AI is a feature, not a product

Infestations of intelligence • January 23 2023

AI tools right now (Stable Diffusion, chatGPT, et cetera) are feature phones: they’re showing up in a bunch of extremely weird form factors with wildly different capabilities and use cases. There’s been no lightning-bolt iPhone moment for AI yet. For now, AI will continue to metastasize into the nooks and crannies of the products we’re already familiar with.

Apple owns the iPhone experience soup to nuts, but tools like Midjourney run inside Discord – an abdication of ownership over distribution that seems bananas in SaaS terms, but is arguably the next shift for product thinking: channel-agnostic functionality. What you’ll be buying in the future is not AI as a standalone product but on-demand access in your favourite app to ‘intelligence’.

Getting GPT3 to run inside of Google Sheets is a great example of this. If you’re already doing your work in Sheets, the switching costs of “an AI Sheets-alike” is too high. But “Sheets with smart automation” is an easy sell.

Closer to home (for creatives) we’re starting to see this kind of platform extension get rolled into new plugins and capabilities within design, video, and audio production tools. The experiential layer of these products isn’t shifting (too much), but the number of one-button ‘magic’ features is growing.

Great, cool, do my job for me, bots! So why is this a problem? Let’s take Magician by Diagram as our example (but by extension any product that works by ‘prompts’, which is a lot of them) – there’s a big hurdle around replicability. AI tools don’t expose their knobs and levers. Their interaction model is magic but that only means that the real controls have been taken away.

AI generation is a black box: you can have an expectation of what sort of output you’ll get, but you’re essentially throwing spaghetti at the wall until the algorithm produces something that you like. Unlike the tools we’re used to, which work via straightforward inputs and produce straightforward outputs, we’re getting value from AI tools mostly by granularly fine-tuning our incantations to appease the AI – this is a capricious genie, not a well-oiled mechanism.

Positing that these tools are, or can be, drop-in replacements for artists, designers, or musicians is absurd; they fundamentally cannot understand the brief. There is no true intelligence there, we’ll never receive an explanation for why certain creative choices were made and not others (as AI is not making choices at all). We are massively overfitting to fast and cheap at the expense of good. To add to this, any business should have concerns about the provenance of AI-created work, and rightfully so. The job of humans in these roles may change in scope, but they can’t (and shouldn’t!) be removed.

The product decisions have already been made: it’s inevitable that our tools will become saturated with AI-style intelligence, for better or worse. We’re charging ahead knowing that AI tools suffer from inherent biases and unsurprising moral failures. I’m not against AI at all, and I think we’ll discover some really good use cases – but we’re still in flip-phone territory: most of our learning is in the future. We should be less worried about our jobs being eliminated, and more concerned about how AI ubiquity will change the perception of the work we create.