Part One of The Theory of Moral Sentiments by Adam Smith is called “Of the Propriety of Action”. Smith argues that we naturally share the emotions and to a certain extent the physical sensations that we witness in others. “Sympathy” is a term Smith used for the feeling of moral sentiments.
In Section One, Chapter Five, Smith writes
In all such cases, that there may be some correspondence of sentiments between the spectator and the person principally concerned, the spectator must, first of all, endeavour … to put himself in the situation of the other, and to bring home to himself every little circumstance of distress which can possibly occur to the sufferer. He must adopt the whole case of his companion with all its minutest incidents; and strive to render as perfect as possible, that imaginary change of situation upon which his sympathy is founded.
After all this, however, the emotions of the spectator will still be very apt to fall short of the violence of what is felt by the sufferer. Mankind, though naturally sympathetic, never conceive, for what has befallen another… That imaginary change of situation, upon which their sympathy is founded, is but momentary. The thought of their own safety… continually intrudes itself upon them…
The modern word “empathy” is the capacity to step into the shoes of another person and feel their pain or joy from within the other person’s frame of reference.
Adam Smith suggests that if we hear a neighbor just experienced the death of a loved one, then we can briefly experience some sadness on their account. The more we put ourselves in their shoes, the more sadness we can experience on their behalf.
We usually think of it as a nice thing to have empathy for others. It can also be instrumental to be able to think through the perspective of another person, in order to predict what they will do next. In practical dealings, it is an economic advantage to make accurate predictions about future behavior.
If I work backward through my 2020 paper “My Reference Point, Not Yours”, then I can start by saying that people can sometimes predict what others will do.
When I encouraged subjects to think about a specific person in a specific situation, they made accurate predictions. Their capacity for sympathy seemed sufficient. They made more money in the experiment if they reported predictions that matched the actual behavior of what people in that situation had done.
However, the first prediction task in the experiment was to make predictions about a group of people who experienced a variety of starting points. Subjects did not do so well here, even though there was a financial incentive to get the right answer. Subjects exhibited a bias as if they assume everyone shares their own reference point, even if they are told that is not the case.
Smith says that people can put themselves in the shoes of another. To some extent, everyone experiences visceral pain when they pass a bad car accident, even though they themselves are unharmed. Smith also points out that we probably experience more upset from losing one of our own fingers than from learning that a hundred million people died on the other side of the world. Sympathy is ubiquitous and also limited.
We can’t be deeply sympathetic to a billion people at a time. That could be a reason why machine learning algorithms can make better predictions than humans in certain circumstances.
Economist Writing Every Day 