Countering AGI with superintelligent docs?
Reason being, AI will eventually replace most of what we do, making it such that when AGI is reached, job displacement for tech writers will be more common because AGI will perform the same tasks, only cheaper. But the likelihood of AGI progressing to Superintelligence seems less likely to me (in the same way that moving from assisted driving to fully autonomous driving is so much harder than anyone anticipated). Striving for superintelligent docs seems like the most logical counter-move against AI's encroachment on tech writer territory.
- Barista analogy
- What do superintelligent docs look like?
- Wicked problems in the tech comm space
- AI tools will help us tackle wicked problems
- Funding efforts to tackle wicked problems
- Conclusion
Barista analogy
To use an analogy, suppose you’re a barista making espresso coffee. An AGI-capable robot trained as a barista is able to make all the coffee that a regular barista can make but twice as fast. Further, the Android barista can create exquisite espresso art in any shape that humans request, wowing them and making the experience novel. Soon the human barista is replaced. After all, the paying customer would rather pay $2.50 for a robot to make a latte instead of $5.00, especially when it tastes the same.
If the human barista wants to stay employed, he or she has to go above and beyond what the robot barista can do. The human barista needs to think more expansively to uplevel their service and delivery to exeed the robot’s. The human barista needs to break through another barrier toward superintelligent barista’ing. What might this entail?
This barista might decide to become a socially aware, hyper-pleasant presence — remembering the names, orders, and past conversations of each person who comes into the coffee shop. Going into the cafe to get a latte becomes more than just fulfilling a caffeine addiction; it becomes a social experience, almost like walking into the Cheers bar where everyone knows you and you feel like you belong and have a place. The barista makes you feel warm inside, recognized and appreciated, listened to. The barista’s face lights up when they see you, and they know your name.
The human barista can’t possibly remember the details of so many hundreds of people coming into the cafe. In fact, Starbucks baristas are frequently lampooned for getting the names wrong on cups. To overcome these limitations, the human barista leverages AI to take notes and recall the needed information. They might use other tools and techniques. However, the human barista does it, they have transcended the robot barista’s level.
Now technical writers might need to do the same when it comes to documentation. Our docs have to be better than what can be performed by AI operating at an AGI level.
What do superintelligent docs look like?
How do you define superintelligent docs? While various definitions could fit this bill, such as automated documentation that predicts topics based on user behavior or errors, or documentation that dynamically writes itself around the user’s specific situation and needs, those imagined experiences are a bit too futuristic for me.
More realistically, superintelligent docs solve wicked problems. Wicked problems are so massive, gnarly, multifaceted, and hard to tackle — for example, they include variables that are constantly changing based on recurring feedback loops, etc. &mash; that they are beyong merely complicated problems; they are complex problems. Solving them seems beyond human capabilities.
Wicked problems in the tech comm space
Do wicked problems exist in the tech comm space? Personally, I’ve been contemplating throwing myself headlong into a potentially wicked problem in my domain: documenting the data model of the world that becomes a map. Basically, there’s a bunch of map data in a giant database, details added by hundreds of map operators — this becomes the source from which geo-related APIs are hewn to surface the data. Without going into too many specifics, this could be a wicked problem due to the following:
- The model exists across multiple groups in different organizations. The groups have different terminology and interpretations of the data, with unique needs. The groups who input the database information differ from the engineers who build the APIs.
- The database information doesn’t necessarily correspond with the API outputs. In other words, the APIs might transform and manipulate the underlying data for specific calculations and needs.
- The model is described in various documents and sources owned by different teams, from engineering source files to knowledge bases to Google Docs and more. The source documentation isn’t consistent, organized, or structured in any readable way.
- The scope of data elements is overwhelming. When diagrammed into a tree, there are many hundreds of elements, with conditional logic and other nuances that make it less straightforward.
AI tools will help us tackle wicked problems
We can’t tackle wicked problems without leveraging AI tools. Commonplace tasks (such as preparing release notes based on what’s changed in an API) can conceivably be replaced by automated AI tools that can decompose the tasks into a few discrete steps that an agent can chain together into a process that yields decent documentation. However, wicked problems don’t easily decompose into a series of small steps that can be chained together into a process yielding an outcome. For example, the scope of content included in a wicked problem probably exceeds any AI model’s token limit to process. Wicked problems require a human mind to bring a strategy and plan to the problem.
Addressing wicked problems will invariably require tech writers to leverage AI tools to perform analyses, extractions, and manipulations of source data. But these wicked problems aren’t something that a machine would likely be able to execute on without AGI transforming into Superintelligence.
Funding efforts to tackle wicked problems
I want to address one more variable with wicked problems. A core challenge in tackling wicked problems is identifying funding. When a problem cuts deep across multiple organizations, the corporate budget model tends to fall flat. Engineering budgets pay for tech writers to document the APIs that its engineering teams create. Spread your wings too far and try to climb too high in your ambitions of tackling a wicked problem and you’ll soon find that you’re no longer in the funding group’s atmosphere and orbit. And they’re not paying for it.
Further, while system thinking (with books Thinking in Systems by Donella Meadows, or The Fifth Discipline by Peter Senge) tends to speak to complex and wicked problems, I find that system thinking in documentation projects usually goes beyond any particular techcomm project or budget concerns.
At my work, our tech writer group once sought to publish a series of lifecycle workflow documentation that would describe the life of something across many different systems, states, and teams. It was fascinating to see how data might be transformed from beginning to end. As an analogy, it was like putting a tracker on monarch butterflies and watching their migration pattern across multiple generations and continents. However, there was little executive support for such a project. Only high-level product managers and executives needed this birds-eye view of the technology, and they didn’t want to acquire it by reading long-form documentation. They wanted someone to talk through diagrams.
This is the conundrum of focusing on wicked problems: no one specifically owns them. If no one owns them, no one funds them. This is part of why the problems grow, extending their roots deeper into the ground. Thus, although I’ve argued that focusing on wicked problems might be the only shield against AGI-capable AI tools replacing tech writers, it’s still questionable as to whether wicked problems will provide the sustenance tech writers need to provide organizational value.
Conclusion
What do you think? Do tech writers need to tackle wicked problems and deliver superintelligent docs to stave off job displacement from AI tools? Or is this scenario too far into the future to be real? Whether the AI trajectories are bogus or not, setting our sights on wicked/complex problems seems like a noteworthy goal. As a profession, we could stand to be a bit more ambitious. With AI at our disposal, we might have the tools we need to be successful.
About Tom Johnson
I'm an API technical writer based in the Seattle area. On this blog, I write about topics related to technical writing and communication — such as software documentation, API documentation, AI, information architecture, content strategy, writing processes, plain language, tech comm careers, and more. Check out my API documentation course if you're looking for more info about documenting APIs. Or see my posts on AI and AI course section for more on the latest in AI and tech comm.
If you're a technical writer and want to keep on top of the latest trends in the tech comm, be sure to subscribe to email updates below. You can also learn more about me or contact me. Finally, note that the opinions I express on my blog are my own points of view, not that of my employer.