Stupid AI tricks you need to stay aware of

The best AI tool is not intelligent. It’s not conscious. It doesn’t “know” anything.

It mimics those things by seeing patterns and spitting them back.

That’s all.

It does that at incredible speed, and it finds and spits back patterns of mind-boggling complexity. But there’s no intelligence involved.

That’s why AI makes bafflingly stupid mistakes.

Here’s what I mean: I gave my AI tool an appeal I’d written and told it to translate it into the style of Dr. Seuss. (It was a clean water appeal, and started with a photo of a very unclean-looking pond that was the water source for a community.

Here’s what I got:

Look there at that picture. Now what do you see?
That water looks murky as murky can be!
It’s green. It is gooey. It’s not very nice.
You wouldn’t want that in your soup — not one slice!

(It did the whole letter. Basically a three times the length of The Cat in the Hat, all about icky water.)

Okay, it really does sound like Dr. Seuss. And it is about the same topic as the original. But it gets weird in that last line: Why bring up “soup” in a discussion of water? And stranger yet, how does soup come in “slices.” There were a lot of weird bits like that.

No human would make those errors. Because we actually interact with soup. We know you don’t slice your soup. Not unless it’s gazpacho chilled a few degrees too much.

AI has never seen, touched, prepared, or tasted soup.

That’s the kind of mistake even a highly trained AI makes. Not a huge problem, but you’d better proof it carefully for that kind of thing if you use AI in your writing.

Some others:

• An appeal about an organization’s financial crisis included this: There are days I lie awake wondering how we will keep up. (Maybe the AI thinks fundraisers are nocturnal. I suppose some of them are.)
  
• In a story about a patient who received an important phone call from his doctor, the AI had it so the patient was both the caller and the recipient of the call. Not sure what happened to the doctor.
  
• This happens frequently: In fundraising written for email: it asks donors to “send a gift.” Even though it has specific instructions not to use direct mail language in digital copy.

These aren’t fatal flaws with AI. They’re easy to fix.

But they remind us two things:

1. You must, must, must edit AI-written copy. For hallucinations, violations of its training, and goofball stuff like the above.
   
2. As human as it sometimes seems when you interact with it, AI is not a person. You will do better when you keep that in mind.
   
   Over at Moceanic, we’ve built (and continue to refine) an AI fundraising tool. It’s called Copy Koala. It has a great track record of producing superb response in direct mail and email fundraising. It combines powerful training, smart input and prompting, and expert human editing. And the cost to use it is shockingly low. It might be right for your organization. If you have budget and the authority to spend it, consider setting up a quick online meeting with Sean Triner to see if Copy Koala might be right for you

More posts on AI and fundraising.