As I’m writing this in early February, 2023, ChatGPT is all over the news. It’s all over the tech podcasts and news sites. I’ve even seen a newsletter for copywriters advertising a class for writing copy with ChatGPT.
More reasons why technology is not magic.
I’ve seen this at least twice and maybe three times before. A few years ago it was machine learning. Long before that, in the 1990s it was fuzzy logic. I think there was another AI (Artificial Intelligence) alleged breakthrough in the two thousand zeroes.
Each time it’s going to replace humans and each time it doesn’t. Each time the hype and hysteria fade and frequently there are some fairly embarrassing faceplants. For machine learning, after all the breathless hype about amazing image recognition, several models were broken by taking recognized pictures and changing a couple of pixels. I’ll repeat that, machine learning recognition of items in photographs was broken by changing just a few pixels. In one example, a toy turtle was labeled a gun after a few pixels were changed.
Yes, there a few successes. Facial recognition has progressed by leaps and bounds. There are also dedicated efforts to finding ways to mess it up, including specialized makeup and specialized eyeglass frames. What wasn’t mentioned was how much facial recognition is hampered by face masks, which have now become normalized in most of the world and are still mandatory in many places.
Getting back to ChatGPT, the current media darling, there’s already been multiple examples of ChatGPT being asked to write an article and getting basic information about the topic wrong.
Two already known circumstances where AI and machine learning fail:
- They can’t understand context, and
- They can’t understand or recreate humor. At all.
There’s probably more.
Humans Do Amazing, Almost Magical Things, All the Time.
Meanwhile, every time researchers try to copy something humans do all the time with technology, it turns out it’s really hard.
Robots transitioned to stationary or wheeled or bug-like decades ago because walking is really hard. We actually go through a series of controlled falls when we walk. I think there’s something like seven different points of stability and balance in our bodies when we walk, which we don’t notice but programming into robots is really difficult.
The first Avatar movie cost so much in part because James Cameron developed whole new headgear to watch and record the actors’ eyes and tongue while acting. He did this because eye and tongue movements are two things computer animation still couldn’t replicate convincingly, so he recorded that from the actors to add into the animated characters.
We can look at pictures that are fuzzy, pixelated, or poorly focused, and still recognize the object.
From what I’ve seen, ChatGPT is useful for quickly producing a lot of text that then needs to be edited and reviewed by a human. And that’s only if the person asking the question does a very good job of setting the parameters. And only if the person editing the response already knows the topic.
Technology isn’t magic, no matter how much we keep trying to convince ourselves it is.