Preface 5

The overlapping and clustering of concepts in our brains can be seen through basic human interaction. For instance, when someone makes a speech mistake and says something that he/she did not mean to say, it is usually due to this overlapping of concepts in our brains. The incorrect word that one says can often be traced back to the similar properties of the word that was meant to be said. One of the most obvious instances of this is when someone mistakenly says a word that is pronounced similarly to the word that was meant to be said. Another example of this clustering example can be seen through mistakes that humans make when writing a program. There are many ways in which different variables can overlap in one's mind, which can cause us to make fundamental coding errors. However, I think that the overlapping of concepts in our brains is a good thing due to the fact that we can quickly relate a current stimulus (object) to other objects that are similar thus creating coherent relations.

Chapter 4

I find the problem of figured out how humans turn sensory information into mental representations quite interesting. There are many different theories out there, and there really is no "correct" one because the way that humans store mental representations is not scientifically proven. I believe that the way we make mental representations is so complex, it will never be correctly modeled by a computer system. In order for a computer system to create mental representations like humans do, it first needs to gather stimuli in the same way that humans do. The main sense that we use is our sense of sight. As we look around, we are constantly recognizing tens, if not hundreds of objects. There is no computer system that comes close to the effectiveness, accuracy and efficiency of recognizing images as our eyes are. Once we come up with a computer system that is effective as our eyes our, then we will only be scratching the surface on correctly modeling human mental representations.

Preface 4

I found the discussion about the Eliza effect to be interesting. I think that we fall for this illusion more than we even know. We tend to give computers too much credit for the tasks that they accomplish. I believe that this stems from the fact that we sometimes think of computers as "thinking things", instead of just devices that follow commands that are given to them. One example of this illusion is with the 20 questions computer game. In this game the user thinks of an object and the program asks the user 20 yes/no questions to try and figure out what object the user is thinking of. Most of the time the program answers correctly. Now some people will go and say that the program "knows what you are thinking." But in reality, the computer just making a conclusion of the most likely object based on the answers given. Essentially, it just comes down to a process of elimination, not the program actually knowing what the user is thinking.

Up to p154

Given the description of how Numbo works I feel that it does a pretty good job of mimicking humans solving these problems. However, even with all the similarities to humans, I still believe that it is still nowhere near close to completely replicating human mental process for Numble problems. This is because every human is different. Some may try to approach the problem different due to the simplest things such as personality differences. An aggressive person may rush and try to compute numbers as fast as he/she can to try and produce a lucky quick result. A passive person may look at all the numbers and develop an efficient plan to solve the problem. Numbo does not have the capability to take these differences into account. Numbo is not specific enough to replicate human mental processes when it comes to these problems. As for what Numbo sets out to do, in a general sense it does have some similarities to human mental processes, but there is just too much missing to call it a model of human mental processing.

Up to p138

I found the Numbo program interesting and much more plausible than the Jumbo program that Hofstadter talks about in the previous chapter. The reason I think that the Numbo program is more plausible is because unlike the Jumbo program, there is a relatively small set of ways that a particular problem can be solved. A human would figure out the solution in the simplest, most straight-forward way. There lies the task of the Numbo program: out of a list of solutions for a particular problem, which one is the simplest? While this is still a difficult problem to solve, it is not nearly as difficult as the Jumbo program where there are no clear solutions. Words are completely independent from each other, and there is no clear simplest way to solve Jumbo problems.

Up to p126

The more that Hofstadter explains how the Jumbo program works the more I think that it is impossible to make. I think that with a dictionary database it is impossible to correctly model how humans solve these types of problems. There needs to be some type of middle ground between no dictionary at all and brute forcing it through string comparison. Even as humans solve jumble problems we are analyzing the jumbled word and extracting information from it so we can eliminate words in our mental dictionary. One obvious example, if you are trying to solve a jumble problem of 8 letters, you are not going to be thinking about a short word because that is impossible. I think in order to make the Jumbo program successful, there needs to be a process that analyzes the jumbled word and uses information from it to create a dictionary subset of the possible words. Creating a program that solves jumble problems without a dictionary just seems like a daunting and impossible task.

p97-111

Reading about the Jumbo program was interesting to me. I was very surprised to read about how Jumbo has no dictionary to match the words up to. When I first read about the purpose of Jumbo I thought to myself that it will need a dictionary of all English words in order to function properly. However, once reading about how they are going to construct Jumbo by making it like a human mental process, I don't think it is possible to correctly create it so it behaves this way. The reason I think this is because at this moment we don't even know all the details about how our actual minds store and fetch words (and parts of words for that matter). So, if we don't even know how our minds store and fetch words, how can we possibly create a program that is supposed to mimic that mental process?