Lewis: Can AI be unbiased?
When I’m driving down the highway, I like to joke that everyone who drives faster than me is a maniac and everyone who drives slower is an idiot. My speed is the right speed. Am I biased? Sure.
I submit we are all biased in more ways than we might think. Everyone has views and opinions on just about everything. In many, if not most cases there is no absolute right or wrong, viewpoints are just that — they are our opinions.
Most of us will say that we would like more “unbiased” news reporting but what does that mean? A reporter might describe a weather event as epic or describe a particular innovation as being revolutionary. This may be the case or maybe they are just trying to embellish their story to get our attention. Is this bias?
Bias gets a lot of attention in politics and politized issues. Groups have gone as far as creating a media bias chart that purports to categorize media outlets from left to right. The data is subjective at best. I would bet that most news organizations believe that they are the truth tellers — they are proving the real facts and it’s all of those other people who are biased.
What is important to understand about AI is that its “intelligence” comes from what are called Large Language Models. Essentially, we train AI by feeding in enormous amounts of content. Content that was written by all of us biased humans. The AI system doesn’t really form opinions, it simply adopts its “perspectives” and knowledge from the content provided.
Support Local Journalism
The good news is that these models are very large, and AI uses probability to determine the best response. This means that, given a vast amount of information, AI systems are unlikely to offer highly polarized responses to questions. The bad news is that regular humans, with our personal beliefs and biases, are the ones supplying these AI systems with reference content. So, it would be easily possible to build extremist AI.
One way that AI systems try to avoid bias is by having the system not attempt to provide definitive answers to many subjective questions. If you ask ChatGPT “How was the world created?” you don’t get a single answer. Instead, it provides a listing of beliefs that reflect the most common scientific and religious perspectives.
AI systems often attempt to avoid bias by simply answering questions with replies like: “There are many perspectives…” When I ask ChatGPT polarizing questions, it always manages to come up with a diplomatic answer. As AI systems proliferate, it is almost a certainty that systems will be programmed to reflect a more conservative or liberal bias.
Want the news to come to you? Get the top stories in your inbox every morning. Sign up here: VailDaily.com/newsletter
This evolution is not due to a flaw in the technology, it simply highlights the fact that AI’s “intelligence” comes from the content we provide to it. Interestingly, it is pretty much the same with us. If we are exposed consistently and exclusively to biased content, we are more likely to adopt those viewpoints.
Bias in AI systems is inevitable. This will not be a problem as long as we understand that AI is not “all-knowing” — especially when asked subjective questions, it reflects the information/knowledge it is given. We interact with people every day who have different biases — we should expect AI systems to be no different.
Mark Lewis, a Colorado native, had a long career in technology, including serving as the CEO of several tech companies. He’s now retired and writes thriller novels. Mark and his wife, Lisa, and their two Australian Shepherds — Kismet and Cowboy, reside in Edwards.
