 |
The fervor over bad AI is another reflection of our times and is mostly rooted in the fear of degraded trust. New technologies can be disruptive and scary but, in most cases, they have moved our societies forward in new ways. AI is already moving us forward and as we have done with other technologies, we need to understand how to use it safely and guide its use for societal advantage.
You have all heard about bad things like fakes/misuse, bias, job replacement and even extinction through the misuse of AI. One of my favorites is where AI hallucinates offering totally wrong answers.
Misuse and fakes are one of the most virulent problems we see with AI particularly with attempted manipulation over political positions. The big tech companies and the government are working to add tools to help with trust regarding what we read and see. Labeling AI generated and modified media and information is one approach being used. A variance to the labeling tool is one that gives you information on the source of the content and who or what created it. These tools put the decision in your hands to decide trust. Instagram is putting an AI-generated/modified detector label in place that lets you know. You already challenge trust when you ask across social media for a contractor recommendation. You choose which one to use based on how many times a positive reference comes up from varied sources and if someone you know, and trust offers a recommendation. This is how we will deal with AI generated information and media by looking for the labels and then challenge the validity.
Bias and hallucinations are byproducts of the data used to train an AI context engine. Engines like ChatGTP are designed to give you an answer and if it does not have a well-defined response, it may choose things that come close and may be totally wrong. Bias is similar in that if the data used for training does not have much information on a particular group’s view the engine may derive a one-sided response. Thankfully these limitations are well known by our technologists, and they are applying guard rails to do things like saying “I don’t know” if the answer is just made up. Also, they are using data training methods that ensure a broader representation of information is used as well as adding ethical controls for a response.
The other way to solve this is to tell what the application constraint is for an AI model, so we know whether its valid or not to trust for our use. Transparency is the key! Remember in my last column about the human doctor knowing to ask about behavioral information that the AI doctor was not trained on.
There is no doubt that AI will replace some current jobs. It is not the first-time technologies have done this, but in most cases, it has resulted in a more efficient and productive world.
Those of us in the “boomer” generation must admit we enjoy having so much more information available to us through the internet that we didn’t grow up with. AI will force us to use it as a partner to learn how to do higher level complexity jobs rather than replace us entirely. There has never been anything easy about transitions like this though, but society has often benefited.
There are many more approaches and tools being added to our AI machines to control bad actions and we could only touch on a few in this article. Security is another huge area that is being built into AI machines.
Keep reading and listening about what companies and governments are doing to build your understanding and ability to trust and use AI for all the good things it offers our society.
Mark Conde of Murphy is a retired chief information officer and self-described technology nerd. Email him at
jmconde818@gmail.com.