Does Design Matter in a World of LLMs?

OK, since I am a designer myself, I think we can get the obvious bit out of the way now: yes, I think that design continues to be important. Let’s look at three reasons why.

First: if we survey what people are either doing, or claiming they will do once funded, a number of attempts at building products around LLMs are kind of underwhelming from a design viewpoint. Which, by the way, can make a good “sniff test” if you are doing due diligence as an investor. In several cases I have seen, there is no apparent purpose for an LLM to be involved at all, which reminds me of the period when everything had a “blockchain” in it somewhere for…reasons.?

As a designer, you have a couple of possible ways to change this, if you are in the product path. The less politic one is to ask why all the drama around an LLM, when something simpler and less likely to get weird, like a good rules engine, might be better.?

A more harmonious approach might be to map out some possible workflows and propose ways in which to make an LLM actually help. LLMs are not magical fixits by any means, but they do some tasks very well, like text summarization or extraction. Focus on those types of known-good operations, and you can avoid being part of the next crowd-pleasing trainwreck.

Second: LLMs, like earlier advances, are not going to suddenly bring about the “zero interface.” There are some potential uses for LLMs at the interface layer, such as making menus dynamic and contextual, but there are very few systems that cry out for an entirely stochastic and non-repeating interface. Don’t get me wrong: the “zero interface” concept can lead to new ways to innovate how we approach UX/UI. But the maximal implementation remains a software manager’s version of The Singularity.

Third: nope, LLMs and other forms of present day “AI” are not about to replace designers this year. If someone still has the idea that designers “come in at the last minute and make it look pretty,” they are welcome to try this, but for anyone not trapped in 1995, there are some issues.

We can start with maybe the biggest single problem common to current AIs: they have no actual world-knowledge or contextual basis for what they do. They emulate the forms of things, but don’t grasp the reality behind them in the way we take for granted, and we should not ask them to try. This is why we keep seeing legal briefs that cite non-existent caselaw, or pictures where there are too many limbs for the number of heads depicted.?

So while a specialized AI tool might, for instance, be able to take a set of style guides, workflows, and UI examples, and flesh out a set of mockups or walkthroughs, someone with real design skills still needs to establish those baselines to begin with and ride herd on the process.