The rise of commercial artificial intelligence (AI) chatbots like ChatGPT, Gemini, Grok, Claude, etc. has brought with it a great temptation. The immense data centers that power them give them useful – in most cases – data identification, collection, and analysis capabilities. This tempts people to use them as “all-knowing assistants”, who can provide answers to just about everything, quickly and reliably. Well, not always reliably, but I’ll get to that later.
With generative AI, people have become increasingly tempted to replace actual human skills with computer-generated content – sometimes mimicking specific styles. This has resulted in a flood of what’s now known as “AI slop”, which is plagued by all sorts of aesthetic and even factual discrepancies. This is even worse with AI-generated texts, especially analyses, essays, articles, etc. Although, by now, more experienced readers have learned to identify AI-generated texts and images quite well, many still slip through the cracks.
Even though, as part of my PhD, I kind of did AI “before it was cool,” I never believed it’d be able to replace humans for complex tasks. My doctoral research, after all, treated autonomous agents as assistants: as software that, based on decision maker input and preferences, identifies alternatives that are worth considering. I never intended for it to replace, for instance, a procurement committee.
That said, I do leverage the data collection and processing capabilities of AI systems for finding sources of information on topics I happen to be working on; for translating these sources, if they’re in a language I’m not fluent in; and for data analysis, with emphasis on the identification of conflicts and inconsistencies among sources, I know full well I must never trust any AI system blindly: personal experience has shown they can be prone to “hallucinations”, and even the best systems can are highly susceptible to introducing errors as a “conversation” with them drags on.
Therefore, I use AI as a tool that has certain capabilities and limitations rather than as a replacement for my personal work. So, I am the one who defines the problem at hand, chooses the context, the data set(s), and checks the results for accuracy, correctness, consistency, and so on. These results are, for me, source material which I’ve personally examined, curated, and signed off on, before I proceed to write my articles. I find the “prompt-and-forget” approach that’s spread across so many fields – especially journalism, unfortunately – like a terminal-stage malignant tumor to be lazy, complacent, and often misleading and fraudulent. Besides, I can’t understand how any essayist worth their salt would allow an AI to replace their personal writing voice and tone.