DJ Strouse

1) Randomness is, in a sense, a measure of compressibility. A physically random system is one whose dynamics we cannot predict and that we must simulate in order to know. It is temporally incompressible. Of course, in reality, physical processes tend to exhibit degrees of randomness, so that we can say some things about them, but perhaps only a limited number of things up to a limited accuracy for a limited amount of time.

Is “that which cannot be predicted except by a full-scale simulation” a proper definition of a random physical process? How does this relate to Kolmogorov complexity?

2) We need a measure of how difficult it is to recognize the order in an apparently random process. Kolmogorov complexity is a good description of the compressibility of a process, but it doesn’t tell us how hard it is to recognize that minimal description. It may be that we think a process is quite random, requires a long description length, and thus has high Kolmogorov complexity but, with just one little insight, we will find some beautiful underlying structure and our minimal description length (and Kolmogorov complexity) drop enormously. Or it could take centuries of great thoughts to make such a simplification. Or it might not even be possible. That Kolmogorov complexity says nothing about the distribution of descriptions in “description space” and the ease/difficulty of discovering the minimal one seems to me to be a serious shortcoming. A better understanding of the “difficulty of description” might give us some interesting information about scientific theory and what types are “easier” or “harder” than others to discover.

These thoughts are inspired by a lunchtime conversation with Henry Yuen today on the nature and importance of randomness and a conversation last week with the USC theoretical computer science lunch bunch over a long-lost, unpublished paper by Len Adleman about randomness.

Tags:   complexity · randomnessNo Comments.

My Goodreads Rating: 4 of 5 stars

Until recently, those interested in learning about consciousness have had just three options: (1) introspection (informative but deceiving), (2) books by philosophers (interesting but completely speculative), and (3) books by crazies (the majority of the literature on consciousness). Consciousness has long been a naughty word in science, but ho! No longer! While the “hard problem” of exactly why phenomenological states arise from the collective squirts of neurotransmitters washing across your brain at all is still a crapshoot, the relatively “easy problem” of correlating certain neural activity with certain phenomenological experiences is well underway.

Christof Koch of Caltech is one of the leaders in probing visual consciousness (or “awareness” if we are speaking to grant committees – shh!) – a particularly “easy” form of consciousness that is amenable to experiment both in humans. Dr. Koch’s “quest” is to identify the minimal set of neurons whose activation leads to consciousness. The book provides a grand tour of all the interesting quirks and subtleties of visual consciousness discovered in the last few decades, painting a picture that is far more fractured and fragile than our daily experience might suggest. If you still cling to the picture of the homunculus dictator riding a meatbag mech warrior around the world, this book will, at the very least, convince you that biological dictators can’t do their jobs without an army of unconscious robot agents. The book also includes some more speculative thoughts on the purposes and general nature of consciousness but, perhaps surprisingly for a book with such a lofty title, consists almost entirely of good old-fashioned science.

What Brian Greene’s The Fabric of the Cosmos is to modern physics, Quest for Consciousness is to modern neuroscience – the finest popular account available for lawyers, stock brokers, and mailmen with a bad science habit. Koch’s book might focus on visual consciousness, but he touches on learning and memory, motor control, and lots more on this quest and does so in his highly readable and concise style. While mucking around in neurobiology can make for dangerous trekking, Koch organized the book very well with small, informatively titled sections, making it easy to remember the salient point of a particular passage if you’re a neuro-rookie and easy to skim a particular passage if you’re a neuro-master.

For a more philosophical approach, I’d also recommend Daniel Dennett’s Consciousness Explained.

WARNING: The rest of this review is a collection of my reading notes. Readability not guaranteed.

• The evolutionary development of the brain makes it the ultimate kluge. New functions are continuous adaptations of old ones, making for some pretty wonky and unintuitive design principles. Speculative example: humans evolve disgust to avoid bad food. As humans form larger societies and viral epidemics become a problem, it becomes evolutionarily advantageous to avoid certain people. Rather than reinventing an avoidance mechanism, evolution simply co-opts the old disgust system (for dealing with bad food) to deal with potentially infected fellow humans.
• Is the function of emotions to allow detected correlations without known functional relationships to influence decision making? (In other words, I might not know by what mechanism two things are related but have unconsciously noticed a pattern in their coincidence.)
• Approach to decoding neural computation: consider that the typical sensory stimuli or brain info that feeds into a circuit will be highly tuned to the absolute scale, relative scales, and symmetries of that which it represents. Considering the nature of the input may constrain neural computation and suggest approaches to understanding it in a particular circuit.
• Does our consciousness integration window increase in unfamiliar environments and/or decrease in familiar ones? Does it increase or decrease in rapidly changing environments? How does one test the integration window in a dynamic setting? Traditional artificial masking protocols used in highly controlled environments don’t seem appropriate.
• As consciousness seems to change continuously, does this mean that the NCCs (neural correlates of consciousness) must be constrained in their activation patterns? How does the continuity of conscious experience arise from the somewhat digital spiking of neurons? Is there perhaps a continuous “read-out” process wedged between NCCs and qualia? Does the distinction between discontinuous and continuous dynamics suggest a useful approach to decoding neural computation?
• How does the brain optimize the balance between zombie agents and conscious processing? The zombie system offers speed at the cost of flexibility, operating through previously detected correlations and heuristics. Conscious processing offers flexibility at the cost of speed, offering a rich simulation of a series of events in order to consider and weigh consequences. (Simulation is the best word for this – you don’t know where your thoughts will end until you “run the program.”)
• How might one quantify consciousness? Do more explicit representations lead to more meaning and hence more consciousness? Would this justify deeming babies “less conscious”, since they have not yet developed the repertoire of explicit representations available to an adult?

View all my reviews >>

Tags:   brain · consciousness · psychologyNo Comments.

Call me a radical, but I have a growing notion that a professor spewing forth information to dozens of students playing classroom stenographer is perhaps not the best way to educate the next generation. Here is my favorite idea for a better classroom experiences that I’ve pieced together from friends, my own experiences, and daydreaming in the shower.

Note that, at least in this post and the motivating conversations, I’m still holding on to the assumption that learning is a process that begins at age 4-6 and ends somewhere between 22 and 30, depending on your interests and levels of masochism. In other words, I’m assuming that learning is not a lifetime experiences interwoven into our daily lives. I’ll delve into that possibility in future posts, but let’s keep it simple for now.

Professor as Advisor
Parents stop reading stories to children sometime around age five or six, so why do professors continue to read textbooks and lecture notes to students until age 22? That’s a waste of both professor and student time. Instead, choose a topic each week, suggest some reading for students, and allow them to, by any means necessary, teach themselves. They might read a textbook, watch some video lectures, and/or gather in small groups to discuss. During this preparation time, students could keep track of questions that arise that they can’t work out between themselves. To further guide students, a professor might point out several subtopics that individual students might specialize in and summarize to the others. Once or twice a week, students would meet with a professor, discuss questions, present on their specialized subtopics and, depending on the type of course, do example problems (math/science) or brainstorm real-world examples (social sciences). During this “class”, professors would not play “sage on a stage”, but rather more of an advisory role. The professor might pose a problem and students would work together at the blackboard to solve it. The professor would be there to offer guidance, suggest alternative approaches, and point out interesting examples.

Advantages:

• more time for student questions
• more time spent solving problem together
• students get the chance to teach
• less time commitment for the professor

Disadvantages:

• requires motivated students

My personal experience is that I learn little from lectures and much more by doing and teaching. I’m currently taking a general relativity course that is a watered-down version of this model and I love it.  Instead of dreading (or more realistically, simply skipping) a two-hour, twice-a-week course, I look forward to it.  Each week, we go off on our own, read a book chapter, work on about five problems, and choose an interesting experiment or physical phenomena to present on.  No homework to turn in whatsoever.  And the result?  Students in the class actually fight with one another over who gets to present on some examples because it’s fun to present interesting examples you’ve worked out for yourself. It’s not fun to work on a half-dozen problems on your own, hand them into a mysterious wooden box, never discuss them, and receive a crumpled version of your work, annotated with a cryptic number, several weeks later.

Summarizing, some key ideas that I’m suggesting need to be worked into the class experience are:

• student-to-student teaching
• live, collaborative problem solving
• student responsibilities outside of class
• no need for graded homework

Thoughts?

(Image above from Blair Kathleen Kelley athttp://students.ou.edu/K/Blair.K.Kelley-1/)

Tags:   classroom · education · teaching2 Comments