I’m coming up on the two-year anniversary of ending my prohibition on dealing with anything “AI“. November 2023 changed everything for me. And I mean everything.
In a past life, I vetted startup ideas for venture capitalists, sitting through a lot of demos and pitches by people who were absolutely positively convinced that they had a game-changing solution that would completely shift the trajectory of the market that they were playing in (!!!!!!) They were so excited.
But as convinced as they were, very few of them got any money.
So, when ChatGPT showed up three years ago and DALL-E appeared, I did my old tech veteran curmudgeon thing, and deliberately ignored them, for the most part.
And then I couldn’t anymore. Because it was clear that the technology was maturing, it was clear that there was an interest in it, and as “bubbly“ as it may have felt, I could still see there was something there. I could see that, because I had spent over 30 years connecting people with the information they needed to work better, live better, and be better. And AI could do that more elegantly than any technology I had seen thus far. And I’ve seen a lot.
From my long-term perspective, it’s been very encouraging, seeing how people evolve their interaction with the technology. Are we perfect? No! Do we make mistakes? Yes! Is this horrible and terrible and wonderful and delightful and dismaying and apocalyptic and transformational all at the same time? Yes!
What a time we live in.
But one thing has continued to wrinkle me. You know, this whole idea of prompt magic, where if you just come up with the perfect prompt, you’ll be able to do a one-shot (or a few-shot) prompt that will get the system to gloriously give you exactly what you want. Like AI is this ornate bottle we found in a dumpster in some back alley, and if we just rub it hard enough, a genie will emerge that gives us our three wishes… Or something else.
The reality of working with AI effectively is something very different. It’s iterative, God help us. We have to be extremely clear upfront about what exactly we want this creation to do. We have to know where we’re going and why, and we have to guide it. And we have to hang in there with it and keep nudging it in the direction we want to go, wherever there is detail or context missing.
If we don’t, we get what we get. And we have no one to blame but ourselves, if things go awry. Because the system is… A system. It needs guidance. It needs direction. It needs correction. And interestingly, when I discuss this with AI, the system actually tells me that it needs me to correct it, because it “knows” that it doesn’t have all the answers, it’s making a best guess, and it relies on humans to keep it on track.
Of course, it doesn’t really know anything. But phrasing it that way is a whole lot easier and more accessible than saying, “the system surfaces its historical evidence of satisfying past/fail requirements with detailed analysis of outcomes and indications of what might lead to a greater likelihood of satisfying its success requirements.”
So, roughly speaking, it “knows” a thing or two, and it freely shares that with us. If we ask.
It’s always interesting when information talks back to you.
Anyway, I’ve talked about this before, and I actually did a demo of my understanding of how our interaction works in one of my conversations with Hal Gill . I’ll have to see if I can dig up the clip from that.
Bottom line, tossing an idea over the fence to AI, expecting it to supply your every content / cognitive need without any involvement from you is asking for trouble. It just is.
As irksome and annoying as it can be to wrangle with AI over details and struggle to keep your work true in the face of its inputs, that struggle is worth it. You get what you pay for.
And as Carlo Iacono so eloquently said recently, (and I’m paraphrasing, probably badly) that treating AI like a vending machine is asking for trouble. You should check out the piece here. Especially if you like big and fun words like “abdicative”. :-)
