Randall, I didn’t intend to make this a discussion about embodiment, nor is it my intention now. However I do not understand how meaningful data make it into these databases. I believe you understand that words are predefined by our experiences. Not explaining how these words obtain qualitative origin without the benefit of experiences leaves a lot to assumption. I think the schemas such as graphing add a layer of conceptualization without offering understanding of the fundamentals of conceptualization. To be useful tools they will still need what they attempt to replace and that is intelligence. Furthermore, using them in the hopes of establishing some direction and that this direction will make clear the goal of intelligence is also a broad assumption. It’s as if we’ve decided to build a building and without any planning and then we build a very complex dome. Perhaps someone not quite as bright other builder may come along and ask “Why no plan?” and get the reply “You just cant understand complex things.” I think the comments from both sides may be true, however are we any closer to getting the building done.

If efficiency is the focus of graphing then I believe the conceptualization of word level understanding to be at too high of a starting point. I brought this up at the meetup. The primitives do not seem primitive enough. With embodiment I can see the ability of natural selection to control the outcome. I don’t see natural selection at work in graphing or other such schemas. And, I see tremendous resources being swallowed up in their pursuit. The words being defined through experience seems to be a necessity at some point. After that further extrapolation is meaningless.

Now that we have picked apart the issues foundational to intelligence, lets re-visit your position. What I understand of what you are saying is that thinking systems, like the brains of biological creatures, lean heavily upon the fact that they are structured primarily by the bunched nerve terminus regions that in some way mimic the sense and motor locations throughout a highly specialized body plan. The brain, as dictated by the wiring scheme by which it communicates with the rest of the organism, must causally predict the physical differentiation of the body. Experimentally, this has been shown to be true (though it is important to remember that there is much of the brain that is not directly connected to sense and motor fibers). The resulting highly differentiated and non-random plot plan exists as valuable information in an of itself. That thinking would evolve towards an efficient use of this structurally significant information is more than a little likely.

I would be very very surprised if a word graph didn't also reflect (at some perspective or unfolding) the same sense and motor topology as found in the embodied brains of the people from who's writing the corpus is collected.

Remember that the word graph is hoped to contain equally pertinent topological information about many other aspects of the lives, interactions, and environments of people than can be found simply in a sense motor connection graph.

But the word graph is only one half of the equation. The other half is the algorithms that both build and crawl the graph. These algorithms will need to take maximal advantage of what the shape of the connection space says about humans and the world to which their written refers.

There will be much to learn by looking for the over-arching patterns that can be found in many connection graph dimensions. What is intelligence like when it isn't as body plan bound as the intelligence we animals have evolved? I am curious. But I must say that the embodiment to which you so often refer, that embodiment which is so literally bound to our corporal structure, is only one of and probably not significantly interesting of and by itself. All information, if it is causally sourced, will exhibit structures that shadow and predict pattern in the subject environment. The resulting graph will always define a type of "body" that is definitionally symmetric to the arms, legs, torso, and head that usually lays claim to its namesake.

Randall

If you are looking for the reason word graph proponents choose written language as source, the answer may be more simple than you expect… computers have been specialized to deal with text, text has existed for a long time, there is a lot of it, and it is readily available. One could conceivably do the same with video libraries, actuary tables, medical records, grocery store sales data, automobile traffic records, etc. It is just that it is easier, cheaper, and more productive to point graphing algorithms at text. And, we have written about more things than we have recorded in any other way.

But I think you are missing the richness of the information that can be gleaned from super-large data sets. For instance, lets say that the only source of data you had access to was an infra-red sensor floating above a city. Just by tracing the motion of hot points you would be able to build a fairly accurate street map of the city as well as floor plans of most of the buildings (where the walls are, where the doorways are. You could presumably even be able to collect the data necessary to understand the topography of the land and the height of each building. You could derive the population, where people worked, where they lived, how many shared each dwelling, even how long they lived and how old each person was. You could work out the familial and power relationship of each person to the other people with which they interacted.

This information would be every bit as reliable as the individual sources we would normally consult if we were asked a question about a city. Word graph people would even defend the notion that a single source (heat, text, whatever) would result in more illuminating, and accurate information in total as it would allow easy conversion and comparison between categories.

Randall

Randall, regarding your statement “I would be very very surprised if a word graph didn't also reflect (at some perspective or unfolding) the same sense and motor topology as found in the embodied brains of the people from who's writing the corpus is collected.” etc.

I understand how attractive it might seem to be drawn into such an intellectually challenging process. But the data isn’t raw enough to allow for the kind of decomposition that a creative intellect will require. (I tossed in creative because I think that’s the next step but the same could be said if you leave it out.) And the structure of the data would be randomized by the learning process, if learning is still included.

“But I think you are missing the richness of the information that can be gleaned from super-large data sets.”

This is true! I see them as flat. You mention the ability a simple “infra-red sensor floating above a city” would have by using these techniques and I have no problem with this. But, to describe their potential you state “you would be able to”, “. You could presumably”,” You could derive”. The point is that these techniques require interpretation, a homunculus. They, if packaged properly, will come to closely resemble intelligence and we would be intelligent to use them. But I see no possibility of autonomy here.
And we seem to be redefining intelligence to suit our current lack of understanding.

It appears to me Lex, that you are presenting a religious argument.

I have reminded you of the fact that structure and information are equivalent. I have shown that a word graph (or any data set) can express (losslessly) any physical structure. It is worth noting that the opposite isn't necessarily true – there are some data sets that describe structures that can not be built in the the physical world. The point is that data, architected as abstraction system, is more flexible a medium than the structure it abstracts – abstraction systems are by definition more flexible than the systems they abstract. So, it is perfectly possible to construct an information system that can simulate every structural aspect of any physical system. If a word graph can capture knowledge, it can capture any knowledge that can be captured by that physical system.

Your argument against this is beginning to feel like a tantrum. Over-protective. Bio-centric. Religious. "Word Graph" is a label that represents a general set that might better be shortened to "Graph" because it is not dependent upon or restricted by the type of data that is related by connections. As a result, the limits of the category "graph" must always be considered separate from any particular incarnation or instantiation. The brain, both structurally and behaviorally, is exactly a member of the class "graph", and as such, suffers the same dependence upon the accuracy and granularity and completeness of the data (sensory input) it is fed and has the capacity and ability to abstract. Graphs are not necessarily perfect (lossless). Graphs represent that class of grammars in which semantic relationships are encoded as connections between data points. It has been proved that graphs can represent with perfect fidelity, any logical statement.

So, the only legitimate argument one could make against graphs as sufficient foundation for abstraction processing has nothing at all to do with graph structure, and has only to do with the completeness of the source from which a graph is built. If you are arguing that a written corpus (no matter how large) is not sufficient source for processing at a level that would allow a similac of human cognition, than you are not arguing for or against graphs as storage structure or processing substrate. This is an important distinction. My problem with the anti-corpus augment is that it suggests that there is some magic invisible line in the sand, before which information is not rich enough for "creative" cognition, and after which, magically, it is. This argument hinges the very possibility of sentient cognition upon an arbitrary quantitative threshold. Such arguments have a tendency to show themselves eventually to be hubris motivated.

Nobody is saying that corpus graphs will magically produce intelligence (at least not at the level of human brains). A graph after all is simply a particularly efficient substrate for the traversal (parsing) of relationships. The same meaning that can be distilled from a graph can be derived from a raw data set… it just takes a lot longer and will involve at some stage of computation, the building of the same or similar graphs! Arguing that intelligence won't arrive directly from a graph is like arguing that humans won't arrive directly from cells. That isn't the position of graph builders. Graph builders say simply that graphs will have to exist as substrate for efficient building of the other higher order processes and structures from which cognition will precipitate.

Had you been tasked with the project of building biological beings, would you argue against the effort necessary to build cells simply because a cell by it self isn't cognate?

Randall

Randall

“Your argument against this is beginning to feel like a tantrum.”

I don’t understand what you mean here. But I’m not against these kinds of techniques used as tools. I do think we agree that there just doesn’t seem to be any reference points that would allow it to have an individual perspective? Regarding this “tantrum” feeling you get from me, I think you’re reading too much emotion in this. I am not “digging in” if that’s what you mean. I feel open to new thoughts and encourage this discussion. I just want to “get the building, built”. The problem is that I don’t hear anything but to keep trying the same old crap with modifications. It’s like Christians wanting to give Jesus a chance. They’ve been wanting that for 2000 years. How much of a chance does Jesus and this stuff need. From my perspective I’m only stating my point of view and this project seems to have a lot of ambiguity to it. I believe I have responded with very specific points and questions. If it is not clear what I want to know let me ask gain. To make it function autonomously, will this software schema while operating require what it is attempting to replace? I am not defending a position just asking a question. This, whether one likes it or not, is a question of how the machine gets motivated. If you don’t like to discus this in terms of biology, then what does that leave?

“I have reminded you of the fact that structure and information are equivalent.”

If you mean that all is perception, I agree. Are you also saying that because structure and information can be described in words the imagery that humans associate with these words serve no function? We dream in imagery - but so what? I have a rational for these kinds of behavior that integrates them into the whole, but my guess is - don’t we all? Right?

“Arguing that intelligence won't arrive directly from a graph is like arguing that humans won't arrive directly from cells.”

Plants also have arisen from cells. While understanding that systems have building blocks you fail to recognize the foundation necessary to provide the desired outcome. You mention that a graph is “simply a particularly efficient substrate for the traversal (parsing) of relationships”. To parse a statement by using words that have no association to imagery will be circular at best. The imagery from some form of sensory system is what gives meaning to the words.