Welcome! Welcome to me, and welcome to you! This is the start of it all. I started my summer research project May 1st. What follows is a documentation of sorts of that first week. I hope that all my weekly updates will be able to take the following form: a section on theoretical discoveries, a section on technical discoveries, and then a section on personal/social discoveries documenting my transition into the lab community. This week is no exception, so here goes.
Readings for the week:
The Atomic Components of Thought by John R. Anderson and Christian Lebiere
Foundations of Conitive Science edited by Michael I. Posner
After meeting Renee on Monday, and a brief introduction (if I was still in math I would make some joke about Renee being the nth member inducting me as the n+1 member into the cog sci research world.. but I'm not. Though it looks like I still did..) into the world of cognitive sciences, I was gifted these two books to read sections of and told to "get my feet wet". The Atomic Components of Thought is a book discussing the utility and theory behind the ACT-R architecture, which is a system developed at Carnegie Mellon University. Because cognitive sciences is such a varied discipline it is blessed and cursed by having an equally varied vocabulary. When we say an architecture, we do not necessarily mean it the same way we do in computing science. There will be no discussion of PLA's. A cognitive architecture is to lay out the building blocks for thinking essentially, to specify the mechanisms for storing and retrieving knowledge and for the creation of new knowledge. I will leave a further discussion of the ACT-R architecture itself for the technical portion of this weekly update.
The Foundations of Cognitive Science provided a broader introduction to cognitive science, starting with a detailed history of the evolution of the discipline and a summary of all the different fields which are interested in cognitive scienes and their unique interpretations and ideas which result from their varied backgrounds. Aside from being a useful guide to the world of cognitive science research, I found this book most interesting because it also made me consider what it is to be a computing scientist. Not often when reading my CS textbooks do I realize the way we approach things and think is somewhat unique. Taking that step back and realizing there is a discernable different process to the way we behave as a faculty, as opposed to assuming the things we discuss and are interested in as computing scientists are equally as natural and interesting to the rest of the world. Realizing this made me step back from everything I'd been reading and reanalyze it with the notion that people think differently than I do. This brief internal reverie on "thinking about thinking" actually wound up being where my most valuable insights into my readings emerged from!
Later in the week found me in another meeting with Renee to discuss how the readings went. We spent a good deal of itme discussing the flaws with current cognitive science research. My most burning questions all had to do with the good deal of magical handwaving researchers do. The ACT-R architecture is a system that runs on goals(what you want to do), productions(how to do it), and declarative knowledge(things you need to know along the way). The problem I had waswhere these goals were coming from. In the beginning, when I push the little go button that brings the program to life, I've given it a beginning goal, and as a result of some of my productions firing, there may be several subgoals created but in the end, when I've completed my initial goal, what do I do? And the thing is, we do not know. I ask again, how do we know that the world gives us data in a way which works with this production system we've created? Can this world be converted into the appropriate symbols to allow for this system to continue? We don't. Cognitive sciences has picked up the task somewhere in the middle. A great deal of it retains a sense of magic and make-believe. We pretend we know how knowledge got into the brain to begin with, we pretend we know how we can interact with the environment, and from there we try and process information. We take a child with certain basic skills already established and show how we think their brain can take a goal and process it. We do not know what to do if that child is empty, or what the BASIC knowledge a baby starts with is. We also do not know how a babies brain is different as it grows, the brain grows in parallel as learning and how does THIS impact learning? Do we need a different model or architecture or machine for a younger learner? Why? What would this do?
And the great thing is, we don't know. People criticize researchers for ignoring these big questions, what if everything they've done is for naught when it turns out all those assumptions we were making were wrong? Well I say, even still, inspite of the fact that these tools we develope now might be inapplicable, the knowledge and insights and experience considering these problems alone is valuable enough to make this research important.
But that is merely what I think, and what I've been thinking about. It is week one, and we shall see where this path lined with conflicting ideas and theories takes me. Tune in next week, same bat time, same bat channel.
This would be where I normally discuss more technical or application type details. This week was more devoted to reading and aquiring a computer and as such I haven't a great deal to discuss here. I will instead provide a brief summary of the ACT-R architecture, which is the system I will be using this summer. ACT-R is very basic, it believes there is a chunk of permanent memory and it is divided into two types of memory: procedural and declarative. The former is like steps to take to achieve a goal and the latter is facts such as "3+4 = 7". Procedures would give us a generic plan for finding the sum of numbers and declarative chunks give us specific facts. When procedures are executed they can create new declarative chunks, for example when the procedures to add are followedto accomplish the goal of adding "3+2" the solution can be remembered as declarative memory so that the next time the system only has to remember the solution instead of computing it. Which procedures fire are driven by the current goal and which declarative facts are remembered is driven by the procedures. When several procedures or facts match a given call and are fighting for supremecy, there is an "activation" value for each that says which will win. This value is determined several factors, some of which include a) it's likelyhood of success b)how "costly" or time consuming it is and c) How often it has fired.
I could go into a great deal more detail but I don't feel summarizing the entire book I'm reading on the architecture is necessary. I have given you enough knowledge to understand my discussion on my work, as more is required so shall I dole it out. I hope this has been some what comprehensive. Now don't just stop reading here, feel free to continue on down to my discussion of hte social aspect of my week.