4 ms·
I think your instinct is both correct and missing something. At some level, 'thinking' involves transmitting information about the world. Suppose you had a poor
by glial 6y ago
I think your instinct is both correct and missing something. At some level, 'thinking' involves transmitting information about the world. Suppose you had a poor model of a new task - the code used to encode information would necessarily be inefficient, since you don't know the probability distribution of anything related to the task. Once you do the task for a while, you get a better sense of the probability distribution of stimuli, and (in theory at least) could develop a more efficient code. Adaptive Huffman coding does exactly this - builds codes on the fly that gradually get more efficient. It's possible that the brain does something qualitatively similar. If so, it would 'look like' information transmission is happening faster.
From this perspective, 'information transmission rate' could actually be constant during learning, when defined as a reduction in subjective uncertainty over time. But because you start with a poor task model, your initial uncertainty is higher when starting learning, due to unfamiliarity with the task, so to an outside observer it looks like information transmission is slower. As your task model improves, your prior for each message matches the environmental statistics better, and information transmission speeds up -- again, just from the perspective of the observer, as your subjective uncertainty now starts low and just goes lower.
- PeterWhittaker 6y agoWell said. Your comment reminds me of a conversation from 30+ years ago with a work colleague who opined - or shared the opinion, I cannot remember which - that our "input bit rate" was fixed. Young and argumentative, I disputed this, but.... I don't remember all of the arguments in favour, but he pointed out reading: We start sounding out individual letters and eventually progress to scanning whole words or sentences or paragraphs. It's not that our bit rate changes, it's that our "bit" changes. Yeah, handwavy I know, I am struggling to recall the conversation. Essentially, though, consider your point another way: We can only process so many concepts/notions per unit time, and initially, in any subject, our concepts/notions are smaller. As our experience grows, they grow, so our throughput increases, even at the same processing speed. Sort of like widening a highway to handle more volume at the same speed. Sort of.
- goldenkey 6y agoIn machine learning lingo, you'd have better 'embedding vectors' for the world at large.
- edmundsauto 6y agoThis is what Barbara Oakley (of the popular Coursera course "Learning how to Learn") calls chunking. I think of it like changing the word size based on the topic and my understanding of it.