People usually ask questions to clarify some confusion. I don't know what yours is, but thought the article might be helpful since it elucidates this subject. Have you read it?

Organisms obviously don't directly optimize their genetic fitness. Deep Blue obviously doesn't directly optimize winning chess. If you want to economically predict their actions however, finding something they seem to optimize works as a rough model. This is easy if you know the process that made them. It's the nature of a rough model you can poke holes to it by finding exceptions, but this doesn't make the model useless.

Tim might be making a stronger claim than this. If that's the case I probably don't agree with it.