Take Luck! Preface 5

I could not help but touch on one of my favorite subjects, and that is the English language. Hofstadter titled preface 5: "Conceptual Halos and Slippability", and it really made sit back and laugh a little bit when I read it. One of my favorite things in life is catching little slip ups in the English language.

In detail he goes about saying how he believes there are conceptual halos in languages so that words have multiple meanings. Some languages have more or less meanings for different words, and this is very unique. There aren't too many words in the world that are universal across languages, and this needs to be taken into account when traversing across multiple languages.

One of my favorite examples of this is, and it also dabbles into how many AI programs do not have a full grasp on our world, are language translators that you can access on the internet. I know many people who have attempted to do foreign language homework multiple times and have failed. Hofstadter gave a great example in Italian where the sentence Lei ha fratelli? - is "Do you have brothers?", but he could answer Si due sorelle - "Yes, two sisters.", where fratelli can float between "brothers" and "brothers and sisters".

On another quick note, Hofstadter spoke of one my favorite mishaps in the English language and that is when there are word substitutions. I couldn't help myself to find a clip of my favorite comedian making a joke on this because I know I have done this many times while speaking the English language.

http://www.youtube.com/watch?v=9do7HRaGd00

Perception and AI Chapter 5

Chapter 5 of Hofstadter's book is a critique of Artificial-intelligence methodology. I found this critique very interesting and I found Hofstadter and his fellow colleagues who made this article, David Chalmers and Robert French, hit the nail right on the head for me.

When I read about and see these artificial-intelligence programs doing "magnificent" things such as making analogies, making conversation with people, and problem solving programs. When first reading about programs like these I thought that they were doing something truly unique. Thinking that maybe these programs had a real sense of "knowledge" and "concept". After reading Hofstadter's critique it made me really question many of these types of programs.

Hofstadter writes about a problem of relevance that I find interesting. Which information does the program decide to use at a specific time or for a specific situation is interesting. I went and read examples from the ELIZA program and this program seemed to impress me less and less. It felt like an empty conversation, and it felt as though there was no knowledge being put forth. There seemed to be an emptiness that wasn't fulfilled when reading different examples of the ELIZA code.

http://library.thinkquest.org/2705/Programs.html

I found that the ELIZA program had some relevance when conversing with people, but some of it felt as though Koko the gorilla was sitting there signing conversation to Penny Patterson.

http://www.koko.org/world/talk_aol.html

I feel as though many of the AI programs such as ELIZA, or ACME seem to lack a specific feel to it. Like they actually have some sort of grasp on the world, and do not necessarily see it as a world full of patterns, strings, or numbers.

Eliza Effect Preface 4

The more I read into Hofstadter's book the more I tend to agree with almost every word this man has put onto paper. In Preface 4, Hofstadter takes a closer look at some of the noticeable "AI" accomplishments and he then proceeds to carry out a sort of critique of these accomplishments.

One of the more notable programs that Hofstadter critiques is a program called "ACME" that was developed by Keith Holyoak and Paul Thagard. This program was supposed to be able to draw analogies between Socrates' remarks about himself being a "midwife of ideas". The program "ACME" seemed to be given a knowledge base of what midwifery is. It could then switch out strings of information for one another and easily make an "analogy" between Socrates' and what a midwife really was.

Hofstadter immediately shoots this idea down and shows that "ACME" does not "know" the analogy, but merely hides it behind the switching of strings and patterns within the remark. I found this quite interesting and it again brought my attention to Searle's Chinese room argument. It is so that a computer cannot really "know" what specific things are in the world without a seemingly infinite knowledge base that it has acquired through humans? Is it possible for a computer to actually "know" what certain items or objects are, and be able to analogously compare different items in the world to gain its own "knowledge". This to me seems like an interesting topic, and I know this has been discussed previously in many discussions about AI.

Numbo and Myself (Pg 138-154)

The more I read about Daniel Defay's program Numbo the more I use introspection to try and decipher how I solve problems of this sort. Defay speaks of Pnet and Coderack and it makes me think of this is how human processing really does occur. I myself sometimes struggle with problems that are in the mathematical realm, and I keep looking into my own mind to try and see how I solve these problems.

Defay easily puts on paper how Numbo works and uses a great description and also great diagram from a trace run of Numbo. The main question that I have in regards to Numbo is if it really shows some sort of intelligence when solving these problems. When first reading about Numbo it seemed to me that it merely took one path and tried it until it failed and disregarded other paths that could be correct. In this section of the reading Defay put that unsettling feeling to rest in my eyes. He states that Numbo, unlike much of the previous artificial intelligence programs out there, keeps goal competition going on without disregarding other paths. Some paths will however be "stronger" than others and I thought this was key to his Numbo program.

Numbo to me shows great promise and it surprised me in a way. The architecture that Defay uses is interesting and to me shows "intelligent" like qualities which can be applied to more areas of AI.

So Defayntastic (127-138)

In this section of our assigned reading Hofstader strays away from his own personal work and looks at the work of one of his colleagues, Daniel Defay, that was on sabbatical from the University of Liège in Belgium. Defay was interested in learning more about Artificial Intelligence and the like. Defay had contacted Hofstadter with a great interest in a specific type of project that he wanted to pursue. Defay had a great interest in creating a program that was very similar to Hofstadter's Jumbo program. Defay however did not want to solve problems dealing with letters and words, but wanted to solve problems with numbers.

Defay's program, called Number, was based on a television show called Le compte est bon. This program would take a set of 5 numbers and would be given a goal number to try and achieve from the set of 5. Not all of the numbers need to be used as long as the goal is reached. I immediately thought of the crypto problems that we had been solving in class (where we have a set of 5 numbers and are given a goal number to reach. We can use the basic operators, +,-,x,/, to solve the problem and all numbers must be used). In Defay's program however only addition, subtraction, and multiplication may be used. I thought this would make it much easier for me, but it didn't.

Defay goes through his program and it made me think about how I actually solved the problems that he had given within the reading. He had given different examples for readers to try and solve and I found a few of them quite challenging when in fact they were really quite simple. It makes me wonder how we really do look at these problems in our minds. Defay states that there are "bricks" of numbers that are quite salient that I overlooked completely. This shot down my confidence more with math and made me appreciate Jumbles just a little more. It makes me feel as thought our brains are sometimes more open to solve language problems due to the use of language more than mathematics, but that is just my opinion ( a very biased one at that since I do not like Math as much as I really should). Defay explains this in his book L'esprit en friche, which I hope to read quite soon. His Numbo program is an interesting program that tries to look at human level problem solving with mathematics. I hope to find a copy of his book sometime in the near future.

Jumbo and Trees

When reading about Hofstadter's program Jumbo there were a few questions that came to my mind in regards to its abilities to solve Jumbles. I looked into my own mind when solving Jumbles and so many things came forward. How I first off don't like to solve Jumbles, secondly, that when I look at a word I just randomly throw letters together when I can't solve a word to hopefully bring forth a coherent word.

It made me think of how Jumbo solved the Jumbles that it was given. Hofstadter clearly states in the assigned reading, from pages 111 to 126, that there is a clear way in which he achieved this. Hofstadter explains that Jumbo uses a "tree" like hierarchical structure to solve the problem at hand. It randomly chooses different paths to add letters and if it does not pan out, Jumbo uses backtracking to randomly select another path to take. I find that Hofstadter's thought process on this is right on the mark. Selecting specific pathways with rules and other things of the like. There are however biases put in so that the Jumbo has gloms that are of higher urgency and also of lower urgency. This can separate out gloms that are not important to Jumbo.

I enjoyed Hofstadter's segment on Jumbo and how he explained how his program really worked. It's a very interesting idea that definitely struck my interest, and I'm excited to see where Hofstadter goes from this point on.

To Arthur!

First off, Happy To Arthur Guinness day! In this weeks assigned reading, Douglas Hofstadter talks about his artificial-intelligence research project called Jumbo. Jumbo’s purpose is to act like a human who is trying to solve a Jumble problem by taking a group of letters, then trying to take the letters and make it so that it can seek an English word out of the given letters.

Hofstadter did not give Jumbo a dictionary, because he thought that it defeated the purpose of what he was trying to replicate, in essence, human thought processes. What he did was give Jumbo instructions on how the English language makes its constructions, such as how consonant and vowel clusters are formed out of letters, syllables out of clusters, and words out of syllables. This immediately made me think of John Searle’s Chinese room argument, and made me think if Jumbo could really show some sort of human like intelligence, or an understanding of what it actually was doing.

I think Hofstadter was going in the correct direction when he was designing Jumbo and his thoughts and ideas that are there are very good. However, there is always the slight bit of doubt that an artificial-intelligence program is merely doing the work, but still does not have an understanding of what it is doing. I look forward to reading more about Jumbo to see the outcome of Hofstadter’s program.