Blithering Idiots 2.0

By Gordon Hopkins While I have shared my panic over the seemingly unstoppable juggernaut that is AI (Artificial Intelligence) with anyone who would listen (and anyone who couldn't get away), it now seems that AI may be it's own worst enemy. The big concern about AI is that stuff produced by AI is becoming ubiquitous. Soon, it will every. Anyone on social media is probably already aware that their feed is filled with AI-generated content. If you aren't aware of it, I'll tell you right now, it is. You just haven't noticed it yet. And that is where the problem for AI comes in. AI language models, like OpenAI, “learn” by gorging itself on content mostly found on the internet. Up until now, that content, be it writing or pictures or video, was produced by live human being with the needed imagination to produce such content.However, as noted above, more and more of the content on the internet is being produce by AI.Recent studies (and nearly all studies on AI are recent) are showing that when AI is “trained” using AI-produced content, the results eventually deteriorate to incomprehensible gibberish. The official term is “Model Collapse.”It's the same principal as making a copy of a copy of a copy. The more copies you make, the worse it gets. The same thing can happen when you clone a clone of a clone. By the time you get to the fifth or sixth generation, that clone is probably going to need help tying his shoes. It is looking more and more like robots need us more than we need robots. In one attempt at some hot, AI-on-AI action, an AI language model started out on a diet of human-created text. Each subsequent generation was “trained” using the text created by the AI. After a few generations, the final result was nothing but the phrase “black@tailed jackrabbits” repeated ad infinitum.I am immediately reminded of the scene in “The Shining,” when the audience learns the novel Jack Torrance has been working on all winter at the Overlook Hotel is the sentence, “All work and no play makes Jack a dull boy,” over and over again. So, basically, we've invented a way to make computers psychotic. The solution posited by some researchers is to make AI-generated text identifiable so it doesn't creep into those language model. This echoes the call for writers and artists and all all other carbon-based human content creators, who want people to know when they are seeing something created by a human or regurgitated by a machine. A story by tech reporter Mack DeGeurin in Popular Science sums it up pretty well, I think, “Without those safeguards of human training data, the internet risks being flooded by a wave of AI vomit. Nobody wants that.”

Gage County Newsletter

Defendant Found Not Guilty in Criminal Trespass Case But Guilty for Failing to Appear for Court

Related Articles

Remembering

The Stuff Sequels Are Made Of

Trendsetters in the News

Gage County Newsletter