The Diplomat
Picture Credits: gettyimages
ChatGPT isn’t the problem, its real value lies in revealing the hidden inefficiencies and outdated practices across education, law, journalism, and bureaucracy. It exposes deeper systemic issues that human institutions have long ignored.
The Problem Isn’t ChatGPT. It’s Us.
Author: David Hutt
Talk to a university professor these days, and it isn’t long before they’re complaining that almost every student is now using ChatGPT to write essays. My response is always the same: maybe the problem is that essays have taken over the way universities judge a student’s aptitude. What about scrapping essays and going back to hand-written or oral exams? Attempting this response recently, a professor friend agreed with me but opposed the solution in principle because “it would mean I have to spend more time with students.”
In the 1930s, the economist Paul Anthony Samuelson coined the term “revealed preference” to explain how consumer behavior could be discovered by simply observing what people bought, not just by asking them what they wanted or by making assumptions. Likewise, as all psychologists know, allowing someone to vent their grievances usually uncovers a “revealed problem,” usually a psychological plight that is causing the physical behavior. Alcoholism or drug addiction, for instance, are often manifestations of depression or social isolation, rather than being the root problem.
On the whole, I’m rather pro-ChatGPT, in part because I’ve come to think of it as the great revealer of problems. As with my professor friend, the problem isn’t that students are using ChatGPT to write essays; it’s that professors don’t want to spend even more time with students that alternative examinations would require. To be fair to most academics, this is because universities now require them to spend almost all their hours researching and publishing, since the number of publications a faculty produces has become the sole metric by which their competence is measured.
Likewise, is it a problem that ChatGPT can write legal copy better than most trainee lawyers? Or is the problem that law firms have, for too long, demanded interns and trainees do endless drudge work because of the legal profession’s money-grabbing obsession with “billable hours,” which has convinced clients that time spent equals value delivered, thus allowing a firm’s partners to earn millions of dollars from slothfulness? One might also ask whether ChatGPT’s ability to write turgid journalistic copy is an existential crisis to my own industry, or whether it reveals the actual problem: so much of journalism has become writing clickbait bumf that requires no actual investigation or original thinking and can easily be done by an LLM that’s far better at searching on Facebook for the latest consumer craze or exaggerated outrage?
I should stress that I’m not someone who heartlessly thinks that anyone whose job is replaced by a new technology is feckless. However, at least in terms of knowledge work (a horrible term, I concede), ChatGPT and other LLMs will separate some of the wheat from the chaff – or, at least, force some professions to consider why they’ve become so bloated with chaff work. Middle management is a prime contender. Yet ChatGPT will never be able to go into a warzone and interview combatants, nor write a wholly original opinion piece. It could make a stab at translating “War and Peace,” but, according to a translator friend, it will never be able to convey what Tolstoy actually meant.
There was the news this month that the Albanian government has created an AI-generated “minister” to tackle corruption. That sounds like a genuinely good idea, since this is one area of governance where Big Data can do what humans cannot, and it’s something that Southeast Asian governments should look into. However, one ought to be suspicious about governments using LLMs in other areas, especially economics, because the last thing we need is bureaucrats generating yet more text in huge quantities that is reliant on either readily known information or much publicized, logical examples.
In a recent column, I argued that governments should stop collecting most economic data for a few years because doing so would compel politicians and bureaucrats to search for information from more unusual sources – maybe even asking ordinary people or small business owners for their opinions. As such, my problem wasn’t with economic data; it was with the way decision-makers increasingly rely only on conventional, short-term, and the quickest to-hand material, rather than attempting less obvious but potentially better and more revealing means of information-gathering that might help solve longer-horizon problems.
The psychologist Paul Bloom coined “the Ginger Rogers theory of information” a few years ago to explain how, in any information system, some ideas will propagate more than others because they appeal to our existing prejudices. Yet information that runs counter to our prejudices will either be rejected or discounted unless it reaches a very high standard of proof. GDP growth is an excellent example. If you’re a Southeast Asian government and you’re achieving 5 percent growth, that’s basically shorthand for “good job,” even if there is more unusual or micro information that suggests otherwise. In a similar vein, there is the danger that our existing prejudices become whatever ChatGPT recommends. Anything that it doesn’t recommend will be instinctively discounted. However, there is also the danger that we develop a prejudice that says the replacement of any job by ChatGPT is the problem, thereby automatically discounting the counterproposal that says this replacement is a manifestation of a much deeper problem.
Credits: TCA, LLC.