The Singularity
- Share
- Description Click to Edit
- Original URL Click to Edit
Comments
- threads
- flat
by
hucheng
5 months ago
Interesting. Why didn't evolution result in us naturally reasoning this way though. Is it computationally too expensive for the human brain (to reason everything this way)? Or is not reasoning this way on a daily basis a big part of what makes us what we are? Perhaps that is addressed in the article (I skim read it). I will probe it more deeply later.
- Reply
by
James Choate
5 months ago
Actually people do think that way, just not numerically. And that's the rub. It's called allegorical thinking, 'what's it like?'.We do it qualitatively, not quantitatively.
Bayesian 'thinking' is having some models a priori and then trying to fit specific instances into the right one. We do this allegorical thinking all the time. Every time you ask, which is the best choice?
The primary different between Bayesian v Laplacian statistics is that Bayes goes in with specific hypothesis before hand (ie H0, H1, ... Hnull) and tries to fit the data into it. Consider ID as an example of emotional Bayesian, and getting it wrong.
Bayesian 'thinking' is having some models a priori and then trying to fit specific instances into the right one. We do this allegorical thinking all the time. Every time you ask, which is the best choice?
The primary different between Bayesian v Laplacian statistics is that Bayes goes in with specific hypothesis before hand (ie H0, H1, ... Hnull) and tries to fit the data into it. Consider ID as an example of emotional Bayesian, and getting it wrong.
by
James Choate
5 months ago
I had to stop earlier because I had to leave for work, but I've got a few minutes, so...
Another good example of how there is some Bayesian process going on is the repetitive behaviour of children like when they pour sand or water from one cup to another over and over. At least some psychologists and learning researchers believe this is the childs brain building up internal examples (ie the Bayesian H#). One standard I've seen is that for the repetition to take (ie some cluster of neurons form a 'vector' representation of the behaviour or character) it can require 500 to 5000 repetitions of the act. So perhaps in some sense we are a tabula rasa at some levels. We have the potential but not the specific.
Another aspect of Bayesian statistics, more an application or configuration issue, that I see is that many folks don't, in my opinion, setup their H* correctly. Normaly H0 is the thing we believe is the most likely, and H_null is that it's false. There may be several H# within H* (eg spam filtering for multiple mailboxes w/ H_null being assumed spam) but in general there is seldom a specific "I don't know" category. In my approach I generaly always set H_null to to "I don't know" so that I have an explicit "maybe something else is going on here" category. It's probably a minor point in most applications but somehow I just don't trust my own ability to make that decision a priori.
Another good example of how there is some Bayesian process going on is the repetitive behaviour of children like when they pour sand or water from one cup to another over and over. At least some psychologists and learning researchers believe this is the childs brain building up internal examples (ie the Bayesian H#). One standard I've seen is that for the repetition to take (ie some cluster of neurons form a 'vector' representation of the behaviour or character) it can require 500 to 5000 repetitions of the act. So perhaps in some sense we are a tabula rasa at some levels. We have the potential but not the specific.
Another aspect of Bayesian statistics, more an application or configuration issue, that I see is that many folks don't, in my opinion, setup their H* correctly. Normaly H0 is the thing we believe is the most likely, and H_null is that it's false. There may be several H# within H* (eg spam filtering for multiple mailboxes w/ H_null being assumed spam) but in general there is seldom a specific "I don't know" category. In my approach I generaly always set H_null to to "I don't know" so that I have an explicit "maybe something else is going on here" category. It's probably a minor point in most applications but somehow I just don't trust my own ability to make that decision a priori.
| Shared or Added By | To | When | Comments |
| scottcollier | scottcollier | 8 months ago | 0 comments |
| scottcollier | Bayesian Programming | 8 months ago | 0 comments |
| jen | jen | 2 months ago | 0 comments |
| jen | Jen's Twine | 2 months ago | 0 comments |
| dean | dean | 5 months ago | 0 comments |
| James Choate | Solar Soyuz Zaibatsu | 5 months ago | 0 comments |
| dean | Artificial Intelligence | 5 months ago | 0 comments |
Tell me about Twine
Twine is convenient, new, powerful, smart, and totally devoted to whatever you're interested in. Read all about it.