Abstract Analogies (Preface 5)

This part of the reading discusses what allows analogies to be detected. He suggests that 'the ability to perceive regular patterns and to formulate rules describing those patterns' is what allows analogies to be detected and it feels to me that he is right or very close to being right. He then goes on to talk about how humans look at certain things in terms of the 'role' it plays. He shows how humans have a tendency to 'feel' like things usually have only one role. He gives an example by asking the reader to think about what the word "hard" means and says that it probably feels like one idea to you. He then lists about fifteen different meanings that the word hard can take on. Even after we are shown that it does have a lot more then one meaning he suggests that we still might think of those things as one idea still. He then gives all the meanings of the german word hard which ultimatley shows how context and culture determines the boundary lines of concepts.

After thinking about the seemingly infinite things that must be considered to identify an analogy it seems like a much more complex cognitive process than I had previously thought. Making a program that can recognize analogies in two different conceptual things seems near impossible after reading how complex the process is behind such a thing.

High-level Perception (Chapter 4)

A point of interest in this chapter was the discussion on high-level perception and what constitutes it. Hofstadter suggests that a critical part of high-level perception involves "mental representaion," which he defines as "the fruits of perception." "In order for raw data to be shaped into a coherent whole, they must go through a process of filtering and organization, yeilding a structered representation that can be used by the mind for any number of purposes." The previous quote explains how humans 'make sense' of what's around them and appropriately react to stimuli in their environment.



He then goes on to explain what it is in my opinion the most important property of high-level perception, and that is that it's extremely flexible. Because of this flexibility input may be perceived in many different ways the book goes on to say. This made mt think of something pretty awesome, I realized that everyone in the world literally has a 'different' kind of high-level perception and that they all have cognitive processes unique to them. Because of this uniqueness of cognitve processes, the term 'human-level cognition' literally has about 6.7 billion different and 'correct' definitions. This also makes me think that when someone is trying to create AI are they being so ignorant as to imply that humans all behave and act the same way, or do they try create a certain 'type' of human level cognition, perhaps the one possessed by the programmer or something.



I don't think it is possible for one single program to represent accurately the level of cognition that humans have, instead one program could only represent a single humans unique cognitive processes.

The Eliza Effect (Preface 4)

In this section of the book, Hofstadter discussed how computers have a sort of "empty" understanding of the world we love in. For example, a computer may be able to make an 'analogy' of water flowing in a pipe and heat flow through a metal bar. Even though the computer may be able to produce a great analogy between water flow and heat flow it still does not have a deep unstanding of water and heat. A computer "has no representation anywhere of what it means to be liquid, or of what "greater than" means, or of what beakers and vials are, etc".

In the world of computer science there exists a thing known as the "Eliza Effect." The "Eliza Effect" is the term given to the illusion of a computer 'understanding' you, 'sympathizing' with you and showing you 'empathy'. In reality, computers have no conceptual understanding of what anything is really, in a way it only 'knows' 'empty' facts. An example of the "Eliza Effect' occuring in the real world would be someone actually thinking that an ATM was actually thanking them when the screen flashed "THANK YOU" after making a transaction. The "Eliza Effect" really interests me because it explains how computers can seem to interact with humans while appearing to display human emotions without actually expressing any emotions at all.

Numbo 138-154

An interesting concept that was discussed in this section was the concept of cytoplasm. Cytoplasm is what Hofstadter calls the place where "the building and dismantling of temporary, puzzle-specific structures take place". Numbo shares this cytoplasm with Jumbo. Hofstadter said the cytoplasm could be thought of as a "black board" or a "working memory", this sounds very intuitive to me. I would thinking it could also sort of be like a "mind" because all processes occur within it, just like everything humans percieve occurs within their mind.

Numbo is desgned to imitate human cognition during the solving of a Numble problem. It sounds like Numble does a pretty good job in replicating human cognition, it even makes mistakes like a human and will go in wrong directions. I admire Hofstadters pursuit of creating human-level cognition and the fact that he even accounts for the flaws that humans have when solving such problems. I never really thought of how hard it would be to create human-level cognition.

Real Artificial Intelligence 111-138

In this section, Hofstadter discusses some very interesting concepts. One of these concepts is parallel processing. He discussed how a program named "Hearsay II" could construct 'multiple top-level structures at once. An analogous vision program would construct several top-level interpretations for a scene, all in parallel.' If I were to compare this feature of Hearsay II to human cogntion it would be like having access to multiple states of perception at the same time which is very hard to imagine and may even be impossible, but it's fun to think about.

Another interesting concept that was discussed is called "intelligent backtracking". This involves taking a step backwards and undoing some prior decision. For this to happen, the program must first realize that where it currently is is not where it wants to be. For example, perhaps Jumbo created a "word-like" word but it is not really a word. It must realize that the group of letters is not a word and then go back and take a different route to wordhood. This interests me because it seems as if Jumbo is actually thinking, and maybe it really is.

Jumbo 97-111

This section focuses on a very complex and maybe even genious computer program named Jumbo. Jembo's purpose was to be able to imitate human cognition that takes place when one is playing the newspaper game "Jumble." One interesting property of Jumbo is that it has no knowledge of the English dictionary, instead it uses pre-set rules that apply to the English language to construct 'english-like' words.

Another interesting part of this section is how intelligence is defined, which is as follows:
"intelligence emerges out of the interactions of many thousands of parallel processes that take place within miliseconds and are inaccessible to introspection." According to this definition it seems like it would be nearly impossible to re-create human-level cognition, but I have a feeling some of these people are going to die trying and may even achieve such a feat.

I find it very impressive that a program that is designed to form English words out of scrambled letters can do so with out "knowing" the words of the enlgish language. I can't imagine how complex the code must be for such a program.

It's All In Your Head 86-97

In this section Hofstadter starts off by telling us a little bit about a program called Jumbo. Jumbo tried to make English-like words out of a set of latters by re-arranging them and putting them into plausable orders. Hofstadter then goes on to tell us that he personally enjoys doing anagrams and is curious what goes on in one's brain while figuring out such problems. Hofstadter claims that words just magically pop into his head and he felt like a passive observer to this phenomenon.

Hofstadter then goes on to discuss knowledge sources, which he abreviates as KS. He claims that knowledge sources have to "sensory organs" to inform it and sense things in its surroundings. This makes me think that humans are probably the most advanced knowledge sources ever to exist.

Hofstadter then goes on to talk about something he read in a paper by Fennell & Lesser from 1975. What he read was: "Preconditions themselves have preconditions, call then 'pre-preconditions.' In HSII, knowledge-source preconditions.... may be arbitrarily complex. In order to avoid executing these precondition tests unnecessarily often, they in turn have pre-preconditions which are essentially monitors on relevant primitive data base events.... Whenever any of these primitive events occurs, those preconditions monitoring such events are awakened and allowed to test for full precondition satisfaction." Hofstadter really thought that this quote was packed with important information even though it was not emphasized in the reading. The fact that Hofstadter saw the value of that quote shows his natural ability to notice important information and use it to further enhance his studies and ideas.