I work for a relatively successful software company. Recently, one of my co-workers joked that really, if you think about it, all we really do is transform data. I responded by nothing that's what every software company does for a living.
A word processor, for example, transforms your input to a visual representation, a storable format, and a printable one. Photoshop modifies image data with your selected graphical transformations. SETI software culls though radio telescope data, looking for (and hopefully reporting) patterns which would already exist. Even games simply transform your input, pseudo-random noise, and stored strategy and visual elements into a game-playing experience.
I just noticed a quote from one Doug Robertson which says the exact same thing another way:
"... no operation performed by a computer can create new information."
"Algorithmic Information Theory, Free Will and the Turing Test," Complexity, Vol.3, #3 Jan/Feb 1999, pp. 25-34.
Is this true? I haven't read his paper, and can't say absolutely, but based solely on my several degrees in Computer Science, and several decades' experience in the same area, it certainly seems to be true.
If so, this leads me to another question:
Is the human mind fundamentally like software?
This question can be asked in a number of different ways. "Can the mind be downloaded to software?" "Is 'the singularity' [where machines eclipse human intelligence] a real possibility?" "Can machines become self-aware?"
Either way, there seems to be a problem lurking here for materialists.
If the answer is "No: unlike software, the human mind does generate new information", then it means that AI projects, although certainly not useless, are ultimately doomed. You're trying to convert one kind of thing into another fundamentally incompatible kind of thing, like trying to map real numbers onto integers. It can't be done.
On one hand, if the answer is "Yes, the human mind IS essentially the same as software", then it raises a larger question: Then where does information originate? If computer software only transforms information, and that's true of the human mind, then something else must be supplying (or must have once supplied) the information in the first place.
Information is not mere data. A unformatted disk drive contains lots of data — a random collection ones and zeros: noise. Perhaps 160 gigabtyes worth of data, but zero information content. If we pretended that "formatting" a drive was setting all bits to zero (it isn't, but for simplicity's sake), then, after formatting the drive, it would contain a small amount of information, equivalent to this short program:
emit zero 160 billion times
Where did that information come from? It exists on the disk because of my intention to format the media. And that specific information came from the "format disk" command on my computer, and, before that, from the software developer who created that command. And undoubtedly (in a real-life example) he would have read that from a DOS, MacOS, or UNIX file system specification, which originated from some author somewhere.
Another important characteristic of information is that it has a specific encoding understood by at least two parties. This can be as simple as a number of lit laterns: "One if by land, two if by sea", or a human head motion — nodding "yes" (vertically, perhaps) or "no" (horizontally).
This seems to imply that information can be created from data. A series of random events isn't usually information, but it seems to me it could become information.
For example, let's say you and I were doing an experiment in probability. You sit on the other side of the room, repeatedly flipping a coin, and calling out the results. I carefully write your results down, for us to tally up later. The coin flip was certainly "random data", yet it became information through our shared intent to observe, transmit, and record it. In contrast, you may have been carrying that coin around for quite a while, setting down every night, and thus generating a "head" or a "tail" (a single 'bit' of data), which was meaningless.
So it seems to me that intention is an important element in converting mere data into information.
Getting back to my initial question, "Does the human mind create information?", it seems to me there are only two answers:
1. Minds do create information, unlike software. Thus minds and software are not convertible entities. Thus AI is dead (though perhaps still useful), and the materialist view of the mind is false. Thus also, 'information' in the genetic code must emanate from some original mind: "God".
2. "Information" does not actually exist, per se; it is illusory, an artificial distinction. Minds don't create 'information', or there's no real difference between 'data' and 'information'. This implies that intent, a key distinction which makes information, is also illusory.
Under the second option, we don't have to deal with questions of "information" being found in the DNA "code", because there is no such thing. Likewise, nothing special happens in our coin-flipping experiment: just as photons mindlessly transmit 'heads' or 'tails' to no-one in particular, even when nobody cares, so your calling out "head" or "tails" is equally meaningless, because, overall, meaning and intent (and thus 'information') are simply illusions. We are simply doing what machines or photons do, albeit in a much more complicated (Rube Goldberg-like) manner.
I don't see a third way out. It seems either we ask questions about information and its origins (seemingly implying a "God" of some sort), or we retreat to the belief that "intent", free will, volition, is illusory. Of course this also means "mind", morals, justice, and even 'meaning' (in even the most mundane sense of the word) are fundamentally illusory also. It also would seem to imply that a materialist lives in constant contradiction, at every single moment, to his own supposed convictions.
If you've read this far, I expect you might have a few questions or objections. For example:
"Doesn't this seem a tad irrational?"
Yes! But I would argue the source of that irrationality isn't from me. I'm simply filing the report, it seems to me, from the field.
"Can't we just say that DNA isn't really 'information' because it doesn't have any intent behind it?"
I sympathize with this argument (which fits within answer #1), but recall: How do we distinguish mind from non-mind? Isn't it by, as the "Turing Test" suggests, communicating with a mind? (That is, evaluating signals to see if they contain information, as SETI does?) Yet each time humans evaluate DNA, don't they conclude it contains "information", hence even atheists call it "code", "instructions", or genetic "units of information"? If so, they're admitting it sure looks like information to them. It passes, if you will, an inadverent Turing Test.
So Dawkins (or whoever) counters that this is only the appearance of intent. There's no mind behind it. Okay, but how do we know that? That seems to be an a priori religious assumption. (Which is fine, but not science, certainly.) But either way, we're reversing things: Before we started with apparent information, and worked back concluding it came from a mind; now we say we must first know if a mind is there before we can know if it's really 'information' or not.
Yet this also implies 'mind' is now a meaningless concept. What distinguishes mind from non-mind? Remember: the old test was that minds emit information. Now that criteria is gone: we start by assuming where minds are. So I can just as easily assume a pencil or rock has a mind as a human. In fact, there's no rational reason to assume humans have minds at all, since saying "they emit information" (in any form of that assertion) is no different than saying "they emit data". It's also just another way of saying: "I first assumed there was a mind in there." Sure, but it's equally rational to assume that for any object, or none.
But we'll give all that up to keep DNA from having 'information', right?
Or we'll live our lives in contradiction, saying or implying this is all true, but acting, with every breath, as if truth doesn't matter at all.
Theists sometimes assert that you can't have morals without God. Whether that's right or wrong (depends what you mean by "morals", I think): The longer I think about it, the more I come to the conviction you can't even have both rationality and knowledge of some very basic scientific observations without God.
If you think I'm wrong, and can demonstrate that, feel free.
Thanks for listening either way.
Hey Tim. This is an interesting topic, though the bulk of the issue seems to be just agreeing on common, objective terms.
"... no operation performed by a computer can create new information."
"Algorithmic Information Theory, Free Will and the Turing Test," Complexity, Vol.3, #3 Jan/Feb 1999, pp. 25-34.
Tim: Is this true?
A quick search didn't help me find his paper, but by 'new information' he seems to mean 'new axioms.' His argument seems to be that each axiom contains within it all the logical consequences of that axiom (which makes little sense to me, since adding a new axiom should create novel conclusions, right?)
I may very well be missing something huge by reading about this guy from a second hand source.
But is that what people normally mean when they refer to information? It seems like a fairly unique definition.
In fact, it seems strange to make definitive statements about a property that is so poorly defined or
that carries so many different definitions, and then assume that one assertion applies to a different definition or that formal
logic could be used. A or NOT A requires a rather strict definition of A.
It certainly seems possible for a computer to do statistical control of processes, a la Demming.
But would this be enough to count as 'information' for many people even though it does seem like a type of inductive
reasoning, the creation of a rule (a statsitical control limit) which was not known before could then be used to create more information.
I don't see any hard, practical barrier here, though we seem lightyears from being able to do this kind of thing in practice.
"No: unlike software, the human mind does generate new information", then it means that AI projects, although certainly not useless, are ultimately doomed. You're trying to convert one kind of thing into another fundamentally incompatible kind of thing, like trying to map real numbers onto integers. It can't be done.
Is the human mind fundamentally like software?
It seems people would much rather use computers as predictible tools than as unpredictible agents in the same way that we breed dogs to serve a function rather than to be wild (is it even possible to "breed" a dog to be feral?)
So the way that humans design programs and the way that natural minds evolve seem likely to be different because they aim at inherently different goals, regardless of the process or medium. At issue in most software design seems to be not "what is possible" but
"what do humans find desirable." Human minds have agency. People don't want computers with true agency. Not really. They want computers
that are extensions of their own will, that increase their own capacity to create new information or to act.
If computer software only transforms information, and that's true of the human mind, then something else must be supplying (or must have once supplied) the information in the first place.
To determine where new information is coming from, don't you have to be able to quantify the amount? As well as I understand the topic, that's a bit difficult.
Even so called "Shannon Information" is more about the capacity of an alphabetic string to contain information than the information actually contained in the string.
And don't all materialists and theists claim that the human mind either came from nature (evolution and adaptation) or God?
In either case, the human mind is being formed partly by an outside source. Further, I'd note that natural selection acts as an information filter. It seems like any system which produces information from nature begins by somehow filtering the data.
Tim:Getting back to my initial question, "Does the human mind create information?", it seems to me there are only two answers:
1. Minds do create information, unlike software. Thus minds and software are not convertible entities. Thus AI is dead (though perhaps still useful), and the materialist view of the mind is false. Thus also, 'information' in the genetic code must emanate from some original mind: "God".
2. "Information" does not actually exist, per se; it is illusory, an artificial distinction. Minds don't create 'information', or there's no real difference between 'data' and 'information'. This implies that intent, a key distinction which makes information, is also illusory.
What does 'actually exist' mean? Aren't we just looking for a predictive model? Technically speaking, do numbers 'actually exist?' Or are they just really useful illusions?
Also, isn't one step in the creation of any information the separation of signal from noise? can't we see that step, at least, occurring via natural selection?
I'm not a big fan of the Turing test, incidentally. It seems like one of those scientific 'bets' that stimulates thought but is a bit of a red herring.
There's no good reason I can see that machine intelligence needs to approximate human intelligence vis a vis a natural language conversation.
So Dawkins (or whoever) counters that this is only the appearance of intent. There's no mind behind it.
Perhaps the purpose of recognizing intent is to recognize a signal as an extension of some agent and model that agent's reaction. "The government that put the stop sign here no longer exists, therefore we don't need to stop."
"the computer won't let me in unless I supply a password." "Xerxes was an idiot for flogging a body of water. The water can't feel pain. (if the story is true)" This is a sort of "if it looks like a duck and quacks like a duck then it is a duck" type of understanding, an attempt to create a predictive model.
The notion of intent seems to accurately model genetic changes. I agree with your criticism. If the model works, use it.
If the predictive model breaks down because of the lack of mind behind a process (that hot babe you were chatting with online was actually a chatbot)
then maybe the criticism of whether information is the product of a mind has some relevance because it impacts how we create the model and how it is likely to function under certain conditions. A genome operates as if it had the intention of self perpetuation.
It seems a short step to actually ascribing intent.
Of course, all of this ignores questions like "can a machine be made to feel pain that we're morally obligated to respect." I have no idea how to begin to answer a question like that.
Hey Tim. This is an interesting topic, though the bulk of the issue seems to be just agreeing on common, objective terms.
"... no operation performed by a computer can create new information."
"Algorithmic Information Theory, Free Will and the Turing Test," Complexity, Vol.3, #3 Jan/Feb 1999, pp. 25-34.
Tim: Is this true?
A quick search didn't help me find his paper, but by 'new information' he seems to mean 'new axioms.' His argument seems to be that each axiom contains within it all the logical consequences of that axiom (which makes little sense to me, since adding a new axiom should create novel conclusions, right?)
I may very well be missing something huge by reading about this guy from a second hand source.
But is that what people normally mean when they refer to information? It seems like a fairly unique definition.
In fact, it seems strange to make definitive statements about a property that is so poorly defined or
that carries so many different definitions, and then assume that one assertion applies to a different definition or that formal
logic could be used. A or NOT A requires a rather strict definition of A.
It certainly seems possible for a computer to do statistical control of processes, a la Demming.
But would this be enough to count as 'information' for many people even though it does seem like a type of inductive
reasoning, the creation of a rule (a statsitical control limit) which was not known before could then be used to create more information.
I don't see any hard, practical barrier here, though we seem lightyears from being able to do this kind of thing in practice.
"No: unlike software, the human mind does generate new information", then it means that AI projects, although certainly not useless, are ultimately doomed. You're trying to convert one kind of thing into another fundamentally incompatible kind of thing, like trying to map real numbers onto integers. It can't be done.
Is the human mind fundamentally like software?
It seems people would much rather use computers as predictible tools than as unpredictible agents in the same way that we breed dogs to serve a function rather than to be wild (is it even possible to "breed" a dog to be feral?)
So the way that humans design programs and the way that natural minds evolve seem likely to be different because they aim at inherently different goals, regardless of the process or medium. At issue in most software design seems to be not "what is possible" but
"what do humans find desirable." Human minds have agency. People don't want computers with true agency. Not really. They want computers
that are extensions of their own will, that increase their own capacity to create new information or to act.
If computer software only transforms information, and that's true of the human mind, then something else must be supplying (or must have once supplied) the information in the first place.
To determine where new information is coming from, don't you have to be able to quantify the amount? As well as I understand the topic, that's a bit difficult.
Even so called "Shannon Information" is more about the capacity of an alphabetic string to contain information than the information actually contained in the string.
And don't all materialists and theists claim that the human mind either came from nature (evolution and adaptation) or God?
In either case, the human mind is being formed partly by an outside source. Further, I'd note that natural selection acts as an information filter. It seems like any system which produces information from nature begins by somehow filtering the data.
Tim:Getting back to my initial question, "Does the human mind create information?", it seems to me there are only two answers:
1. Minds do create information, unlike software. Thus minds and software are not convertible entities. Thus AI is dead (though perhaps still useful), and the materialist view of the mind is false. Thus also, 'information' in the genetic code must emanate from some original mind: "God".
2. "Information" does not actually exist, per se; it is illusory, an artificial distinction. Minds don't create 'information', or there's no real difference between 'data' and 'information'. This implies that intent, a key distinction which makes information, is also illusory.
What does 'actually exist' mean? Aren't we just looking for a predictive model? Technically speaking, do numbers 'actually exist?' Or are they just really useful illusions?
Also, isn't one step in the creation of any information the separation of signal from noise? can't we see that step, at least, occurring via natural selection?
I'm not a big fan of the Turing test, incidentally. It seems like one of those scientific 'bets' that stimulates thought but is a bit of a red herring.
There's no good reason I can see that machine intelligence needs to approximate human intelligence vis a vis a natural language conversation.
So Dawkins (or whoever) counters that this is only the appearance of intent. There's no mind behind it.
Perhaps the purpose of recognizing intent is to recognize a signal as an extension of some agent and model that agent's reaction. "The government that put the stop sign here no longer exists, therefore we don't need to stop."
"the computer won't let me in unless I supply a password." "Xerxes was an idiot for flogging a body of water. The water can't feel pain. (if the story is true)" This is a sort of "if it looks like a duck and quacks like a duck then it is a duck" type of understanding, an attempt to create a predictive model.
The notion of intent seems to accurately model genetic changes. I agree with your criticism. If the model works, use it.
If the predictive model breaks down because of the lack of mind behind a process (that hot babe you were chatting with online was actually a chatbot)
then maybe the criticism of whether information is the product of a mind has some relevance because it impacts how we create the model and how it is likely to function under certain conditions. A genome operates as if it had the intention of self perpetuation.
It seems a short step to actually ascribing intent.
Of course, all of this ignores questions like "can a machine be made to feel pain that we're morally obligated to respect." I have no idea how to begin to answer a question like that.
Posted by: Ryan W. on July 31, 2009 03:06 PM