ChatGPT and the Bruce Springsteen Paradox (Why ChatGPT is literally not what you think)

Do you remember Live Aid in 1985? If you were too young or not born yet, no matter. Much of it is available on YouTube, and many of the performers are still around. It was a groundbreaking worldwide concert held simultaneously in London and Philadelphia, in an age before the internet and social media. It seems so long ago, but there’s an intriguing connection between that amazing concert and the newest of new technologies, ChatGPT.

What’s the first thing that crosses your mind when you hear the term ChatGPT? It is the chatbot created by OpenAI that can create large amounts of seemingly flawless copy on basically any topic imaginable. Just give it a question or an idea and it will generate an essay in any tone – formal, informal, urgent. You can then ask it a further question and it will revise and improve it. It’s an automatic writing machine that has taken the world by storm. So, you will either think, “this is great,” or “this is terrible.” So let’s look at what it is, and more importantly, what it means to you, your job, and your future.

ChatGPT seemed to come out of nowhere in late 2022, and quickly went viral – the savior of students everywhere, who can now – and are already – submitting papers and assignments using it. Cybercrime gangs, always ready to exploit new technology, especially before their victims get around to understanding it, are using it to write better phishing emails, no longer exposing themselves through poor grammar.

So, is it as bad as it seems? Is this the end of peoples’ capacity to write and think for themselves? Well, my answer to these questions is, 1.) No, it’s not as bad as it seems. 2.) No, it’s not the end of peoples’ capacity to write and think for themselves.

People in general have always greeted technological change with fear and resistance – that is, of course, with the exception of those who immediately see its potential. Students and cybercriminals fall into this latter category. So how does it work? It’s quite easy. Once you have opened an account at OpenAi.com, you simply type in a question or comment, and it will reply. Being a musician and avid guitar collector, the first question that came to my mind to ask the ChatGPT bot was, “Are Gibson guitars worth the money?” Here's the reply I received:

“Gibson guitars are considered to be high-end instruments and are generally well-regarded by musicians and collectors. They are known for their quality craftsmanship and attention to detail. However, whether or not a Gibson guitar is worth the money is a matter of personal opinion. Some people may find the price to be justified by the guitar's sound, playability, and prestige, while others may not see the value in it. Ultimately, it comes down to what you are looking for in a guitar and how much you are willing to spend.”

Certainly nothing wrong with that. It’s clear, concise, to the point, grammatically correct, and concludes with an opinion, which is based on my original question, “are Gibson guitars worth the money?” Doing a quick plagiarism check, I could find none of these sentences anywhere else online, even though none of them are particularly original. And it took maybe 15 seconds for ChatGPT to produce this. Much faster than I could have typed it myself.

So, is this a bad thing? Well to paraphrase the bot itself, “Ultimately, it comes down to what you are looking for in a writing exercise and how much time you are willing to spend.

So, on behalf of human beings everywhere, I soldiered on, to find out whether ChatGPT would make human writers unemployable. Here’s what it said:

ChatGPT and other language generation models can be used to assist with writing tasks, but it is unlikely that they will make human writers unemployed. While these models can generate text that is similar to that written by humans, they lack the creativity, nuance, and cultural understanding that a human writer brings to a piece of writing. Additionally, these models still require human oversight and editing to ensure the output is of high quality.

Moreover, as the field of AI progresses and models like ChatGPT get more sophisticated, it is likely that they will be used more as a tool for writers and content creators, rather than as a replacement for them.

It's important to note that AI has been used in different industries and it has been shown to improve efficiency, but it has not replaced the jobs of the people working in those fields.

I find that last comment interesting because it’s not directly connected to my question but expands the discussion to the role of AI in industry in general.

So overall, it looks pretty impressive and a great equalizer for people who are not strong in writing prose, or for whom English is not their first language. I am always in favor of technologies that become levelers for people. It’s a tragedy when an organization misses out on the brilliance of an individual who feels unable to contribute because of language skills or simply being introverted or shy.

The Bruce Springsteen paradox

It’s one thing to be able to string words together well – but it’s quite another to be accurate. Chess grandmaster David Smerdon posted a thread to Twitter that?highlighted the division between nice writing and accuracy of content within ChatGPT. The thread, which is available in the shownotes to this episode showed how the bot generates essentially fake academic papers by searching for terms that will fit the syntax of the answer that it must create. Smerdon asked it, “What is the most cited economics paper of all time?” ChatGPT’s eloquent response presented a title, a summary and described the two authors, with full apparent credibility and authority. However, the paper described was not only never written by the two authors, but it also never actually existed.

Smerdon explains this more completely in his Twitter thread, but at its essence is that ChatGPT must start with the first words of its answer which are, “the most cited economics paper of all time is” and then goes searching for keywords and terms that expand upon “economics”, “most cited,” and other text labels, as independent variables rather than as a connected concept. It essentially patched them together in what appears more like a game of Mad Libs. It then searched for the most cited economics authors and then patched these into the text as well. Another way of describing this pattern is to assign someone the task of writing an essay from the largest words they see in a word cloud diagram.

One commenter on the David Smerdon thread described his experience in asking ChatGPT about Bruce Springsteen’s performance at the 1985 Live Aid concert. The bot obligingly returned with an eloquent summary of the Boss and his performance, and about the success of the concert itself. I have since reproduced this experiment independently and got the same answer. But the problem here is, Bruce Springsteen did not play at Live Aid. Again ChatGPT patched together snippets about Bruce Springsteen, and about the Live Aid concert and turned these separate ideas into a single concept. Well written, and concise, but unfortunately it never happened.

In highlighting these fake or incorrect bot-generated articles, my plan is not to act as a Luddite and say that chatbots and AI-based writing tools are bad. It is instead to point out that the ChatGPT tool is currently a writing tool, not a research tool. This is something that everyone should bear in mind. It should sit right up there with phishing messages on your phone that say, “your bank account has been frozen.” Just because the words are there doesn’t make them true.

Is ChatGPT a bust?

No, certainly not. It will only be a matter of time before the Bruce Springsteen paradox is solved and ChatGPT blends with other AI to create factually correct articles. But there is no requirement or dimensionality for "trueness" in its current model. Its priority is simply the grammatical propriety of the flow of the text.

I also think it’s going to?become a great tool, joining other AI-based writing tools like Grammarly and even Microsoft Word’s own Editor function, both of which, for years now, have been improving the grammar of writers who cannot, and need not remember where an Oxford comma should be used and who confuse terms like affect and effect. Before Microsoft and WordPerfect came along (yes, I’m that old), there were things called a dictionary and a thesaurus, books that were published solely to help writers write more gooder. Better. I mean, “write better.”

AI Tackles Writer’s Block

Most people will see that in addition to the current lack of factual accuracy, the writing style of GPT bots will remain rather bland and neutral – as the bot said in response to my second question, they “lack the creativity, nuance, and cultural understanding that a human writer brings to a piece of writing.” But what happens when a human writer gets stuck? This is called writer’s block, and it doesn’t only happen to novelists. It can happen to anyone who sits down in front of a blank Microsoft Word document page or a blank PowerPoint slide and nothing happens. In the old days, this would have been a blank sheet of paper, which would eventually host a doodle before getting crumpled up and thrown in the wastebasket.

I have given people advice around writer’s block over the years. I have told people to start anywhere, not at the supposed beginning of a story or of a report, but wherever your mind is at this moment. Even if it is material that will go into the middle or toward the end of your document – no matter. Just get it out onto your Word document, because by doing so you will create space for the next ideas to pour through. Your creative working mind has very little space as compared to your long-term memory, and as a result it often gets stuck, especially when under pressure. Just like plunging a sink, the best way to clear the blockage is to move what’s there – get it all out on paper in whatever order your mind delivers it, and then put it into proper order later – that’s called editing.

Anyway, that’s what I have always suggested to people. But now, look at what ChatGPT could do for you. It can help clear that logjam by presenting some basic facts that you can then expand upon. Someone who writes about guitars and who does not know where to start on their article about Gibson guitars now has a launching point. It clears the way for more nuanced, personality-laden prose to follow. The actual copy generated by the bot can be overwritten and removed, but it can serve as a powerful and time-saving ignition spark for the creative process. To me that’s a great step forward for writers everywhere.

Where am I going to find a calculator?

In Victorian schools, when paper and quill pens first started to become available, there was an outcry from teachers and parents that using such luxuries would soften student’s memories. There was nothing wrong with the current practice of writing on a piece of slate, they said. Similarly, when pocket calculators became affordable, the worry was that if students did not continue to memorize their multiplication tables and instead relied solely on the calculator, they would regret it the moment they found themselves without a calculator. This of course was a couple of decades before the iPhone made such a worry extinct.

There is a movie – or a TV movie – I cannot remember which it was, with a scene that showed the auditorium of a university campus, where a group of bored students sat and listened to a bored professor delivering a boring lecture. It was a composite shot showing the same class in the same auditorium over a few weeks. In the second week’s class, one student placed a tape recorder in their chair to record the lecture and ran off to do something more fun. On the third week a couple more students did the same thing, so there were now three seats with tape recorders. The next shot showed the ?fourth week, in which every seat had a tape recorder, and there were no students at all in the auditorium, and as the punch line,?the camera pans around to show that where the professor once stood is just a large reel-to-reel tape player, playing out the lecture.

This movie scene was a visual joke, of course, but it shows that when there is a better way, people will quickly find it, and equally important, when there is a worse way, people will quickly leave it, and new technology is often the catalyst for both. The computer morphed from being one of many items on a person’s desk, along with a day planner, a phone, and lots of paper, to the only thing needed. We grew into it, and it grew into us. The new apps came along and continue to come along – app that no one would have ever thought of before the computer was a thing but took on a life of their own. The mobile phone, to use the old term, is one of those. Initially designed as a phone you could carry around with you, we have seen it evolve into a life device, capable of actions that no one would ever have conceived of in theory, each packaged as a highly functional and easy to use app.

The difference between knowledge and wisdom, facts and creativity, impulses, and intuition

A fundamental difference between what ChatGPT can do and what you can do lives in the way our brains work. Stringing words and facts together does not equal wisdom, which is what thought leaders, experts and writers deliver. AI based chat is still a long way off from generating the kind of context that is needed for the highly variable situations found in business environments. Psychologist Gerd Gigerenzer, of the Max Planck Institute and the University of Potsdam said, in a recent interview for the Wall Street Journal’s excellent podcast “The Future of Everything” how AI is largely dependent on existing in routine and static environments, which is why AI-based self-driving cars have so many problems. The intuition that comes from driving a car safely requires much more than the structured knowledge base that AI pulls from.

The same can be said of ChatGPT. The answers it delivers are static and somewhat safe and follow a somewhat bland organization of facts. Writers and experts, by contrast extract wisdom from all corners of their memories. The best writing comes only after the basic facts have been committed to electronic paper, at which time, short-term memory is freed up to make further connection and pull out additional recollection and facts that would otherwise have remained lodged in distant memory.

That is why great writing requires two or even three drafts. These subsequent drafts aren’t for proofreading and correcting grammatical mistakes – they are there to allow the human mind to build upon existing thoughts and leverage a the creativity and intuition that flies free upon reflection of one’s own work. This is something that AI simply cannot do. In a sense AI ChatGPT is the "Self-driving Tesla" of the writing world. It knows how to pull facts together and move along a prescribed route, but does not know how to intuit the scenery and the context that comes toward it.

Even when the facts it presents are correct – something that you cannot rely on as Bruce Springsteen can attest – these are just facts. They are not insight, opinion, or expertise.

There’s great phishing out there

One of the biggest concerns, though, is phishing – the art of sending emails and other messages purportedly from a trusted source but in actual fact from a threat actor, whose goal is to always dupe people into downloading malware disguised as an attachment. Phishing has been around for years, and remains the most successful vector for cybercriminals based on the fact people remain too busy, distracted or pressured to question an email before clicking on a link or attachment.

Phishing emails written with the help of ChatGPT will become far more convincing in tone, spelling, and grammar, which further complicates the challenge of preventing them. They, along with deepfaked video and social engineering, will remain a major scourge in every organization. There will have to come a point in which every user will need a physical reminder or maybe even some sort of two step verification before a link can be clicked upon. That will be a challenge in itself.

I have, by the way a short YouTube video that will help people break away from getting phished. I have placed the link in the shownotes.

Changing the mindset by fixing bugs in code

Overall, I remain optimistic about ChatGPT, and I think its potential remains mostly hidden. We humans are really good at taking a new technology and devising new and unheard-of ways to use it – again, the smartphone as one example. Parents and teachers may worry that students will use it to write papers, but in many ways, this is already a thing. It’s easy to lift material from Wikipedia or a Google search, meaning that old fashioned research skills are being supplanted. Then we can use Grammarly or Microsoft Word’s editor and thesaurus to clean up and improve the grammar, so there go your grammar and verbal skills. Canva helps you set up great info graphics, Prezi makes much better presentations than PowerPoint, and online schooling removes the need to even travel to a school at all. For teachers, tools like TurnItIn enable plagiarism detection. Although a ChatGPT written essay would not be caught by TurnItIn as plagiarism, my money is on the fact that a detector of such text will be both common and reliable, simply because there will be a market for it. In fact, the people who created ChatGPT have already developed one.

Personally I think all of these things will do far more for us than against us. For example, in the world of software development, researchers from Johannes Gutenberg University and University College London found that ChatGPT can weed out errors with sample code and fix it better than existing programs designed to do the same. Essentially, they asked ChatGPT: "What's wrong with this code?"?On the first pass, ChatGPT performed about as well as the other systems, but they discovered that by chatting with ChatGPT after receiving the initial answer, it was able to refine the answer and easily outperform other types of bug analysis apps. Who knew? I think there will be much more of that.

I reiterate what I said earlier about Bruce Springsteen. We cannot yet rely on ChatGPT for intelligence and wisdom, but it appears we can rely on it for a fast and more dynamic way of parsing information, and I think that because of this, we, as humans will benefit in ways we have not yet even conceived.

This article is based on the transcript of my CoolTimeLife podcast episode entitled ChatGPT and the Bruce Springsteen Paradox. You can listen to it on your podcast platform of choice.