As a scientist, I am motivated to build an objective model of reality. Since we always have incomplete information, it is eminently rational to construct a Bayesian network of likelihoods — assigning a probability for each possibility, supported by a chain of priors. When new facts arise, or if new conditional relationships are discovered, these probabilities are adjusted accordingly — our minds should change. When judgment or action is required, it is based on knowledge of these probabilities. This method of logical inference and prediction is the sine qua non of rational thought, and the method all scientists aspire to employ. However, the ambivalence associated with an even probability distribution makes it terribly difficult for an ideal scientist to decide where to go for dinner.
Even though I strive to achieve an impartial assessment of probabilities for the purpose of making predictions, I cannot consider my assessments to be unbiased. In fact, I no longer think humans are naturally inclined to work this way. When I casually consider the beliefs I hold, I am not readily able to assign them numerical probabilities. If pressed, I can manufacture these numbers, but this seems more akin to rationalization than rational thought. Also, when I learn something new, I do not immediately erase the information I knew before, even if it is contradictory. Instead, the new model of reality is stacked atop the old. And it is in this sense that a mind doesn't change; vestigial knowledge may fade over a long period of time, but it isn't simply replaced. This model of learning matches a parable from Douglas Adams, relayed by Richard Dawkins:
A man didn't understand how televisions work, and was convinced that there must be lots of little men inside the box, manipulating images at high speed. An engineer explained to him about high frequency modulations of the electromagnetic spectrum, about transmitters and receivers, about amplifiers and cathode ray tubes, about scan lines moving across and down a phosphorescent screen. The man listened to the engineer with careful attention, nodding his head at every step of the argument. At the end he pronounced himself satisfied. He really did now understand how televisions work. "But I expect there are just a few little men in there, aren't there?"
As humans, we are inefficient inference engines — we are attached to our "little men," some dormant and some active. To a degree, these imperfect probability assessments and pet beliefs provide scientists with the emotional conviction necessary to motivate the hard work of science. Without the hope that an improbable line of research may succeed where others have failed, difficult challenges would go unmet. People should be encouraged to take long shots in science, since, with so many possibilities, the probability of something improbable happening is very high. At the same time, this emotional optimism must be tempered by a rational estimation of the chance of success — we must not be so optimistic as to delude ourselves. In science, we must test every step, trying to prove our ideas wrong, because nature is merciless. To have a chance of understanding nature, we must challenge our predispositions. And even if we can't fundamentally change our minds, we can acknowledge that others working in science may make progress along their own lines of research. By accommodating a diverse variety of approaches to any existing problem, the scientific community will progress expeditiously in unlocking nature's secrets.