A heuristic is a ‘rule of thumb’, a shortcut to solve a problem quickly in a way that usually works reasonably well. Heuristics are essential for problem solving because nobody has enough time or ability to solve every problem perfectly. Much of learning any field is learning increasingly effective heuristics for solving the problems in that field.
Consider this ‘shortest path’ problem. I’ve added two points A and B to this map of Paris: what is the shortest route between them? Without Google Maps or any other program it would probably take you quite a while to find the exact right answer, even if you used some smart tricks. How would you get a reasonably good answer in say, ten seconds?
If you were to draw a path from A to B with a pen or you finger, you would probably make use of the fact that short paths between two points tend to be close to a straight line. You could use a strategy like this: starting from A, every time you reach an intersection you travel down the road that points in the closest direction to B (without backtracking.) Using this strategy I’ve drawn a path from A to B in yellow.
See that the yellow path looks mostly reasonable, except that it turns to the left after crossing the river instead of turning right. This makes the path end up being a bit longer than if it had taken the route I’ve highlighted in dark blue. The strategy used is still a useful heuristic though, because it will work well enough for most problems, most of the time.
One way we use heuristics all the time is to quickly do statistical inference*.* Inference is just estimating a probability that something is true based on what you’ve seen before. If you look up and see dark clouds, you can infer that it’s likely to rain: you have probably seen dark clouds many times before, and a large portion of the those times it began to rain shortly after. In a sense we use inference for really everything we do. You probably pick up your cellphone without worrying it’ll be burning hot, and walk up stairs without worrying they will collapse. It’s not because these things would be impossible, but because it has happened so rarely to you (or not at all) that it’s extremely unlikely to happen now. Because humans are not computers, we don’t perfectly record every thing that’s ever happened to use and count exactly how many times in each kind of situation some outcome occurred. You don’t look up and see dark clouds and think ‘I’ve seen clouds like this 485 times before, and 37% of those times it rained within the next 30 minutes.’ We rely on more imprecise mental heuristics to get a rougher estimate. When we need a more precise estimate we have to rely on more precise data collection and statistical models, like those used to make weather reports.
We also have to make inferences in social settings, often relying on quite superficial information. If you are quickly looking for a doctor in hospital hallway, you’re probably going to look for someone in a lab coat or scrubs. This method is not 100% effective, you may miss a doctor who happens to be in regular clothes, or find someone wearing a lab coat who’s not a doctor. You could take time to gather more data, or spend time thinking about every visual clue, but if you’re in a rush then that’s too slow.
There’s nothing wrong in and of itself with using superficial information, in fact, you wouldn’t be able to get through your day otherwise. But the problem arises when you lose track of the limitations of heuristic strategies, and start believing their conclusions to be certain despite any evidence to the contrary. Quickly judging someone based on their appearance can be useful in some ways, but it’s a big problem if:
In other words, your first impressions are inaccurate and/or you stubbornly stick to your first impressions despite new information. A lot of prejudice like racism, sexism etc. can be thought of as examples of these two errors. For example, suppose Alice erroneously believes that people from some group are on average not as trustworthy. When Alice meets someone from that group, she therefore assume that they are probably not very trustworthy. But then even after Alice has interacted this person multiple times and they have always proven trustworthy, she she only slightly improves her assessment of them. Clearly, we would therefore consider Alice to be prejudiced against people from this group.
Cognitive biases like these are covered in more details here, but research shows that people tend to not update enough from their first impressions. And initial inferences can be inaccurate because of a variety of cognitive biases, like recency bias.