I found this article in my Fortune magazine this month and thought it was really thought provoking. It brings a new perspective to the artificial intelligence (AI) craze which I hadn’t considered before. The new information is helping me to reconsider some of the rosier predictions of the uses of AI.
The constant gathering of new information is why there are times when I feel I should adjust my views. When people don’t adjust their view in light of new information, it is referred to as the commitment bias. In the commitment bias, once someone has committed to a point of view, they tend to ignore valid information that conflicts with their view. As humans, we are all guilty of this bias.
Like anything in life, you get the bad with the good. Our relationships have both aspects. They are also a major component of any decision, such as a purchase of a stock or selection of a career path.
The majority of AI now is called supervised learning. The algorithms in the software learn from the data that has been feed to it. AI doesn’t evaluate if the information is right or wrong. It just absorbs it and adds it to its knowledge bank.
The huge advantage of AI systems is that they can process data much more quickly than the human mind, but they haven't developed methods to discern which information is right and wrong.
This article described an AI system that Microsoft put in place to field customer calls. It did an outstanding job learning from the individuals it "spoke to". Unfortunately, many of the individuals conveyed extremely negative biases and prejudices to the system. In a short time, the system was insulting customers. Needless to say, Microsoft quickly took it off line and issued an apology. Microsoft corrected the issue to some degree by “blacklisting” conversational items that the system could learn from.
Clearly, there are still some significant hurdles to cross before AI can become the future of logical thinking and action (Think Dr. Spock in Star trek). I am not sure that some of these hurdles can ever be overcome. While AI may not be as useful as I once thought it would be, it will still have an enormous number of great applications, but on a more limited basis - at least for now.
Most importantly, these unfortunate results say volumes about us as humans. Our prejudices and biases are the result of what we hear and are subjected to. We then absorb, repeat and spread both good and bad. Unfortunately, AI systems decrease the “learning” time because they process input so quickly.
