3 ms·

Sounds to me like you're already familiar with ACH [0]. I was sure that there was a free online platform, but I'm not seeing one. But if there was, it would b

by woodman 10y ago

Sounds to me like you're already familiar with ACH [0]. I was sure that there was a free online platform, but I'm not seeing one. But if there was, it would be easy to plug in what you've posted and enable a structured collaboration. [0] https://en.wikipedia.org/wiki/Analysis_of_competing_hypotheses https://en.wikipedia.org/wiki/Analysis_of_competing_hypothes...

IgorPartola 10y ago
Oh man. Nope, not familiar with ACH. Thanks for the link.
- rdl 10y ago
  It'd be fun to recreate http://competinghypotheses.org/ http://competinghypotheses.org/ in a more modern technology.
  - woodman 10y ago
    It would be fun to implement it to take advantage of the benefits that a restricted natural language would bring, like Attempto [0]. That would allow you to really automate the evaluation of hypotheses against new data points. You might be able to wedge a SAT solver in there as well. I've got it on the ever growing todo list. [0] https://en.wikipedia.org/wiki/Attempto_Controlled_English https://en.wikipedia.org/wiki/Attempto_Controlled_English
Lxr 10y ago
Interesting, what about prior probability and Occam's razor? The 'process' section mentions "less consistency implies a lower likelihood" but makes no mention of preferring simpler hypotheses which I find surprising. Edit: 'Bayesian analysis' is briefly mentioned further down the page.
- woodman 10y ago
  ACH is just a very high level framework for shaking bias, the methods are implementation details. This is for humans, so you have to keep that in mind when you are choosing the analytic method - the most difficult that I'd press for is first order logic. Having a machine come along behind the human analysts and perform the more difficult forms of analysis is basically what Deep Green [0] was about. A more manual version of that would be the Lockwood Analytical Method for Prediction [1]. The funny thing about Deep Green was that developers had to address the fact that military commanders didn't like having a machine perform the planning process, so they had the algorithm include sub optimal options in the selection menu - allowing the human operator to feel good about choosing the obviously correct solution. [0] https://en.wikipedia.org/wiki/Deep_Green https://en.wikipedia.org/wiki/Deep_Green [1] http://lamp-method.org/ http://lamp-method.org/

by woodman 10y ago

IgorPartola 10y ago

Oh man. Nope, not familiar with ACH. Thanks for the link.

rdl 10y ago
It'd be fun to recreate http://competinghypotheses.org/ http://competinghypotheses.org/ in a more modern technology.
- woodman 10y ago
  It would be fun to implement it to take advantage of the benefits that a restricted natural language would bring, like Attempto [0]. That would allow you to really automate the evaluation of hypotheses against new data points. You might be able to wedge a SAT solver in there as well. I've got it on the ever growing todo list. [0] https://en.wikipedia.org/wiki/Attempto_Controlled_English https://en.wikipedia.org/wiki/Attempto_Controlled_English

Lxr 10y ago

Interesting, what about prior probability and Occam's razor? The 'process' section mentions "less consistency implies a lower likelihood" but makes no mention of preferring simpler hypotheses which I find surprising. Edit: 'Bayesian analysis' is briefly mentioned further down the page.

woodman 10y ago
ACH is just a very high level framework for shaking bias, the methods are implementation details. This is for humans, so you have to keep that in mind when you are choosing the analytic method - the most difficult that I'd press for is first order logic. Having a machine come along behind the human analysts and perform the more difficult forms of analysis is basically what Deep Green [0] was about. A more manual version of that would be the Lockwood Analytical Method for Prediction [1]. The funny thing about Deep Green was that developers had to address the fact that military commanders didn't like having a machine perform the planning process, so they had the algorithm include sub optimal options in the selection menu - allowing the human operator to feel good about choosing the obviously correct solution. [0] https://en.wikipedia.org/wiki/Deep_Green https://en.wikipedia.org/wiki/Deep_Green [1] http://lamp-method.org/ http://lamp-method.org/