Starting more simply, without even going into the details of the algorithmic models yet, if we can assume that evolutionary processes had a role in our development then it is all the kinds of experiences that we normally encounter which played a role in evolving the mind and it’s training. We rely on experiences coming from a “common source in the world”. When this becomes less true, that is when we can have observable bias emerge.
The classic experiment demonstrating the Anchoring bias prompts individuals to focus on an initial numeric memory recall task, then subsequently on an unrelated but numerical estimation task, which in turn demonstrates an influential bias of the first task on the second. We can certainly demonstrate the bias, but keep in mind the experiment itself uses an artificial series of events: it’s unpredictable!
Perhaps getting the estimation wrong makes sense? If you got the estimation right you might still wonder if the second task followed the first for a reason? Getting things wrong to speed learning new things is exactly what we should be doing isn’t it? Perfectly rational!
Studying the HTM actually offers much more explanatory and potentially useful models of biases. We’ll take a another pass at the anchoring bias reflecting more on the actual HTM soon.