Re:Coders Automating Their Own Job
by godelski on Wednesday October 03, 2018 @01:09AM
(#18127760)
In the effort of simple counter proof: you can dig a hole with a spoon.
That requires a substantial amount of hard work but unlikely results in a measurable amount of success. Certainly one less than would be if the proper tools were used.
And I do agree with you that it is not the only factor. But this is why I differentiate cause and correlation. And the disagreement is likely to how we are using those terms. I am using "cause" in the colloquial sense (because context from above) of how most people do not see multivariate equations, but single factor (which is an unfortunate thing) ie. "cause and effect" (but we both know better). I am using "correlation" in the more rigorous manner meaning "related to; a statistical variable; or to be in association with".
And going back to your RL example, I think you mention one part which proves my point as well.
> the exploration/exploitation strategy
You may notice that many learning algorithms have a substantial amount of failure. After all, it is the failure that leads to success. BUT a key point is that it is not the effort put into the attempt that led to success, it is the adaptation of a new strategy after recognizing the failure. You could just as easily never visited that attempt method and thus your efforts would lead to a higher success/work factor. You could also never change strategies and your efforts would result in a substantially lower success/work factor. More explicitly "hard work" is a variable that is coupled with "strategy", and the value of "hard work" can be astronomically high but the value for success can be zero. And for this reason we say it is not "the cause" but rather "a factor" or "correlates with".
Also, as mentioned in a grandchild to my post, this is a saying which I gladly will admit does not cover the entire scope of what factors lead to success. Even the unpacking of implicit information cannot do that, as a full description of the path to success is rather cumbersome.
That requires a substantial amount of hard work but unlikely results in a measurable amount of success. Certainly one less than would be if the proper tools were used.
And I do agree with you that it is not the only factor. But this is why I differentiate cause and correlation. And the disagreement is likely to how we are using those terms. I am using "cause" in the colloquial sense (because context from above) of how most people do not see multivariate equations, but single factor (which is an unfortunate thing) ie. "cause and effect" (but we both know better). I am using "correlation" in the more rigorous manner meaning "related to; a statistical variable; or to be in association with".
And going back to your RL example, I think you mention one part which proves my point as well.
> the exploration/exploitation strategy
You may notice that many learning algorithms have a substantial amount of failure. After all, it is the failure that leads to success. BUT a key point is that it is not the effort put into the attempt that led to success, it is the adaptation of a new strategy after recognizing the failure. You could just as easily never visited that attempt method and thus your efforts would lead to a higher success/work factor. You could also never change strategies and your efforts would result in a substantially lower success/work factor. More explicitly "hard work" is a variable that is coupled with "strategy", and the value of "hard work" can be astronomically high but the value for success can be zero. And for this reason we say it is not "the cause" but rather "a factor" or "correlates with".
Also, as mentioned in a grandchild to my post, this is a saying which I gladly will admit does not cover the entire scope of what factors lead to success. Even the unpacking of implicit information cannot do that, as a full description of the path to success is rather cumbersome.