Anagram

Semantic Halos

Hofstadter starts by stating that the foundations to perceiving patterns and ascribing rules about those patterns lie in the ability to recognize analogies and abstract resemblances. He continues into an analogy problem between American and English government structure, which leads into conceptual halos of words and concepts.

One concept or word has numerous meanings, each of which can fall across many different contexts. These concepts/words are not merely one idea but are composed of multiple ideas and are linked with other concepts, words, meanings, and ideas. In Hofstadter's example of English and German concepts for hard he concludes that context and culture play a huge role in determining the boundary of concepts. Ideas about language apply only in the context of the language being discussed. The concept denoted by hard in English or by schwer in German are not universal, but instead apply only in each language and are quite different when compared to another language or considered in another language.

This reminded me of a question I had about conceptual spheres after a discussion with my boyfriend on a topic we briefly discussed in class: chairs. His conceptual sphere of what constitutes a chair differed from mine, which led me to believe concepts cannot be concrete if what we associate with that concept or consider that concept to entail is not universal by any means. Professor Graci briefly mentioned an exhibit he attended in which there was a room filled with assorted chairs; I believe he said if it had not been noted beforehand one may not have known that each object was a chair. As I told my boyfriend about this he insisted that the majority of these could not have been chairs, especially if they did not fit into certain requirements which he gave me.
1) A person must be able to sit on or in it
2) It must raise one from the ground
3) It must have some sort of back support
...and I believe the list went on but my head began to slip into wondering why certain things would fit into my concept of "chair", which was pretty flexible, but they would not fit into his rigid concept of "chair". For instance, he would not consider a stool to be a chair, whereas I would. He would not consider an object to be a chair any longer if the legs were cut off and it were placed on the ground, whereas I would. I had an extremely loose concept of what a chair could be, simply an object made to be sat upon, a category to which other objects fall into (e.g. booth, stool, sofa, etc). We discussed the requirements for a computer to recognize a chair, and concluded that we both associated objects outside of the computer's conceptual sphere (e.g. a chair on the ceiling as mentioned by a fellow classmate when Professor brought up the exhibit). However, we could not come to a conclusion as to what fit into the conceptual sphere of "chair"; we simply could not agree on a rigid definition or idea. Once again, this led me to wonder how there could be stark differences in the idea of what constitutes or pertains to certain concepts.

"conceptual slippage" = "context-dependent tolerance of conceptual mismatch."

BACON

Chapter Four illuminates the argument presented in Preface Four, which stated that computer technology is not capable, or even near being capable of, human cognition. The omission of information, exaggeration, or misleading explanations of capabilities of computer programs, so that it is believed they are able to use some aspect of human cognition is called the Eliza Effect. Chapter Four goes further into detail on such misleading stories of computers and perception. In one case a program titled BACON is said by its developers to be able to deduct universal laws from "original data". It turns out the program is given all the data necessary to discover the laws; data and information that may not have been available or possible to know at the time of their original discovery. Chalmers et al. states that "[t]he program was given precisely the set of variables it needed from the outset (even if the values of some of these variables were sometimes less than idea), and was moreover supplied with precisely the right biases to induce the algebraic form of the laws, it being taken completely for granted that mathematical laws of a type now recognized by physicists as standard were the desired outcome." (p 178) This is a prime example of the problem with all computers in perception. It is necessary to give the information to the computer beforehand in which it can deduce the answer from. No program runs on knowledge that is not supplied. It can assign new variables but it has to have been told in advance some new variables will be creased that it will have to assign some variable to, it has no autonomy in the matter. It cannot think on its own. It needs a programer who has supplied the information necessary.

Although this does make me question the point of Hofstadter's programs at all. His work focuses on mimicking human cognition with his programs, yet even when the computer finds, for instance, underlying patterns by using methods similar to human cognition, the computer is not aware. Descartes, James, and many others would say that without being aware of oneself and cognition nothing more can exist besides an inanimate object. However, by modeling human cognition with Hofstadter's programs we may be able to better understand the functioning behind cognition. Maybe it is not Hofstadter's goal to create a self-aware computer, but just a computer which models some aspects of human cognition.

Eliza Effect

So I was I was buddies with Hofstadter so I could recommend this great movie to him and chat him up about misconceptions about artificial intelligence, specifically the Eliza Effect, which he discusses in Preface 4. I just watched this great movie, The Surrogates (which the worst aspect of the movie was the leading character, Bruce Willis). It is the first movie I have watched about computers in the future where it didn't focus on the intelligence, or human-like capabilities in thought of computers. There was no jumping from modern technology to a future of computers who completely understood concepts and could think like a human; it did not question the morality of artificial intelligence. There was no good versus bad computers, it was all about brain-computer interaction. The machines did not understand concepts or think on their own, they simply moved and obeyed commands from their human counterpart.
Point of it all, it was refreshing to see a movie about computers of the future which did not focus on computers that are just as capable in thought as human, considering we are far from making computers understand the most, trivial to humans, but complex facts for computers. As we discussed in class, computers cannot identify a chair in the conceptual sphere that a human can. When thinking about possibilities for computer intelligence, humans quite often forget about how much knowledge goes behind the simplest concepts, how much background knowledge it takes to read and understand a paragraph. For instance, in Hofstadter's preface, philosopher Margaret Boden refers to a program "ACME" which is supposed to be able to understand Socrates' analogy between philosopher's(as teachers) and midwifes. The program seems to have no understanding of any of the words which are being presented, it simply follows a pattern which is given to it in two instances. It has no clue about the differences between variables, they could be synonyms or antonyms for one another. The program simply plugs the variables into a given formula without understanding any relationship between midwives and philosophers, or that it is even dealing with substantially different terms, it does not even know they are made of a different string of characters (the program compares them and receives the information that these strings are not the same, but it does not actually know or learn). Hofstadter goes on to show this with his example of how a relationship between a completely irrelevant story could be drawn to resemble what the program supposedly understood. I completely understand why Hofstadter was so upset about the misconceptions that the Eliza Effect causes, it is why we have so many bad AI movies about the morality of computers in the future.

Math Methods

The sample run of Numbo illuminates Numbo's attempt at modeling human cognition. The program uses the same methods that most humans use when trying to solve a problem as well as the declaritive knowledge stored in the Pnet. Some of these methods were described earlier in the chapter. I thought about how useful these techniques really were and decided that they are little methods I don't even realize I use because it happens so fast; I only notice when they don't work because then it gets more difficult and I have to really concentrate on the problem. I think Hofstadter says something to that effect about Jumbo...but anyways as I was looking through my materials that are supposed to prepare me for the math section of the GREs I realized that I use these methods, just mostly in some what of an unconscious or mechanical way. I like that Numbo puts these human methods to use, especially the knowledge of approximate sizes of numbers and rote small-number arithmetic. I think a lot of knowledge can be derived from using these simple methods, to build a more complex idea to solve a problem.
In Defays' second to last paragraph he articulates the capabilities and possibilities of Numbo as a program attempting to emulate human cognition, or at least be a step further in that direction.
"The Numbo project has shown that, with an appropriate architecture, a system can behave, at least in a limited domain, in a very fluid, humanlike way, combining the ability to spontaneously perceive chunks, the ability to manufacture groups, and the ability to achieve goals through the chaining of different operations. Of course, Numbo's capabilities could be greatly imporoved."

Seeds of the Mental

Daniel Defays' paper discusses the number problem Numbo, similar to Jumbo, a slightly different form of Crypto. Numbo attempts to simulate human cognition, namely pattern recognition. Solving Numbo and Crypto with a program are both good models of human cognition, with Crypto being slightly simpler, considering there are less possibilities due to more rules. Numbo's Pnet seems particularily complicated; I feel like much of the knowledge stored in the human brain is not easily brought to the surface when trying to think of relationships between numbers that we call upon to help solve problems such as Numbo or Crypto. It would prove difficult to capture every connection we might make between numbers to solve a problem; as the paper says, knowledge of relationships between numbers varies from one individual to the next. The knowledge used may differ each time as well, as the preface says. Depending on the context we may draw on a different method to figure out the relationship between bricks and the target. It would be quite a complicated system, that would constantly be changing, to store "common knowledge" of relationships between numbers.
I did, however, like the idea of giving links weights and nodes different degrees of activation. This reminded me of neurons in the brain. Not only does there seem to be a degree of plasticity in this idea of degrees of activation and weights but it also seems to communicate much like neurons in the brain. The nodes communicate by activation spreading throughout the nodes in an uniform pattern (okay so it differs slightly from the brain, but it is the idea of activation spreading from one node or neuron to its neighbors that resembles human cognition). I kind of saw something similar to plasticity in the way that Defays descripes how due to the context and other intrinsic factors, some nodes or links will be more prone to activation than others. This vaguely reminds me of plasticity or how some pathways in the brain that are activated more than the surrounding areas have stronger connections. I love that Defays thought this vital to his program, and I agree it is integral in modeling human cognition, for the functioning to resemble is as closely as possible.
Numbo also seems to have something similar to the "commonsense halo". Defays says that "[a]ssociation (simulated by spreading activation) is clearly the key notion here -- but total reliance on blindly spreading activation can give rise to uncontrolled, chaotic behaviors of the network." Therefore, some system must be implemented to keep the program in the "commonsense halo" of relationships of the numbers it is working with. As we have learned, this is also similar to human cognition; the "commonsense halo" keeps our cognition on track and from drifting from one relevant topic to another completely irrelevant topic.
Overall, I enjoyed Defays' paper and Numbo. I do like Crypto more, probably because it is something hands on in class for us to try ourselves, but does Numbo seem to be the basis of our project.

Multiple Musings

I like Hofstadter's two ideas of ways to transform pseudo-words by using entropy-preserving and entropy-increasing. This reminded me of the word scramble game I play. I use these techniques of regrouping, rearrangement, and disbanding. I use the scramble feature to rearrange the letters, or if I create one word I use regrouping to see what else can be made. When I run out of words to create I start disbanding and then working with what is left. I think regrouping could be improved by the rules that would be given to jumble, so as to prevent the unacceptable initial consonant clustering, as seen in the first paragraph in Hofstadter's 'Entropy-preserving Transformations' section. But then again there might be too many irregular rules in English for Jumble to actually follow any. Such as the single/double syllable problem of "soap", "boa", and "coop".
I think some of the methods that Hofstadter presents would prove useful though because they resemble human cognition. I think that spoonerism, forkerism, kniferism, exchange, and reversal all definitely coincide with human cognition. These are all techniques I think that I use when playing a word scramble game.
So many of the different options for scrambling the letters to pseudo-words in jumble, may yield results in a minuscule amount of instances. How would jumble discriminate when to use each method? Jumble should first use the methods that are most similar to human cognition, then these others. I think human cognition has adapted to recognize or create words, so we are mildly decent at it through these methods.