Podcast: Download
Subscribe RSS
I’ve been playing with an AI art generator, thinking it could provide me with some post images. So far, it’s not been much help. The image here was generated with the prompt “middle age couple on couch kissing”.
The image it okay, but there’s more to the story. This was try sixteen out of twenty. It was the only one that showed anything close to a kiss. Beyond that, I would not call this couple middle age. Every one of the men’s hair was at least half grey or white. Two were partly bald, but none more than that. Half the women were the age of the men, but the other half were decades younger than the man. Then there was the image of two women rubbing noses as a man behind one of them was smelling her hair. And of forty-one individuals in the images, the only one who was not lilly white was the hair sniffer.
The problem is the training set used to teach the AI. Imagine if I had tried my prompt on a program trained on porn. To get what I wanted, I would have had to ask for a “clothed couple” and specified kissing on the mouth. Even then, I doubt I would have generated something I could show here. What if I tried it on a program trained entirely on Disney animated movies?
So what, you’re wondering, does this have to do with marriage? You and I are a lot like AI programs. What we think and do is strongly influenced by our “training set”. And unlike an AI, we can’t wipe our memory and start over. As we add new data that does change things, but the old is still there, and human nature is to default to what we learned first.
This means some of the fights you have with your wife are because one or both of you reacted from a flawed training set rather than logically examining the information and making an unbiased decision. You allowed your expectations and fears to colour what you thought.
The advantage you have over an AI is that you can question yourself. You can ask if you keep getting the same answer because of what your wife is doing and saying, or because your training set is making you see things that aren’t there.
A post worth reading:
ScienceDaily | Robots turn racist and sexist with flawed AI, study finds ◄ Neural networks built from biased Internet data teach robots to enact toxic stereotypes
Image Credit: © Paul Byerly via AI
Shop Amazon ♦ Shop to give links page
We’re donation supported Thanks for your help!
This post may contain affiliate links, see my disclosure for info.
Leave a Reply