tag:blogger.com,1999:blog-1689592451067041352.post4571221512561396078..comments2023-05-08T07:41:01.071-07:00Comments on DiEbLog: The Search Problem of William Dembski, Winston Ewert, and Robert MarksDiEbhttp://www.blogger.com/profile/02099109109735165335noreply@blogger.comBlogger1125tag:blogger.com,1999:blog-1689592451067041352.post-30184999987758721582018-01-30T15:47:56.265-08:002018-01-30T15:47:56.265-08:00Hi DiEb,
Two areas where target sets are detached...Hi DiEb,<br /><br />Two areas where target sets are detached from objective function extrema are surrogate functions (i.e., when you want to optimize one function, but it is intractable, so you settle for a surrogate whose extrema are hopefully close to the true target), and unreliable objective functions (such as noisy evaluations, which means you may actually be optimizing towards the wrong target). These are ever present issues in applied data science and I'm sure you can think of other cases without too much effort. So, allowing different possible relationships between target sets and information resources like objective functions makes sense. If nothing else, it is a more flexible model that includes the usual state of affairs where the target set is fixed to be the minimum/maximum of the function as a special case. What matters, in the end, is the degree of dependence between the target set and the objective function, namely, how much information about the target can you gain by exploring the objective function?<br /><br />Since DEM have evolutionary concerns in mind, it also makes sense that their targets need to be defined independently from the objective function, or else when we say "Evolution can find functional small targets in large spaces, such as eyes, echolocators, wings, and flagella" we are essentially making a tautological statement, because the targets of evolution are defined simply to be the things it produces. Of course evolution can produce what it produces. However, these targets are identifiable independently of the fitness functions of nature, and there is no a priori guarantee that such fitness functions will have maxima that coincide with these targets (and have smooth monotonically increasing fitness slopes leading up to them). If you want to say the targets are exactly those places where the fitness functions are maximized, there is no theoretical guarantee that the targets will exactly overlap with the set of living things. Unless you can show otherwise. I'm open to being persuaded.Unknownhttps://www.blogger.com/profile/06929435419596641287noreply@blogger.com