The inefficacy of AI writing

 

It is an accepted vision that AI will soon entirely eliminate human writing. We old-timers will brag about how we wrote something by hand, just as some of our grandparents proudly declared that they walked through 5 miles (or kilometers, as applicable) to get to school instead of getting in the toasty car equipped with snow tires as they do now. Progress by definition, irresistibly moves forward, it is claimed. However, I would argue that AI writing may supplement homo sapiens muses but will not eliminate them, but not because there is anything intrinsically wrong with AI writing. The actual reason is that AI writing is simply not effective in many, if not most, situations involving learning, explaining and persuading.

Proper AI output requires directed prompting and editing. The initial output generally contains several stylistic errors in terms of the desired result. The inputting party must refine the prompts until the AI app produces a truly acceptable text. The a priori condition here is that the person framing the prompts is capable of identifying the issues. Therein lies the problem. Without learning how to write using the human mind alone, making mistakes and receiving feedback, a person does not develop a proper template for comparison’s perspective to even judge the AI version. Many students, including adults studying law and finance, avoid the challenge of putting their text on a page and immediately turn to AI. They lose their prime opportunity to learn how to write text and, consequently, to edit AI text. They may pay a high professional price for their shortcut.

The other aspect of AI text output involves factual errors. There is no shortage of reports in the media of AI actually inventing law cases and medical studies as it overstrives to meet the demands of the inputting party. This situation is humorous if the human party catches the error before submission or publication but can have unfortunate consequences if not. While laziness and ignorance sometimes play a part, a more pernicious factor is the illusion that AI is infallible. No AI engine makes such a claim but it is so easy to believe everything a computer tells us. Ideally, everybody would check the facts but, in practice, many writers are lazy or ignorant. Therefore, for many professionals, AI does not in fact reduce the effort involved in writing, but merely changes its form.

The main reason that AI will probably not replace human writers is far more universal. Simply put, for many purposes, its output, even guided by trained writers, does not attain the goal. Long before AI erupted on the scene, the average person did not trust language that was too sophisticated or smooth. They called it “highfalutin” or whatever the local word term was and immediately suspected either someone was hiding something or trying to make them feel stupid. As an example, if a car salesman claimed that a given vehicle was “aerodynamically and ergonomically designed to maximize power and fuel usage”, customers would look confused or suspect that the engine was too weak. Neither result would lead to sales. People prefer to hear direct language using everyday words. Ignoring his politics, President Trump may be crude but everybody understands what he is trying to say at any given moment, unlike most politicians. On LinkedIn, many writers complain about “AI-written content” being published. The issue is not the use of AI but the lack of connection to the reader. AI, even with human help, tends to “automate”, which is much less effective than less polished but honest words.

There are clearly exceptions to this tendency. Academic writing loves soufflé writing, i.e., fancy words with hidden, implied or even no meanings. AI is outstanding at producing these texts, far better than most human beings. Other fields use a specific form and lexicon to express their ideas, often rather formalized. These areas include law and business. Many legal claims and financial press notices sound almost identical aside from the relevant names and numbers. AI serves as an “expert system” in producing this coded text whose audience is in the know of the meanings and expects a certain style. In fact, these readers may not appreciate anything less than it.

Still, AI text is not appropriate for those who are learning, researching and persuading. Even with editing, it quite often triggers an intuitive resistance to the message and the suspicion that something is amiss. It will take a period of time before the impact of these AI inadequacies strikes home to the enthusiastic AI users out there, but the issues will not disappear. In summary, realistically, taking a walk is an enjoyable and healthy experience. However, there is nothing wrong with staying dry and comfortable in a car on an unpleasantly hot or rainy day. AI and human writing are not mutually exclusive.