Automate or articulate? Certainly, AI can handle the routine, the repetitive, the mundane. But turning to generative AI for first drafts of creative or learning oriented work defeats the purpose.
Generative AI is an apt tool to help speed up the process of writing that needs generalisation, simplification, or summarisation, freeing us from monotonous paperwork. However, there is still a place for painstakingly building ideas on a blank piece of paper or screen.
The core process and immense learning value of writing is lost when a first draft is automatically generated instead of constructed from scratch. In delegating a first draft to AI, we diminish the opportunity to meaningfully learn from the act and to create writing that utterly captivates the reader not because it was generated by artificial intelligence, but because of its exceptional and authentic content.
The mechanics of ChatGPT
In a recent New Yorker piece, Ted Chiang uses the analogies of a photocopier and a blurry JPEG file to explain the abstraction of knowledge that underpins how ChatGPT functions. Like a photocopy of a photocopy or zooming in on a blurry JPEG. The reconstruction of information based on probability rather than retrieval, which explains why it sometimes gets things wrong.
By the end of the decade much of the content online might be AI generated. Perhaps one day AI will be able to generate content tailored to individual tastes and preferences, and that its even preferred by people. Nonetheless bringing anything from ideas into reality is as much about the journey of thought as the final product.
Efficient and generic
Experimenting with the popular chatbots, I’ve consistently found the writing generated is acceptable, but the overall result is generic, and I’d prefer to write the content myself, even when I simply need summarisation. The Chatbots can’t identify the pieces of information I’ve found most interesting or useful from my own reading. This process, although taking longer, ingrains the insights in my mind.
When a piece if content is summarised for us, the main general idea can be construed, but not information that sparks our curiosity from our own experiences, or the elements that connects with unique connected knowledge. These can’t be pre-empted in a prompt (unless the prompt includes all the things we have ever learned and experienced).
We all perceive information slightly differently, and utilise the knowledge constructed in different ways. This is lost if AI systems summarise everything for us with generic understandings of texts.
The cognitive journey
When we write, the writing process is a thinking process. It enables us to develop the skills to be able to articulate arguments which can thereafter be voiced, to develop creativity and to innovate. Even our first draft, or ‘bad’ writing is an important part of the process, as Ted Chiang argues:
"If you’re a writer, you will write a lot of unoriginal work before you write something original. And the time and effort expended on that unoriginal work isn’t wasted; on the contrary, I would suggest that it is precisely what enables you to eventually create something original."
Although Chiang is referring to the craft of creative writing, I would argue this applies to leaning too heavy on generative AI for any intellectual or creative pursuit. At least in the development phases of our building our capabilities, or in aspiring for mastery, it’s crucial we do the work not for the outcome, but for the process of doing the work.
This need for doing hard things is not just about rewards-based performance. There may come a time when much of the thinking-based tasks, we do now for paid work can be delegated, automated or is no longer done. But expanding our capabilities through learning is central to wellbeing, sense of purpose, and to be able to participate in and contribute to communities and societies.
Treading carefully
For the time being, until we better understand how AI models work and have clear delineations regarding optimal and ethical use of generative AI work, I’m skeptical that the benefits of using AI for the first draft of creative work outweighs what is lost and what is risked. This is not to mention debates on whether the quality of Large Language Model outputs may be declining. There is also the problem of model collapse when an AI model is trained on data produced by earlier models, which is a real risk in light of AI generated content already proliferating.
That said, there is a lot of paperwork, documentation, and routine output that generative AI can be utilised for. As someone that is no stranger to being stuck doing repetitive mind-numbing tasks ripe for automation, there are plenty of tasks in people’s workdays that can and should be in the domain of AI systems so that people can pursue more complex, creative or fulfilling tasks.
But let’s not rely on it every time there is a blank screen; that would be a disservice to the potential of our work, as well as our own development.
...
ChatGPT was utilised during the drafting of this piece to help choose wording for the header summary and subheadings.
Future ready?
Comments