Editorial on "Mirrored Images"

(from our correspondence files - this comes up often)

        Some of the private sites displaying features found in Mars images have taken to using a technique called "mirroring", whereby an axis is drawn and part or all of a feature is reflected across that axis to create a new feature with bilateral symmetry. Significance is often then attached to the original image because of its striking appearance after being mirrored.

        Anyone can make familiar shapes out of truly random patterns by mirroring. This practice is demonstrated by classical Rorschach inkblot tests, which are often mirrored to make the inkblots look less random than they really are. The first rule in science is to devise a protocol to test hypotheses that cannot be influenced by the biases of the experimenters or subjects. Mirrored images fail that criterion badly because the viewers biases come into full play when judging mirrored images. I see no easy way to set up unbiased, objective criteria for judging features once they have been mirrored.

        Mirroring half a feature throws away key information that might help determine artificiality; specifically, the real data that was replaced by the mirrored feature. For example, in the original Cydonia Face image from Viking, we had just half a face. We could then predict that, when the other half was finally seen, it would look random if the mesa was natural, and somewhat like a mirrored half face if it was artificial. But mirroring the original half face would tell us nothing useful. It would merely create unrealistic expectations for the unseen side that even an artificial object could not fulfill because of natural asymmetries and erosion. A posteriori findings (such as half a face) almost never have significance, but mirroring them will create the impression that they do. Only unbiased things such as predictions made before anything is known about their probable outcome have value to distinguishing hypotheses.

        On the Meta Research site, we do something that creates similar biases when we present versions of images with features colored in, or separate pictures of terrestrial counterparts of features suggested by the pictures. Doing so can create bias, making the viewers' responses invalid to scientists testing hypotheses about the features. However, that is done for the benefit of viewers who don't plan to do any analysis or hypothesis testing of their own or to participate in such tests. The conclusions presented with these images were drawn before such "keys" existed. The keys simply provide the quickest way to communicate results to a wide variety of people having a vast range of perceptual capabilities.