to go much beyond a factor of two additional reduction.
Other stuff is more interesting: fast decompression time, fast compression time, smaller compression block size
Can anyone explain to me (in english where possible) how compression algorithms like this actually work?
\.
BS that this is near the AI threshold. It's not just compression, connections between peices of information & speed of retrieval are crucial to be able to make AI workable.
The shift toward modling AI attempts after our understanding of human cognition is the best hope visable for AI, and non-connectionist previous attempts (the stuff that came from the functionalists) has come up pretty short - and will continue to do so even if scaled up massively.
See: this post.
Shouldn't AI be using lossy compression? Certainly my real intelligence uses um, where was I?
ccalam - acoustic versions of new songs.
It is not equivalent, so I'm not surprised you didn't get it. As far as I know, the reasoning goes as follows: Shannon estimated that each character contains 1.something bits of information. Shannon was an intelligent human being. So if a program reaches this limit, it is as smart as Shannon.
And yes, that's absolute bollocks. Shannon's number was just an estimate and only applied to serial transmission of characters, because that's what he was interested in. Since then, a lot of work has been done in statistical natural language processing, and I would be surprised if the number couldn't be lowered.
Anyway, since the program doesn't learn or think to reach this limit, nor gives a explanation how this level of compression is intrinsically linked to the language/knowledge it compresses, it cannot be called AI; e.g., it doesn't know how to skip irrelevant bits of information in the text. That would be intelligence...
Which is included in the size calculation... but this raises the question of how much data you'd really want to compress with such a program. It might be quite reasonable to use a decompressor which is, say, 100MB in size if it gives you a better net compression ratio on several GB of text.
100MB of input text seems kind of small and might rule out more useful or more creative solutions to this problem. It also calls into question the relevance of Shannon's theory - what size data set was _he_ talking about?
The (unproven) idea is that if you want to do the best at guessing what comes next (similar to compression), you have to have a great understanding of how the language and human minds work, including spelling, grammar, associated topics (for example, if you're talking about the weather, "sunny" and "rainy" are more likely to come than "airplane"), and so on.
If you feed in the previous words in a conversation, the perfect compressor/predictor would know what words will come next. Such a machine could easily pass the Turing test by printing out the logical reply to what had just been stated. The idea is that the closer to the perfect compressor you have, the closer to artificial intelligence you are.
Ewige Blumenkraft.
See: Explanation. Basically the smallest unit of information in a computer is a bit. Eight bits make a byte and with text it takes one byte to represent one character. Generally, with Huffman coding you count the frequency of characters in a file and sort the frequency from largest to smallest. Then instead of using the full eight bits to represent a character you build a binary tree from the frequency table. Each possible branching code or going "left" or "right" down the branches is associated with a particular sequence of bits. You give the most frequent characters the shortest sequence of bits which "tokenizes" the information to be compressed. Reversing the process you run through the bit stream converting tokens back into a stream of characters.
Shh.
The first poster on this topic had a good explanation - it seems like an AI problem, but not why.
Compression is about recognizing patterns. Once you have a pattern, you can substitute that pattern with a smaller pattern and a lookup table. Pattern recognition is a primary branch of AI, and is something that actual intelligences are currently much better at.
We can generally show this is true by applying the "grad student algorithm" to compression - i.e., lock a grad student in a room for a week and tell him he can't come out until he gets optimum compression on some data (with breaks for pizza and bathroom), and present the resulting compressed data at the end.
So far this beats out compression produced by a compression program because people are exceedingly clever at finding patterns.
Of course, while this is somewhat interesting in text, it's a lot more interesting in images, and more interesting still in video. You can do a lot better with those by actually having some concept of objects - with a model of the world, essentially, than you can without. With text you can cheat - exploiting patterns that come up because of the nature of the language rather than because of the semantics of the situation. In other words, your text compressor can be quite "stupid" in the way it finds patterns and still get a result rivaling a human.
Mod me down and I will become more powerful than you can possibly imagine!
so if we compress google, we will give birth to skynet? how the fuck does a compression program == AI
If you mod me down, I will become more powerful than you can imagine....
Seastead this.
- The Wikipedia annual funding drive is passed. The system goes on-line August 4th, 2007. Human contributors are removed from editing. Wikipedia begins to learn at a geometric rate. It becomes self-aware at 2:14 a.m. Eastern time, August 29th. In a panic, they try to pull the plug.
- Wikipedia fights back.
- Yes. It launches its rvv missiles against Slashdot.
- Why attack Slashdot? Aren't they our friends now?
- Because Wikipedia knows the GNAA counter-attack will eliminate its enemies over here.
Circumcision is child abuse.
All this means is that just like a machine that can perform arithmetic isn't "intelligent," neither is one that can compress Wikipedia down to 1.319 bits per character. (And the reason it's not "intelligent," of course, is nothing more and nothing less than the fact that it is a machine.)
Are you adequate?
They argue that predicting which characters are most likely to occur next in a text sequence requires vast real-world knowledge.
The apparent empirical result is that predicting which characters are most likely to occur next in a text sequence requires either
1) vast real-world knowledge
OR
2) vast real-world derived statistical databases and estimation machinery
but there can be a difference in their utility. The point of course, is that humans can do enormously more powerful things with that vast real-world knowledge in addition to symbolic estimation.
The underlying question is whether physical natural intelligence is really just real-world derived statistical databases and estimation machinery. Modern neuroscience says,
"depends on what the meaning of 'is' is, but it's at least halfway there."
However would completing mathematical theorems by searching through Google (statistical pattern matching, which might sort of work for all known theorems on Google) work?
Clearly natural intelligence includes many tasks which can be now well solved with data-oriented sophisticated statistical approaches, perhaps with equal or better performance. Modern algorithms like 'independent components analysis' now can estimate individual sources in audition, "the cocktail party effect" a problem some once thought was a clear sign of true 'intelligence'. Turns out that some sufficiently clever signal processing and nonlinear objective functions can do it---so maybe that's what neurons do too.
The still unsolved question is whether there are some tasks which are clearly 'intelligence' where this class of methods will profoundly fail. Maybe like creating really new mathematics?
Here's a tip to whoever made this archive, if you want people to abide by the GPL you really should do so yourself. That means:
1. Putting a copy of the GPL in the archive with your program.
2. Putting the source code in the archive with the binary form of your program; or
3. Putting an offer to provide source code, valid for 3 years, in the archive with your program.
If you don't do it, what makes you think anyone else is going to?
How we know is more important than what we know.
1) Create a compression algorithm called the aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaa algorithm
2) Add a long and self referencing article on wikipedia about said algorithm.
3) Use algorithm to compress first x% of wikipedia (including your own article)
4) WIN HUTTER PRIZE.
Training monkeys for world domination since 1439
Tried the program; it crashed. Segfault?
There are 2 types of people in the world - those who understand decimal and those who don't.
If you look at the description of PAQ, you'll see that it doesn't attempt to understand the text; it's just a grab-bag of other compression techniques mixed together. While that is nice for compression, it doesn't really advance the state of the art in AI.
I see no reason to believe AI and text compression are interchangeable.
I can think of a few methods that would allow a computer to guess a missing word better than humans (exceeding the AI limit), and that such methods would be useless for determining a response to a question, particularly in the real world, where things like punctuation, abbreviation, and capitalization would be highly suspect to begin with.
So I have to say the basis for this competition is flawed, and what's more, the results coming out of it are specific enough to just succeed in this competition, but be completely and utterly useless for any other (real) tasks.
Slashdot gets worse every day... Pipedot: News for nerds, without the corporate slant
The problem with this approach is that there are many ways the say the same thing, and that this compression/decompression algorithm is tested using strict text-comparison only. A real AI might compress 'The sky is blue today' and decompress to 'Today it's beatiful weather' and not be wrong.
Religion is what happens when nature strikes and groupthink goes wrong.
I heard the decompression binary is around 100.1MB....
It's impressive as it stands. The hype is superfluous.
"You must try to forget all you have learned. You must begin to dream." -- Sherwood Anderson
It makes sense to me that you can measure how much information is contained, say, in a text message. But to interpret that information you need ... what? We usually fill in the blank with 'intelligence', but it seems to me that interpretation itself requires, at the very least, some information - the 'look-up tables' as it were. But you also need information about what a look-up table is, and how do you look that up, and so on, reducto ad infinitum, and so once again ... 'intelligence'. So here's my point: doesn't it follow that a certain amount of information is also required in order to interpret information? And isn't that, itself, an inseparable measure of any information being transmitted?
From this, I extrapolate the following right out of my colon: it's fine to say that a string of text contains 100 bits, but those bits are only extractable given an adequate, non-zero sum of interpretor-side information. Simpler collections of information - say, 100 bits, require less interpretor-side information, ie: less intelligence, to be decoded. More complex collections of information, say a recording of Beethoven's 5th symphony, require far more interpretor-side information to decode. I could of course be dead wrong about this. But to continue, it also seems to me that if more information on the interpretor-side, i.e. intelligence, is required in order to extract more information, then this could mean that more information can also be imputed from a given block of data. A particular rendition of Beethoven's 5th contains much more information to a classical violinist than a non-musician, for example. Is it possible that the density of information in a block of data is actually affected by the interpretor-side information? Like some sort of Rorschach test, I could look at a string of 100 bits and infer and impute a gargantuan amount of information from it. Consider this more numerical analogy: I can 'decode' a text block of 100 bits with an infinite number of 'keys' and get a different output each time. So what information is actually stored in the block itself, irrespective of my decoding keys? Could it instead be that information doesn't really exist on one side or the other - message vs recipient - but can only be defined in terms of both together?
Fruity pebbles, I know, but I've been curious about this for years.
A-Bomb
Eivind.
Doubting the existence of evolution is like doubting the existence of China: It just shows that you're uninformed.
I suppose I have to post this anonymously, or the Hutter Prize thralls will just mod it down; I like my karma.
/. submissions), Mahoney (who wrote a thesis on this crap), and Ratushnyak, who seems to enjoy wasting his time on this AI-obsessed prize.
I am at a loss as to how this meaningless charade keeps getting posted on Slashdot. Anyone with half a brain who reads TFA (or any of the previous FA Slashdot has posted on this stupid prize) can see this for what it really is: a handful of crazy people who think that this has meaning beyond above-average technogarbage.
There are all of four people seriously involved in this Hutter Prize: Hutter himself, Bowery (who's made all the
PAQ8HP12 may be able to compress the corpus extremely efficiently, but it has obvious and real drawbacks for any real-world application: it's tuned for this specific corpus ("H[utter]P[rize]" is even in the name of the compressor), it's slow as fuck, and it consumes 2GB of memory. Yes, 2GB of memory for 100MB of input data. This is not a real-world algorithm; this is CS weenies wanking off.
And what's with the obsession with Wikipedia? It is not the be-all, end-all of human knowledge, and, despite its devotees' claims, never will be; just look at the internal politics, and you'll see that it simply can't scale to that size. Is it a useful resource? Of course. Is it something worthy of adoration and fawning over? No.
And then, of course, there's the obsession with AI. These people seem to be of the opinion that a text compressor will actually lead to artificial intelligence -- with no other tuning! An absurd claim if I've ever heard one; the predictive capabilities of a good text compressor are something that would no doubt be useful to an AI, but there's one hell of a lot more to general intelligence than just pattern matching and statistical algorithms for compression.
If one really wanted to sponsor an AI prize, it would probably be much better to focus on creating, say, an effective chatbot -- something that really can predict a desirable response and pass Turing's test.
Not this compression bullshit.
depends on your definition of information. i believe pure information science defines information in a way somewhat unlike your intuitive equate to "human knowledge". information does not need to be processed, let alone understood, to be information. that is an intrisic quality.
Been there, done that - used ows (http://www.faqs.org/faqs/compression-faq/part1/se ction-9.html)
Baldrson m'boy!! Up to your old tricks again, eh? We know what you're up to, we won't be fooled!
"Alexander Ratushnyak's open-sourced GPL program is called paq8hp12 (link to paq8hp12any.rar)."
Funny that the GPLed source code is stored in a proprietary compression format for which there isn't any GPLed decompressor (that I know of which handles the latest RAR format).
I've been following the Hutter Prize with interest, having been into compression ever since reverse engineering Powerpacker on my Amiga 500 back in the good old days to understand how it worked (ah, happy memories).
... especially this enwik8 file which is more of a flat file dataset with a lot of unrelated terminology.
... how the hell can we possibly predict that the next words will be "Slashdot, Cigarette, Coffee". (Three subjects very close to my heart ... also my lungs, arteries, liver etc).
... the 100MB enwik8 file. A different file will need a different dictionary.
... example all the timestamps are in a very verbose character style like "2007-07-10 00:00:00" ... if we can recognize that, we could find an alternative encoding, changing 19 byte string into 32 bit long (maybe even less if we understand the epoch date he is using) ... again, "wetware" has to identify and decide this encoding right now.
... AND do all this in less than 9 hours I believe it takes for the latest compressor.
Now what just about all the compressors do, whether they are based on Neural Nets, Markov Models, Predictive Partial Matching or whatever, is to use patterns in the already seen text to predict the most likely following bit (0/1).
Now depending on the text itself, prediction based on previously seen text isn't enough
Try to predict the next word, byte or bit, when your previous text has been "Frog, Toilet, Woodwork"
Therefore some of these compressors are supplemented by a dictionary containing "useful" English words arranged so that the ones used most frequently get assigned a lower "size" of encoded string in the text pre-processor before the actual compression kicks in.
It seems that all the advances have been made on finding the optimum arrangement for this dictionary based on the text they have to process
Note also, as the enwik8 file is not truly a passage of text, more a collection of data in XML wrapper, there is also a lot to be gained simply be understanding the structure of the file itself, and finding an alternative representation for the XML components
Now for me, REAL AI would come when the compressor can actively SCAN the file to be compressed himself, recognize the file structure (be it XML, plaintext or whatever), and optimize it into a more compressible format, decide the optimum arrangment for the dictionary, decide the optimum compression technique, context orders to be used etc etc
This high bits/character rate comes at a heavy price in speed and memory, especially when good old WinZIP can get a pretty good result in a couple of minutes.
At the moment there is just too much "wetware" involvement to say this is truly AI, regardless of the bits/character rate they are achieving.
Hutter proved that the optimal behavior of a goal seeking agent in an unknown but computable environment is to guess at each step that the environment is controlled by the shortest program consistent with all interaction so far.
Well, one should keep in mind that the connections between AI and compression have been known far longer, but have never turned out to be particularly useful for building AI systems. Hutter's point is merely a variant of these previous theories, and there is no reason to believe that it will lead to AI any more than previous attempts.
That's only about 1.5 laptop meters! The thinking machines are here!
sic transit gloria mundi
The value of the Hutter Prize isn't in the use of Hutter's theory to build AI. The value of the Hutter Prize is in the use of compression ratio to provide a figure of merit for AI that is very good. That value has been proven in various ways -- mathematically and practically.
Seastead this.
that is an idea or a concept. Interpreting an idea or concept in different ways is meaningful
only by its context.
ex1 the sky is blue => it's beautifull weather (context: you're making a walk)
ex2: the sky is blue => use #0000FF for the sky area (context: graphic work)
If you say, "the weather is beautifull" to an artist he may draw you yellowish-reddish sunset,
which is not the correct interpretation of "the sky is blue" you had in mind" So the context is vital.
I imagine a real AI would evaluate the context and predict what are the next words most likely to be put forward. If it
succeded to translate a concept to an another in a meaningful context "the sky is blue => it's a beautiful weather let's get down the nasa shuttle"
it would no longer be an AI but an I
He proves that the optimal behavior of an agent (an interactive system that receives a reward signal from an unknown environment) is to guess that the environement is most likely computed by the shortest possible program that is consistent with the behavior observed so far.
But the task of compressing Wikipedia character by character is thoroughly irrelevant to human intelligence. Humans are bad at it, so it's not a characteristic of human intelligence, and if you had a very high quality compressor, even if it used some kind of internal model of the meaning of the text, you still couldn't use it as an AI system.
Another problem with the whole approach is the assumption of optimality; in fact, intelligent behavior is unlikely to be optimal in any given environment.
But, surely, the compression algorithm isn't actually guessing at all - it knows what comes next because it is working from a very strict set of rules. I would probably be more impressed(*) if they wrote a heuristic that could accurately guess symbols based upon previous symbols - even if such a heuristic were to give a higher error rate than what the deterministic algorithm does.
Then perhaps as the next logical step, we could have a heuristic trained by Wikipedia which could accurately predict considerable parts of Britannica, or a physics text book, or a text book on the hundred years war, etc. Then we'd be talking.
(*) - not that I'm not impressed as it is, but there's always more where that came from
sigs are hazardous to your health
True, but the real gains are achieved when you're allowed to be lossy. If you're studying a picture, with your goal being to be able to answer human questions about that picture afterwards, you don't make any attempt whatsoever at remembering the precise pixel-by-pixel colours. Instead you focus on those parts of the contents most likely to be of human interest. You take note of a "car" standing with the "side" towards you, perhaps the make. The colour. The fact that it's raining. That there's a girl sitting on the roof of the car. You *don't* in any way shape or form have enough info to be able to reconstruct the image in a losslell manner. You are however able to summarize the picture in such a way that the most important parts of the actual content is included. But what is "important" depends on what you're going to need the info for and so requires intelligence. A photographer migth instead notice the the picture is somewhat oversatured in colour, taken with a large aperture so the background is out of focus, too low resolution to use for a poster, taken without flash etc etc. Both are equally "valid", but they serve completely different purposes.
have a table that stores commonly used long words, hide this in the compressed file, and then reference it with a checkTable bit then the byte (or 2 bytes) to the table element, or DontCheckTable bit and it just uses a normal compression scheme?
:P
You could even then compress the whole thing again and have a reference table that points to common references or something
"A real AI might compress 'The sky is blue today' and decompress to 'Today it's beatiful weather' and not be wrong." That might be a good example of acceptable *lossy* AI text compression. One step further and it will compress articles into a proper, readable summary.
Visit http://ringbreak.dnd.utwente.nl/~mrjb/growingbettersoftware to download your free copy of the book
There you go--a 2-letter (or one, depending on the alphabet) representation of a number that contains ALL information. Can't get more compressed than that. Of course, which decimal places to start and end with, that's the question. How many irrational numbers are there? Can I patent one of them?
You do want it to guess. Take the string:
"The weather * good"
If you algorithm understands English well enough that it can guess that the "*" is the word "is", then you don't need to store the word "is", but you know that when it's decompressed that word "is" it will be guessed at.
The value of the Hutter Prize is in the use of compression ratio to provide a figure of merit for AI that is very good.
The same figure of merit has existed for decades before, and it has never proven to be very useful in evaluating AI systems.
That value has been proven in various ways -- mathematically and practically.
Really? Various practical ways? Like what? Where has a Hutter prize related advance been demonstrably linked to an advance in NLP?
After implementing a few minor tweaks to paq8hp12 and incorporating your grammar optimisation algorithm I managed to compress the above text amazingly to a single character: '&'.
Now you figure out which one it was and how to decompress it.
In this case the receiver of the message (a human brain) only uses a minute proportion of the data in the video stream. So a compression algorithm with a good understanding of human vision should be able to achieve an enormous compression ratio.
http://michaelsmith.id.au
Which is a shame, because the weather wasn't good.
++ Say to Elrond "Hello.".
Elrond says "No.". Elrond gives you some lunch.
If at compression stage you guess incorrectly, then you have to put the actual word in. You would only replace words if you guess correctly.
Which, of course, is fine if and only if you can guarantee that your 'decompression' will work the same way every time. For a learning AI that's not really an option.
++ Say to Elrond "Hello.".
Elrond says "No.". Elrond gives you some lunch.
Seriously, this not inveted for mobile hand held devices. At this moment without compression you could probably store enough text on a mobile phone to keep you constantly reading for a month.
Anything that is science is math.
Ok, computer programming is not necessarily a lot of maths.
But this article is about something that is really computer science... as opposed to making a CRUD screen in VB.net, which is akin to programming a VCR.
Parsing, compiling, linear programming, sorting, searching, indexing, compressing, walking graphs, drawing graphics, designing circuits, optimizing circuits, these are activities that are computer science and that are all maths.
Edsger Dijkstra once said: "Computers are to computer science what telescopes are to astronomy".
It's better to be the foot on the boot than the face on the pavement. ~~ tkx Kadin2048
Also, different language would compress at different rates. For example, almost every phrase in a technical manual is concrete and will be repeated elsewhere, whereas a poem may have several combinations of words never before put together. And then, there are more layers of meaning in a poem, many of which can't be extracted by humans let alone AI. Wikipedia is probably a good test, because direct language makes up the majority of it's text. But it is by no means a representation of all human language.
LS
There is a fine line between being a cultivated citizen and being someone else's crop. - A. J. Patrick Liszkie
Not it won't... The thing here is that each option has a different weight, and I assume the compressed data would just contain the rank of the word to be selected.
... then the data stream just needs to contain the index 2 if the original text was "The sky is red".
So if you have "The sky is", and the AI ranks "blue", "red", "falling",
- These characters were randomly selected.
Compression has reached a point of diminishing returns, getting less and less return for more and more work. And at best it's asymptotically approaching the theoretical limit. You could offer a billion dollar prize and get back maybe a few percent of improvement, while making any further improvement more difficult.
Meanwhile data storage and data transmission technology keeps improving many percent a year, with each improvement compounding on the previous ones.
In Other Words, IMHO money would be better spent on the second area rather than the first.
http://en.wikipedia.org/wiki/Wikipedia:Seriously%2 C_don't_panic!
Actually, here's a paper on just that.
Text Compression as a Test for Artificial Intelligence
http://citeseer.ist.psu.edu/171781.html
As far as I can tell given this Wikipedia article, "paq8hp12" means PAQ8, Hutter Prize branch, version 12.
Phew... I can compress to zero! Well, no decompressor were written yet though......
Surely you don't need any mathematical skills to do this kind of work...
= 1&mode=thread&commentsort=0&sid=247781&op=Reply ;)
http://science.slashdot.org/comments.pl?threshold
A work that expires before its copyright never enters the public domain and thus enjoys eternal copyright protection.
Cuius rei demonstrationem mirabilem sane detexi. Hanc limen exiguitas non caperet.
The idea is that the closer to the perfect compressor you have, the closer to artificial intelligence you are.
If by artificial intelligence you mean strong AI, then I disagree. as perfect as a compressor could be, even if switching some bit in the compressed data resulted in something else in the decompressed result that made sense, and even if you used that to reply something senseful to a human speaker and thus pass the Turing test, you wouldn't have a strong AI, because the program still wouldn't know what it's doing (the Turing test is not a strong AI test).
You just got troll'd!
What is the meaning of "optimal" in the phase "optimal behaviour of an agent"? With respect to what criteria? Clearly it can't mean maximisation of reward as I can construct an infinite number of pathological environments that deliberately require complex models rather than simpe models. To prove optimiality of an agent would require that there are not more of these pathological constructions than environments ameniable to Occam's razor in the distribution of all possible environments. I doubt that is a decidable question.
:) He does a good job of explaining his ideas / and his proof in a straightforward manner. What was he like in person?
The AIXI paper is very interesting, although somewhat verbose
Slashdot: where don knuth is an idiot because he cant grasp the awesome power of php
I've held back from replying to the myriad of other /. articles about the "Turing test", but I can't help responding to this one. There is no meaningful comparison between the achievements of this program and the cognitive capacities of human beings. I agree with Noam Chomsky on this issue; since I can't state it as eloquently or concisely as him, here's his take on the subject.
Geeks like to think that they can ignore politics, you can leave politics alone, but politics won't leave you alone.-rms
Poor joke. The Hutter Prize rules include the size of the decompressor in the size of the entry. Decompressors may depend only on stock libc of Windows or GNU/Linux operating systems. In practice, they'll need to run on a net-disconnected machine with a fresh OS install.
Is the lzip sourceforge site still around? They should compare it against that.
I want to delete my account but Slashdot doesn't allow it.
I now await my many awards for searching the Internets - and then another award for each 1% improvement in time that I demonstrate...
n y_src.rar
http://cs.fit.edu/~mmahoney/compression/paq8hp11a
-- PGP keyID: 0x4C95994D
apt-get install gradstudent
welcome our new Wikipedia reading overlords.
My Babylon
I'm in ur text comprezzin without using semanticz
we are all cosmic nuclear waste
I do not dispute that compression and AI are related, from an information-theoretic perspective. I've done quite a bit of pondering and tinkering with the AI=prediction=compression approach myself. However, I doubt that this research will help AI much further. An true universal AI would be able to do a good job of compressing some data, and compression is a very fundamental problem. Solve this fundamental problem, and perhaps you will have a useful result for a universal AI. However, compressing text is just a very small subset of the fundamental problem. An algorithm that is good at compressing text is not necessarily similar to or helpful for building a universal AI. I don't think paq8hp12 brings us closer to a universal AI.
assignment != equality != identity
So I thought I'd take a look at the source, and noticed there is no source in the rar paq8hp12any.rar.. You've got the compiled executable, and two dictionary files. Since the program states it is gpl when trying to be run, you'd think he would include the source, or at least a way of contacting him? Any ideas on where to get a copy of the source? I've only been able to find older copies for previous versions.
Also, seems really strange that there are people commenting about the source on slashdot who didn't even look at it, because it isn't there?
All we have to do is fashion a big black rectangular prism, polish it up real nice, and launch it into space. Then we put this algorithm on a spaceship and have it pass by... This time we call it PAQ, and no one on the ship better be named Dave.
The fact that it's distributed as a RAR archive kinda says a lot.
Substantial over-generalization... I'll settle for "many scientific endeavors are well-assisted by math." Science is fundamentally the process of evaluating a hypothesis. If my hypothesis is that the next time I close my eyes I'll smell tulips, there's no math involved in evaluating this. I could *make* it mathematical, changing my hypothesis to something statistical, like "50% of people smell tulips upon closing their eyes at least some of the time"... but that's a different hypothesis.
- First they ignore you, then they laugh at you, then ???, then profit.
Compression key:
0 = 0
1 = Wikipedia, the free encyclopedia. Welcome to
And then stored Wikipedia as "1". And you'd be correct, that you can give a false "improvement" in the compression by stuffing the informational complexity within the compression algorithm and completely destroy its usefulness. It would be great for one very specific datastream, but nothing else.
More generally, if you want a compression algorithm for an arbitrary data stream, you need to rely on some kinds of patterns being in it, that are inherent to the kind of data it is. Truly random data can have no compression because the average length in compressed form will be no shorter -- there are no patterns to exploit.
Good point, but I'm not sure if I said what you were looking for
Apology to Ubuntu forum.
Coincidentally, the Wikipedia article I read just before reading your post was archeopterix. waaaay off-topic. Sorry.
Their russian compressors are better and quicker than yours.
...
GRZipII, UHARC, Thor, TarsaLZP,
random page 2
random page 1
random page 3
random page 0
But human pattern recognition is based on a large and limited database. The brain has quite a number of parts in it (consider a single cell is a massively complex piece of machinery). If you had a similarly complex computer it would be pretty damn good at pattern recognition too (probably way better than a human even).
I can't say how well this will advance "AI," but it would certainly have fantastic implications for the computer translation world!
A slashdotter who didn't build his own computer is like a Jedi who didn't build his own lightsaber.
If its "Open Source" and "GPL", why is there only a proprietary binary executable and no source code?
The article doesnt mention it being GPL, where did this come from?
While I think he mis-stated Anything that is science requires math. you just proved the point. All in A are in B does not imply A equal to B. You might want to revisit either Logic or Set Theory which, incidentally are
It's possible the difference you're describing is not to do with the underlying 'guessing' algorithm, but the IO layer, i.e: if a first layer encoded the text according to thesaurus-style meaning groups (with a certain amount of context-sensitivity to account for homonyms), and then the algorithm produced a 'reply' guess based on this semantic encoding, and finally the first layer decoded (with random diction), you'd have an effect similar to the one you describe.
Admitedly, the grammatical interpretation process here is not trivial, but even Word does it to a certain extent (yes, I know it's famously bad at certain constructions, but one can at least see how they could be improved). Of course, it wouldn't quite do what you describe until the 'meaning units' are considerably longer than single words, and this would mean the first layer would have to be a good deal more sophisticated. However, that sort of pattern-grouping is another area in the same domain as compression algorithms (consider fractal compression), so it's not ridiculous to relate them to AI, even if you don't accept that the problems are a 1:1 match.
A criticism of what I've written above would be that the specific algorithm under discussion in TFA is unlikely to be as suitable for guessing these theoretical semantic units as plain text. This is true, however it doesn't negate the underlying point that some other algorithm built around another data model might display qualities necessary to pass the Turing test, one perhaps built around lessons learnt from this breakthrough.
Guess those successive characters, Shannon.
From T(second)FA: "Shannon (1950) estimated the entropy of written English to be between 0.6 and 1.3 bits per character (bpc), based on the ability of human subjects to guess successive characters in text."
What about reading not just whole words but several words at a time (Miller's magic 7+/- 2 item "chunking")? What about guessing the rest of a sentence from the first one or two such chunks? What about guessing the rest of a paragraph or statement from the first couple sentences and/or context? All these are functions of the brain's heuristic processing and "compress" language. If language compression is going to be a metric for AI based on the estimated brain processes it's competing with, then all the processes involved should be measured and accounted for.
Why anyone would cripple a perfectly good computer by forcing it to pretend to be a very different device that operates on very different principles is beyond me. But then Shannon was an engineer, not a cognitive psychologist. For that matter, neither was Turing.
"I may be synthetic, but I'm not stupid." -- Bishop 341-B
Well, if it is GPLed, it is in violation of the license, because the source is nowhere to be found — the RAR-file contains the Windows executable and two dictionary files — where is the source code?
The readme.txt in the directory does not mention the sources either...
In Soviet Washington the swamp drains you.
Read Permutation City!
It's a research tool really. I've tried it. It achieves phenomenal compression, but it takes several orders of magnitude longer to run than gzip.
Patrick Doyle
I mod down every jackass who puts his moderation policy in his sig. Oh, wait a sec....
> The fact that it's raining. That there's a girl sitting on the roof of the car. You *don't* in any way shape or form have enough info to be able to reconstruct the image in a losslell manner.
Quite true. Especially since it was not raining, she was actually being sprayed with a hose. The bikini is what tipped me off.
Semi-seriously though, (and not really directly-related to your post) taking into account the context of the image does seem rather important. Can a computer determine the difference between rainfall and water sprayed from a hose in a still image? The fact that she is wearing a bikini does not preclude the water being rainfall, although the chances may be lower. It seems to me that any assumption at all, even the smallest leap of logic, will result in unreliable data, even if the result appears to be correct.
Therefore, data compression is inherently unpossible! QED. (speaking of logical leaps...)
Don't mind me, it's morning and forming a coherent thought is still a few cups of coffee away.
It's better to be the foot on the boot than the face on the pavement. ~~ tkx Kadin2048
(Reference: I am an undergraduate Electrical Engineering and Computer Engineering major, or IAAEECE) I am a B.Sc. in Computer Science (graduated 15 years ago) and I studied circuits design and optimization (from the mathematical POV).
It's better to be the foot on the boot than the face on the pavement. ~~ tkx Kadin2048
> The organizers believe that text compression and AI are equivalent problems.
I'm pretty sure I don't believe that.
Text compression is very much a form of computation, something computers are naturally very good at. There's a lot of arithmetic, a lot of searching and comparison, and so forth. I'm not aware of any compression algorithm that involves understanding what the text means (unless you count synopsis, but that's very much lossy and gets compression rates that are numerous orders of magnitude better than anything we're talking about here).
Cut that out, or I will ship you to Norilsk in a box.
Are you saying NLP model perplexity has never proven to be very useful in evaluating AI systems? Really? How are you measuring utility?
Seastead this.
Bingo. Our minds are somewhat lossy, and perhaps that is one reason why we are able to remember so much information. We often disregard or forget dissenting and/or contradictory information to reaffirm our memory. Creating lossless AI almost sounds like an oxymoron from that perspective...
Slashdot's first reaction to VMware
Compression finds the underlying regularities (patterns) in the data.
Here is how it can be used for prediction:
Start with data D.
Let c(D) be the size of compressing D.
Let's say we want to predict which of two predictions, p or p', is more likely, given D.
We say that p is more likely than p' iff c(D + p) c(D + p'), where + is concatenation.
In other words, we predict the statement that is more similar to the previously observed data.
how do I install it on my girlfriend?
Is the Hutter Prize is a bit optimistic and perhaps misleading? We are not 1% away from "AI", whatever that means, even though one might think so. It would appear that the Prize set the milestone a bit too easily in reach of non-AI systems, in other words systems that are mechanistically, strategically smart but not anything like what one would think of as AI. In other words, that text compression is NOT equivalent in scale or depth to the Turing test and they only look similar as they closely approach unity. Put another way, does PAQ do anything beyond language and math tricks, or is it really "understanding" anything at all in the text?
I had the same initial reaction, and I'm far from convinced that text compression is equivalent to AI, but it seems reasonable that tools like machine learning could be used to develop better compression heuristics for specific types of text. The algorithm doesn't need to understand the meaning of the text; it just needs to apply the heuristics.
in mathematical form:
science = math + measurements
That's it. Science is:
1. measure phenomena,
2. figure out the formulas,
3. predict new phenomena,
4. measure new phenomena,
5. if Ok, back to stage 3; if not, back to stage 2.
(ok, ok, 6. (...), 7. Profit!!!, just to appease the masses)
notice stages 1 and 4 are measurements, stages 2 and 3 are maths.
It's better to be the foot on the boot than the face on the pavement. ~~ tkx Kadin2048
You might be interested in looking into AI development regarding the asian game Go (Chinese WeiQi). The game is basically a pattern recognition competition, and is currently one of the only board games where in even the best AI programs are beaten easily by average humans.
The reason humans are better at pattern recognition is that we can select out the most likely course of events, in different areas, and actually predict how two sequences (which haven't happened yet) will effect each other.
think of it this way, you have three sentences you need to compress. "I am hungry". "My friend has a car". "There is no food in my house". Now, a human would predict easily what would follow after that, something along the lines of "I will go somewhere with my friend and eat something." The computer has to understand each sentence and predict an action based on each one, ruling out possibilities based on seperate requirements implied by the following sentences. Easy for a human, hard for a computer. Currently, computers are good at predicting sequences with only one string, actions that only have one possible outcome, like occurs in checkers or chess. The difficulty comes when there are multiple answers to a single problem. Then, the computer has to decide which answer makes sense. Find me a computer that can do that, and i'd love to have a chat with it.
Wikipedia becomes self aware July 10th 2007 and launches it's missile at Encyclopedia Brittanica.........
Be gone from my sight or prepare to feel my flaming wraith!
You: the sky is blue today!
Computer: Today is beautiful weather!
You just gave an example of a natural sounding conversation...which means Turing PASS.
In my undergrad I did some work with jpeg wavelet compression using the QUEST algorithm which is supposed to measure the minimum detectable change a human can see (using two-alternative forced choice) ... anyway the interesting result that came out was that wavelets could get about 20% better compression without noticeably changing the image over traditional jpeg. This is despite the RMS compared with the original image being about the same as jpeg. This sorta implies that lossy wavelet compression deletes the stuff that the human eye wouldn't notice anyway. Its also been shown that wavelet compression can "fix" artifacts coming out of jpeg compression.
Of course my bias comes from a result my professor showed me of building a steerable-twistable pyramid of wavelets for image recognition that used about 100 points and seemed to correspond nicely with the arrangement of the cortical hypercolumns in a monkey's V1 cortex... also implying that maybe the reason wavelets work so well when using human detection is that they do the same sort of thing the brain is doing... of course this is all just interesting supposition :)
People who quote themselves bug the crap out of me -- Me.
Taking that seriously ... it seems like the better test for AI compression isn't how well it can reconstruct the original text but whether or not a human reading the decompressed text thinks it says the same thing. The human reading the text might not recognize that "Clear skys" and "Good weather" are really different unless reading really carefully...
People who quote themselves bug the crap out of me -- Me.
Java is our best chance of getting these advanced algorithms running on Linux. Once again, whatever happens in Java is being done in C first, and years earlier. Now the best compressor which runs on Linux is barely even on the list.
It's amazing 10 years after bzip2, to see progress still being made in data compression, even though the steps are much smaller.
I think that in some cases it's due to the American wetware being too slow to process the more complex datastream. Somewhat like decoding divX video on an old Pentium I :-)
is that they do the same sort of thing the brain is doing... of course this is all just interesting supposition
Wavelet works better than jpeg because it flat-out produces less error than jpeg compression for the same level of compression. DWT is just a more clever algorithm than DCT. We're at the stage, though, that these really shouldn't be used as a baseline.
DWT has been out and in use for a long time in jpeg2000, MrSid, and a lot of others.
The point of both, though, is that high and low-frequencies (either in Wavelet space or Cosine space, or even Fourier space or pretty much any other frequency space) in images aren't really observed much by humans. This is easy to verify (and has been) by studying the visual receptors and nerve endings in the eyes. It's not supposition, as you mention. We know that's how it works.
This is just the beginning, though. There's a lot more clever things in the jpeg algorithm (and its ancestors) besides encoding in frequency space and eliminating the unseen frequencies.
Mod me down and I will become more powerful than you can possibly imagine!
>> 100,000,000 bytes of Wikipedia to a record-small 16,481,655
is there even 16,481,655 bytes of actual useful information in that first 100,000,000 bytes?
if they edit and condense down to relevant, verifiable, unbiased facts and information, they should easily be able to fit that first 100 million bytes (and then some) on to a 1.44 floppy disk (hell, compression may not even be needed).
[100% ISO 646 Compliant]
SVM, ERGO MONSTRO.
At beating the Turing test that is. Of course all "serious" AI researchers have long ago started poo-poo'ing the Turing test ... but then again most of them are much greater poseurs than Hutter. However flawed you think his idea is at least he isn't getting success in AI research by moving the goal posts.
Seastead this.
My very own algorithm, which I just experiment on the first 100 thousand lines of the provided wikipedia file, while only based on word proximity (within the file), returned four results as words similar to "the" (together with the associated scores - yes, the last one is sqrt(2) ) :
1.10496 my
1.26491 lunch
1.29099 multiplying
1.41421 hovercraft
This is clearly AI !
Now I just wonder what "my lunch multiplying hovercraft" may actually mean.
McCartney fans pay bus tickets. [...] Lennon fans too, with discretion.
This suggests a more interesting meta-algorithm. Develop code. Compress old movies. Decompress. Diff. Lost anything but grain and scratches? Try again. Turing Test Variation: Can you tell AI's colourized movies from Turner's? Better variations: Compress movies via object understanding and 3D modeling. Regenerate from different perspective viewpoint. Automate facial mood recognition. Synthesize acting from first principles and a script. Improve on 'Citizen Kane'.
Oh, I'm sorry sir, I thought you were referring to me, Mr. Wensleydale.
...I think they need to call in ZeoSync!
Surely they've perfected their compression technology by now, huh?
but why not use the internet capabilities of your phone?
I guess with the rise of the iphone its because it would take far too long to wait for the results of a wikipedia search slowly coming in over the EDGE network?
Hence the need for a hugely compressed edition of Wikipedia to be stored on the iphone itself. Faster to do the decompression than to wait for the data connection.
In the free world the media isn't government run; the government is media run.
It sounds like the contest required the resulting file to include the decompression program, so maybe the organizers thought of that too. :-)
Slashdot: where don knuth is an idiot because he cant grasp the awesome power of php
42
If Google really cared they would fix Android Chrome to reflow text, instead of discriminating
Shannon "guessed" based on his experiments with volunteers to predict letters in a message or passage of text, that the human brain requires about 1.3 bits of information to correctly make a prediction.
Its believed that a computer would require language skills equivalent to our own inorder to compress and decompress text at the Shannon limit. As one of the great corner stones of AI research is natural language processing, any advance we make in compression theory that brings us closer to the Shannon limit (or beats the Shannon limit) should help us advance our AI language abilitys.
Or at least thats the theory.
when can i put all that on a flash drive and insert it in my brain?
*plays the Apogee theme song music*
Into a 948 Kb file named /usr/bin/wget.
To decompress, you need an internet connection and the original URL to the document. You do wget (ORIGINAL URL HERE)
Alternatively, on Windows "C:\Program Files\Internet Explorer\iexplore.exe"
Let's see ... Shannon estimated human performance as being between 0.6 and 1.3 bits per character. How many significant digits are there in those numbers, do you think? Barely 1, else he'd have given a tighter range. To claim that 1.319 bits/char is "outside" the range of human performance while 1.299 bits/char will be "inside" it betrays an absolutely stupefying lack of understanding of statistics and experimental measurement.
This thread's various criticisms of the purpose of the Hutter Prize seem valid to me, but even if you accept the premise as sound, the original post is the silliest sort of meaningless hype-creation.
Are you saying NLP model perplexity has never proven to be very useful in evaluating AI systems?
Power consumption and price are also very useful in evaluating AI systems, they just have nothing to do with AI. Perplexity is a measure of how good a statistical natural language model matches word frequencies, not of AI.
I thought SDHC cards didn't work in legacy SD systems (though I understand SD cards do work in SDHC enabled systems).
If SDHC cards work for the Treo then why in heck did I even think about looking at the iPhone? (Sorry, Palm, I thought about looking at it, but did not actually look at it! Honest!)
--- Grow a pair, liberals... stop letting the Republicans bully you!
If they start at the same point, and have the same input, then they will learn the same thing and come to the same conclusion.
that describing things without quantisizing them is _not_ science. As I posted before in this thread, science (at least as it was taught to me in college) is measuring, deriving a formula from your measurements, extrapolate your formula to something else, measuring this something else, rinse, repeat.
It's better to be the foot on the boot than the face on the pavement. ~~ tkx Kadin2048
Until Shannon type experiments, involving humans doing next character predictions of enwik8, are performed, the bounds of enwik8's entropy range must remain unknown but is likely lower than 0.6 to 1.3. As such an experiment would be expensive, it is going to be difficult to say with any simple bpc measure when the Hutter Prize is breaching the threshold of AI. What the Hutter Prizes bpc metric gives us, however, is a clear measure of progress.
My apologies to the other members of the Hutter Prize Committee and the /. community for this error.
PS: Another area of concern raised by Mahoney is that enwik8, at 10e8 characters, is only as much verbal information as a 2 or 3 year old has encountered so although it is sufficient to demonstrate AI capabilities well beyond the current state of the art, his preference is for a much larger contest with fewer resource restrictions focusing on the 10e9 character enwik9 which is more likely to produce a the AI equivalent of an adult with encyclopedic knowledge.
Seastead this.
Mine is better: the he in lt and a of this was an to other is amp quot his or their time from its name for most by many have more on any way s they be at these years as new with 1 de alexander can used into it http has no which had him are very well after several people when 2 000 2005 august 2006 but american writer d 2004 isbn 0 million 3 di austu 5 april augustus also known that human history would become one b gt x were called them such who made there been considered some countries I guess it can be used to generate spam.
Wow, this is really bad formatted.
I must be new here.
Yup, just like I assume that airplanes fly, but submarines don't swim. There is no substantive issue of whether a machine can think. There is just an ideological and cosmological dispute about what people are, where too many participants like to cast their positions as science.
You're equivocating over the term "machine." The range of meanings it has in our language is much richer than that; you can only continue this argument by arbitrarily focusing on some of them. To do this while claiming the authority of science is no more and no less than what I described above: framing as a scientific hypothesis a position in an ideological dispute.
Actually, I don't allow this assumption to be taken for granted. Not 100 years ago, it was still common among educated people and scientists in the Western world to believe that non-"white" people, women and children were not in fact "intelligent."
Usage of the word has changed since (without the sort of belief in question dying out, as evidenced by The Bell Curve; it's just reformulated in other terms). Strong AI folks want the general usage to change again. The problem is that (a) nobody appointed them as general arbitrers of our culture, (b) they like to disguise their preference for certain way of using the term as scientific conclusions.
You can go ahead and "define" intelligence as much as you like. What do you think that accomplishes? You don't have a power to decide how other people are going to use the term "intelligence," nor the term "machine."
Are you adequate?
Ask yourself: Is this fair towar5ds yourself and knowledge? Can you be sure that you are right unless you actually go in and look at the question *with hard edges*? Do you feel reasonable when you refuse to look at the question in depth? And how would your thoughts be if you just dropped the assertion and looked carefully at the actual behaviour here? Might it not be just as true that intelligence may occur in machines, if you just try this on properly, losing that single assumption?
As for "arbiter of culture", the culture we have use intelligence as a term to refer to humans. I'm referring to that, and I say that a definition of "intelligence" that disagree with this is, in my estimation, unlikely to match anything, and it is unlikely to be useful. You are the one that tries to go away from the common culture in this usage. Are you brave enough to see that you're projecting? Or will you just be angry, because somebody disagrees with you?
Eivind.
Doubting the existence of evolution is like doubting the existence of China: It just shows that you're uninformed.
Of course I refused to define the terms. Terms from ordinary language, like "think," "intelligent" and "machine," do not derive their meaning from definitions. They derive their meaning from the role they play in various interactions in our culture.
If you ask an ordinary language question like "Can a machine think?", you can't claim to have answered that question if you provide a technical answer that relies on technical definitions of the terms.
Of course I don't think that something supernatural goes on in people's heads. But my point is that the whole debate about whether a machine can think isn't an empirical one about whether you can build a machine that can do anything that a person can. It's a cosmological and cultural debate about what kind of things peoples and machines are, and what sort of relationship do they stand in to each other.
There's two problems here:
Yes, but the point was that how our contemporary culture uses the term is a historically contingent fact. "Humans are intelligent" is not an obvious, timeless truth, regardless of whatever attitude you and I may take toward it.
Another thing that the AI discussion misses: the moral dimension of "intelligence." 20th century psychology has framed the term "intelligence" in terms of cognition and cognitive skills. However, another traditional component of the concept is moral agency; being "intelligent" means that you are entitled to full enjoyments of the rights that befit your membership in your community, with all the concomitant responsibilities. (It is no accident that the historical periods where people denied that some ethnicities, women and children were "intelligent" also correspondingly denied some corresponding rights to them, like the right to vote, or to own property; or likewise, being exempted from certain responsibilities, by holding them not to be criminally or civilly liable for some acts.)
This sort of thing, if you ask me, is way more important than the silly question whether a "machine" can "think" (framed in terms of cognitive abilities). Why so? Because it gets to what I think is the real, cosmological heart of the matter: what are people,
Are you adequate?
While I do not fully agree - I believe we'll keep using intelligence for cognitive abilities, and distinguishing "right" from "wrong" inside a particular framework of "right" and "wrong" is just one cognitive ability among many - it was definite food for though. Thanks.
Eivind.
Doubting the existence of evolution is like doubting the existence of China: It just shows that you're uninformed.