I discovered this book while I was a graduate student at CMU. Not taught in any of my courses. Underlied a lot of what was going on at CMU. That's where EDM was first invented, to a large degree. Not a coincidence. Note that this is the same Simon as in the previous week's readings.

Today's class discussion questions
Simon, ch. 1, 2, 5, 6

Core Questions

When we study curricula or online learning environments, are we studying the natural world or the artificial? What are the implications for the long-term longevity and applicability of our findings?

When we study learners, are we studying the natural world or the artificial? What are the implications for the long-term longevity and applicability of our findings?

What about the effects of culture on how learners respond to learning situations? How fast can we expect the culture relevant to education to change? How would we even know that it had changed?

What steps would be needed for educational interaction be reduced to an optimization problem, a set of mathematical function for which an optimal solution to be found? Is this possible? Desirable?

"In the past much, if not most, of what we knew about design and about the artificial sciences was intellectually soft, intuitive, informal, and cookbooky." -- Is this true of educational design today? What about educational data mining? (Note my claim in Feature Engineering Studio that modern EDM is like ship navigation in 1000 BC). How can we surmount this problem? Is it worth having classes like Feature Engineering Studio that are "intellectually soft, intuitive, informal, and cookbooky."?

A core use of EDM/LAK is prediction, and the way Simon thinks of prediction applies to more than just "prediction modeling". Simon argues that in some cases, homeostatic mechanisms (regulatory mechanisms to keep another variable constant) and feedback (in this case, feedback from the student/teachers to the system or designers) can be more effective (and that the three should be used in concert). Is he right? How do/should we use prediction, homeostatic mechanisms, and feedback processes in educational systems?

Who is the true client of education and learning analytics? What are the implications of choosing different clients to focus on?

Secondary Questions

Given that precision estimates are available for most EDM models, why aren't error bars or probabilities used very often in reports to teachers or system-internal decision making? Should they be? How could they be used?

How do we avoid local optima in educational design? (For example, steadily improving a type of online learning system until it is as good as it can be, while being decidedly inferior to a different type of online learning system's asymptote?

Simon claims that (in the early 1990s) natural sciences have almost driven out the sciences of the artificial, that one does not see first-rate dissertations at first-rate universities concerned with problems of design. Is that true of Teachers College, and of schools of education in general? What is right and wrong about the current state of affairs in Schools of Education (with regards to this)?

In what ways can cost-benefit analysis play into learning analytics design?

In what ways can means-ends analysis play into learning analytics design?

In learning analytics design, the cost of designing plays a relatively prominent role compared ot the cost of the design (e.g. implementing the design is not radically more expensive -- and may even be cheaper -- than collecting and analyzing the data and developing models). How does this impact choices about what to analyze in learning analytics?

Simon says that the behavior of an artificial system should be adapted to the outer environment. How would a rationally designed educational system (writ large) behave? Do our current systems behave like that? How do the properties of the inner system show through?

Simon discusses the value of computer simulations, however imperfect, of key phenomena. Why do we have so few simulations of educational phenomena? (One counter-example is SimStudent by Noboru Matsuda and Ken Koedinger... but it has not been adopted beyond that group. Another example is Anderson's ACT-R Theory, but its modern use is radically diminished.) Is it just fashion, or are there key limitations of symbol systems for modeling education?