in partnership with Providing Outreach  in Computer Science Bringing Bayesian solutions to real-world risk problems

# How certain are you of what you see? (confirmation bias)

Have a look at the image here (which we will call the source image) and make a judgement about what this image is. Make a mental note of how certain or uncertain you are about this source image.

Now look at this image which we will call the target image.

Return to the source image.  The chances are you are now certain about what this image is. Suppose you had to describe the source image. Would you describe it the same way now as you would when you saw it before the target image? It would be impossible to do so because it is impossible to get the target image out of your head once you have seen it.

So although the source image has not changed, what you see in it has changed simply because you were 'exposed' to the target image.

No matter how much you try to resist knowledge of the target image, once you have seen it, it becomes impossible for it not to affect your judgement about the source image. This is a well-known cognitive illusion called confirmation bias.

Here is another example. What characters can you see in the source image below?:

Again make a mental  note of how certain you are about these characters.

Now look at the target image.

If (as is very likely) you changed your mind about your belief in the certainty of the characters in the source image having seen the target image then again you are the victim of confirmation bias.

Confirmation bias appears in many different guises and has serious implications in all walks of life, but especially science, the law, and medicine.  For example, forensic and other scientists in legal cases have to make a judgement about a piece of evidence (the source) at the crime scene which could be:  a partial DNA sample;  a footprint;  a fuzzy image of a number plate;
glass fragments; etc.  In each case they have to examine the source and identify its characterstics (the DNA profile, the shoe type;  the characters in the number plate; type of glass, etc). Once they have done this they can then compare the analysis with a target. So is the source DNA profile consistent with that of the suspect's; are the visible characters in the number plate consistent with the defendant's number plate etc.

You may be surprised to learn that in such cases the scientists sometimes see the target before they perform their analysis of the source. This opens up the possibility of confirmation bias. Although scientists in such scenarios often say that their analysis is unbiased, it rarely is. Indeed the above examples provide some evidence of how unshakeable confirmation bias can be.