Can We Build Unbiased Robots?

People are persuaded by past experiences, upbringing, opinion, circumstance, mood, and pretty much everything around us. We might make a decision because of a scent that reminds us of a memory, or choose to treat someone a certain way because of a completely unrelated event that occurred in our life. Our understanding of a situation influences our thoughts and actions at any given moment.

In other words, people are biased.

Even people who consider themselves logical and rational are subject to bias. But that’s to be expected of free-thinking and emotionally intelligent beings. Thought, emotion, and feeling are what separate man from machine, right? While machines don’t have emotions, opinions, or an understanding of past, present, or future experiences, they aren’t immune to bias.

How are robots biased?

AI systems can make millions of decisions every minute through machine learning algorithms. Even the slightest hint of bias or prejudice in the program can run rampant in a short amount of time. A robotic arm’s preference to put things in the blue bin, or penchant for picking up apples over bananas, isn’t the issue here. As AI continues to become more advanced and find its way into more fields, the danger of biased robots grows.

We’re seeing machine learning being used in medicine, law, and law enforcement, it’s being used by the average person — through our smartphones, online searches, and social media — on a daily basis, and an increasing number of companies and businesses are relying on AI to make decisions.

Because people are biased by nature and people choose how AI systems make decisions, it follows that artificially intelligent systems are subject to the same bias as their creators. To make matters worse, algorithms compound initial bias.

So is it possible to build an unbiased robot?

There are tons of unbiased robots; your Roomba and your industrial motion control system aren’t judging you. The question is whether it’s possible to build an unbiased AI system. Methods such as removing unnecessary variables from an algorithm, or allowing a model to identify bias could help. However, the problem still remains: biased people programming the systems.

AI and machine learning will continue to grow in importance, so it’s important that we continue to strive for a way to eliminate bias in theses systems. If unbiased robots are impossible, we must move forward with that in mind.

How are your unbiased robots doing? Call 479-422-0390 for repair, maintenance, service, or support for Indramat industrial control systems.