Very interesting article thank you for sharing your thoughts !
Still, while reading I shuddered on your description of the scientific method. Because the method you describe is the Frequentist one, not the Bayesian one that I follow and love.
The Bayesian one fixes both the problems you mention in this article:
- 1) Potential false positives: In bayesian thinking, no theory ever "pass" or is "rejected", their credibility just gets updated as more experiments are done.
- 2) Asymmetricity: In bayesian thinking, tests can update credibility of propositions in both ways. And credibility of theories are updated relatively to all their alternatives. In your example, ather was a good enough theory until the Michelson-Morley experiment lowered largely its credibility, making alternative theories where results of the experiment are less surprising more credible, nothing "wrong" with that.
Still Bayesianism doen't solve everything ! In fact how do we come up with alternative theories ? When do we need some ? What propositions are even stated ? This is good fuel for thinking ...
I fully agree with you! I thought about adding it to the post but it would have been too much of a detour. Bayesianism assigns gray-scale credibility scores to propositions, and it is beautiful because it allows us to maintain quantitative uncertainty (or certainty) scores on our convictions. Friston's Free Energy Principle and Solms' affect-based consciousness theory both require uncertainties, so it's a must. I think frequentism is more used because we like black and white answers, and also because it works when we are close to 100% certainty. But in the creative chaos phase, when we are just a bit better than random, maintaining credibility scores is very important.
The reason I didn't bring this up because it would have taken the spotlight off the question I wanted to introduce. Credibility scores are still orthogonal to importance scores. As you say, Bayesianism still doesn't answer which hypothesis we should invest in to test. It also doesn't handle the most interesting problem: self-transformation, where the organism changes itself, its value landscape, and the reality it sees. That will come later, in the 3rd and 4th Ps (perspectival and participatory).
That said, for the 1st and second Ps (propositional and procedural), I believe that Reinforcement Learning + Bayesian decision making is sufficient as a formalism, of course with a lot of research to be done to make it work.
thanks for posting this, I eagerly await the articles which discuss the other types of knowing.
Thank you, will do.
Very interesting article thank you for sharing your thoughts !
Still, while reading I shuddered on your description of the scientific method. Because the method you describe is the Frequentist one, not the Bayesian one that I follow and love.
The Bayesian one fixes both the problems you mention in this article:
- 1) Potential false positives: In bayesian thinking, no theory ever "pass" or is "rejected", their credibility just gets updated as more experiments are done.
- 2) Asymmetricity: In bayesian thinking, tests can update credibility of propositions in both ways. And credibility of theories are updated relatively to all their alternatives. In your example, ather was a good enough theory until the Michelson-Morley experiment lowered largely its credibility, making alternative theories where results of the experiment are less surprising more credible, nothing "wrong" with that.
Still Bayesianism doen't solve everything ! In fact how do we come up with alternative theories ? When do we need some ? What propositions are even stated ? This is good fuel for thinking ...
Thank you, awesome comment!
I fully agree with you! I thought about adding it to the post but it would have been too much of a detour. Bayesianism assigns gray-scale credibility scores to propositions, and it is beautiful because it allows us to maintain quantitative uncertainty (or certainty) scores on our convictions. Friston's Free Energy Principle and Solms' affect-based consciousness theory both require uncertainties, so it's a must. I think frequentism is more used because we like black and white answers, and also because it works when we are close to 100% certainty. But in the creative chaos phase, when we are just a bit better than random, maintaining credibility scores is very important.
The reason I didn't bring this up because it would have taken the spotlight off the question I wanted to introduce. Credibility scores are still orthogonal to importance scores. As you say, Bayesianism still doesn't answer which hypothesis we should invest in to test. It also doesn't handle the most interesting problem: self-transformation, where the organism changes itself, its value landscape, and the reality it sees. That will come later, in the 3rd and 4th Ps (perspectival and participatory).
That said, for the 1st and second Ps (propositional and procedural), I believe that Reinforcement Learning + Bayesian decision making is sufficient as a formalism, of course with a lot of research to be done to make it work.