Today I want to talk about forgiveness of mistakes. In particular, I want to ask whether we forgive more mistakes made by humans or by non-humans, algorithms, robots, and so on.
Let’s start by imagining that you ask somebody to go to the bank for you or to go to the cleaners for you, and that person makes a mistake. Now imagine you ask a robot to do the same task, and the robot makes a mistake. Imagine you ask somebody to drive to pick up your kids, and they are in a car accident. Now imagine that you ask a robot to drive to pick up your kids and they are in an accident.
In each of these scenarios which will we have a worse reaction to? Which will we assume more from and be less likely to forgive? One day, things might be different, but right now we get more upset, angry, and less forgiving of mistakes from algorithms. Before we explore big mistakes, like driving and so on, let’s think about the frustrations we have when we ask one of those agents to do something.
For example, I continuously try to get them to create an image of me with half a beard, and they can’t. I guess symmetry is the priority. By the way, if you ever try this and are successful, please let me know the prompts you created. Now, even if we think these agents are amazing and wouldn’t have imagined their capabilities a few years ago, they are nevertheless frustrating. And the frustration is that I gave a simple instruction. What’s the problem? When I ask one of those agents to draw a picture of me with glasses and half a beard and I get things that are not what I asked for. The problem is, of course, that I don’t understand how they work. They don’t work like humans, and they don’t make corrections based on my feedback or more specific instructions. Part of the issue is that when an agent or an algorithm makes the mistakes, we don’t understand where the mistake is coming from.
When a human makes a mistake, we’ve met humans before, we know they make mistakes, and we accept it as part of the standard. We know people make mistakes. When an algorithm or a robot makes a mistake, this is new to us, and we ask wonder what else could be wrong. With people, we know that people make mistakes, but we also know people generally do things right. The issue is that we understand human mistakes, and we understand humans also do lots of things right. When algorithms make mistakes, these are a different kind of mistake, and they make us wonder about the whole underlying structure. If I ask an algorithm for a joke and it gives me a pun, that’s a mistake. If I ask for an illustration of me with half a beard and it doesn’t do that, what else is it doing wrong. I think that one of the biggest issues is that it makes us wonder what else is broken and where could things break next.
Quite a while ago we did an experiment where people got feedback and advice when they invested in the stock market. They got it either from a human agent or from a computer agent. These were the days before their computer agents were so good, but we thought that they were. The advice was good for a while and then at some point, the advice was terrible. And the question was would we forgive a human or an algorithm faster? And people forgave a human faster.
There are two reasons. We know humans, and we know that they’re wrong from time to time. And after they forgave the human, they were more likely to take his advice. On the other hand, when algorithms get things wrong, it’s new technology, it’s baffling, and we wonder what else is wrong and what will it get wrong the next time? One day we’ll get better at forgiving robots and algorithms. But until then, we’re going to hold them to a higher standard than we hold ourselves. Just because we don’t understand the mistakes that they make to the same degree.
