I was reading the AP news feed and came across this article about how a man was sent to prison because of Artificial Intelligence, AI. An AI, developed by a software company using a secret algorithm, led prosecutors to believe the man, Michael Williams, committed murder. Eventually a judge dismissed the case due to lack of evidence, but by then Williams had spent nearly a year of his life in jail. As we use AI for high stakes decisions, such as in court cases, school admissions, or employee hiring, we need to be sure that we use the tools in an appropriate way so that people are treated fairly. In other words artificial intelligence must be ethical.
What Is AI?
AI is simply the use of a computer algorithm as a decision aid. You feed information into the computer, such as an audio clip of city background noise, and it can help search for specific sounds. Rather than programming the computer directly to detect a given sound, you “train” the computer by giving it samples of the sound you want it to detect. Over time the computer learns what gun shots sound like, and given a particular sound to evaluate, it compares it to its prototype of a gunshot. A problem with this approach is that there is no guarantee that the computer identifies the sound correctly, and as the AP article points out, the AI can mistake a car backfire or fireworks for a gunshot.
A limitation to this sort of AI is that it makes decisions similarly to the way people make decisions based on prototypes. A prototype is a person’s mental representation of what something should be like. For example, if you are hiring a high level executive, research suggests that the prototype for many people is a male because most of the executives they have known are males. When evaluating a female candidate, if the hiring manager relies only on the prototype, the woman will not look like a good candidate and will be unlikely to be hired, thus the glass ceiling. In my field of industrial-organizational psychology there has been a century of research on using assessment tools that avoid this sort of prototype thinking. Rather than basing hiring decisions on gut feelings that this person is or is not a good fit, decisions are based on a careful analysis of job requirements and personal talents. Although AI might seem to be the way forward, in many ways it is a step back, just substituting the prototypical thinking of humans with prototypical thinking of machines. If we train an AI to pick executives by feeding it examples, that AI is going to learn in much the same way as a person. Introduce the AI to hundreds of executive who are mostly men, and the AI will learn that executives should be men. This can produce adverse impact against women–the AI gives in women less of a chance of being hired than men.
Artificial Intelligence Must Be Ethical
There are growing voices warning about the ethics of AI, or more correctly about how we use AI. Keep in mind that AI is just a tool, and no tool should be used blindly. AI merely provides information; we should not turn the decision process over to the machine. We should remember that AI judgment is not perfect, and even if it is more accurate than human judgment as sometimes claimed, it can still make mistakes. What it tells us should be considered one piece of information to be weighed with other pieces of information. We should not be intimidated by an algorithm because we do not understand it, and give it undue weight in a court case or other high stakes situations. The ethical use of AI requires that we understand how the AI works so we can know how best to utilize it as one of many decision tools, and be willing to ignore it when what it tells us doesn’t make sense or cannot be confirmed by other tools. If we are going to use it in ways that significantly affect people’s lives, artificial intelligence must be ethical.