Generative Doesn’t Equal Intuitive
"ChatGPT has sh-t for brains."
So went the spiciest part of a text exchange with a friend tasked with operationalizing AI for his non-tech employer. It is going, well, not so well. Hallucinations, self-plagiarism, generally slow-witted behavior.
He is not alone. Right now, there are thousands upon thousands of information economy workers fighting a similar fight. This is the vibe shift from I-can't-believe-what-AI-can-do to Why-can't-AI-just-do-the-thing-I-want.
It's driven by a typical user experience of gen AI that goes like this:
You navigate to ChatGPT or Midjourney or some other site where you are there to do AI. You're writing prompts. Rewriting prompts. Reading prompt advice newsletters. Walking away. Research a metaprompt. Or maybe you just give up and go to Google.
You're spending significant amounts of time on a website that does AI, trying to understand how to make it do AI. And it's exhausting.
This is AI-as-destination, when what we really need is distributed AI.
We're seeing early signs of this in how Adobe is putting AI to generative work, but doing so as part of a fully functioning program. And Apple seems to be taking a similar tack.
Here's how Wired’s Khari Johnson insightfully summed up the use of AI in the new iPhone 15: "The company appears focused on AI that is intuitive, not generative, making artificial intelligence a part of your life that smoothes over glitches or offers helpful predictions without being intrusive."
The notion that we can't have both intuitive and generative in one package is telling. AI is the one slice of the tech world where we're willing to put up with an experience that seems almost designed to be obtuse. Sure, ChatGPT and the like promise to heed your natural language queries, but really these things speak their own languages. If they didn't, prompt engineering wouldn't be a thing.
What we need is an AI that is intuitive but also ambient. An AI that flows into what we are doing yet still helps us to create.
I'm seeing a bit of this in Grammarly. Typo-prone as I am, I turned on the Grammarly Chrome extension and now its cursor thingy hangs out with me as I write, marking up spelling and grammatical errors and suggesting fixes. There's also an AI layer that tries to help with tone and voice. It's not groundbreaking or fully there but it is interesting as a way of thinking about the role of AI in our work.
The Grammarly experience floats around, hanging out with you as you write. It is not the end, but it is helping you get to your own end. You can imagine training on your voice, helping you expand your thinking, perhaps being a research assistant. But as useful as Grammarly could become, at no point in this scenario do I envision going to a digital property owned by Grammarly. The work is happening in my usual workflow and AI is sometimes helping it to flow better.
There are probably a few dozen metaphors we could apply to the role AI is playing in this model. Helper. Assistant. Whisperer. Partner. Maybe navigator. It's helping you get somewhere. As we design AI tools and experiences for the future, it feels important that we're not setting people up to do more AI. Which is maybe a strong argument to not join the many organizations out there building their own LLMs and designing their own standalone tools.
Distribution, not destination.