Inhuman quality
This recent viral post about AI’s capabilities prompted a few thoughts about repeatability, craft and quality. Like many people, I think one of the most destabilizing aspects of AI will be its apparent ability to do high-quality work repeatably and at scale. This ability already extends to things like writing, design, research and analysis—things that previously required craft—and AI’s range of abilities will almost certainly grow. But that’s only the effect on, so to speak, the supply side of quality. I also think AI will destabilize the demand side, and it’s worth thinking about how that could happen. I’m looking at this mostly from the perspective of a writer, but I suspect the effects will be similar in other areas of knowledge work.
Repeatability
The first part of this is pretty straightforward, and it’s an extension of the sort of productivity gains that digital technologies have been providing for decades. AI will definitely make knowledge work faster, and will probably increase the quality of that work for business uses.
Again, I'm looking at this from the perspective of a former IBM writer. Think about what it means to “sound like” a particular brand. IBM is a big company with a long history and a well-established corporate identity, so one of the priorities for the marketing and communications teams was to have a consistent tone and style. This varied a bit for things like ads and slogans, but for instance if we’re producing solution briefs about different products and services, we want those to look and sound alike. This was difficult simply because so many people, from so many different parts of the business, could have input on how something was written and designed. So we writers had training, and refresher courses, and a whole internal style guide and word usage database that were supposed to answer whatever questions we had about how to write something. I’m sure this sounds familiar to anyone who has worked in a large organization. And also, during my time there, we went through several iterations of technical systems for making assets more consistent. These included prefilled templates that the design team could lay out without needing to make judgment calls, and website modules that dictated what you could write and how it could fit together. The most recent thing, deployed before I left, was an in-house generative AI tool that was trained on IBM assets and style—prompt it with what you want, and it would produce a first draft of the asset, in an appropriate format. So this is the ability of AI to drive supply, by making production faster, more uniform and more aligned to corporate standards. This reduces the impact of “craft” in the process—and that’s the point.
Craft
Using skilled people to achieve a uniform result has always been a sort of uneasy truce; it’s just that craft used to be the only way to achieve quality. In many ways, the purpose of the training and organization was to reduce the scope for individual creativity. This makes sense; you can’t have every piece of business writing be its own unique thing with different style, format and design. Being a good professional writer meant figuring out ways to communicate within those established limits. And of course that’s the inevitable trade-off between creativity and scalability, or between creativity and consistency. AI can homogenize that craft, taking all those preexisting examples that human writers have produced and spitting out a variation that conforms to the requirements. It’s tailor-made to produce a sort of quality that’s relevant to business use. And it’s getting better and cheaper constantly. I think the effect on the “supply” side of quality knowledge work is obvious.
Quality
But the second potential effect is different and, I think, unique to AI. The economic forces that prioritize consistent and repeatable creative work also drive a dissociation between two different understandings of what “quality” entails.
Quality from the creator’s perspective means writing something effective and original—a new take on the material or a new way of presenting it.
Quality from the business’s perspective is more formulaic: effectiveness of purpose divided by cost of production. There’s a little wiggle room because that purpose might be directly measurable (like sales of a product), indirectly measurable (like changing the company’s net promoter score), or only tangentially measurable (like improving company reputation). The cost is the salaries for that whole apparatus of production and homogenization, as well as the opportunity cost of doing things at human speed.
You’ll notice that this business quality metric isn’t at all the same as the writer’s fulfillment or the reader’s enjoyment.
That’s why I worry that the effect of AI on the demand side may well be more insidious. Because I strongly suspect that AI is so scalable that the prevailing notion of “high quality” will become more and more synonymous with “the smooth and effective product that AI produces.” Individual variation, inspiration, fallibility could be seen as inherently undesirable because the metric for quality business writing will become “to write like an AI.” And once that becomes the metric, why would people need to be involved at all? Their only contributions would be to make the writing sound less like the machine’s…and what if no one wants that? I’m sorry to say that in the business realm, that eventual outcome seems very likely to me, and the effects will probably be similar in areas such as political speech and journalism.
I retain hope that the creative realm will be different, if only because the point there is at least partly to connect with a creator, not solely to have a particular outcome. But I am nervous that a reading public saturated with “effective” AI writing may lose interest in retaining humanity behind the words. Maybe there will be a backlash and a new flowering of appreciation for individual craft. I guess we’ll see, because as Matt Shumer persuasively argues, AI is happening.