ChatGPT Can’t Take Your Job Unless It Figures Out How to Do One Really Difficult Thing
Is the artificial intelligence revolution about to fizzle?
Are you worried about ChatGPT taking your job? Me too! After all, writing is a huge part of my job, but, in the time it takes for you to reach the end of this sentence, ChatGPT can spit out 2,000 coherent and compelling words about the exact same topic.
Clearly, I’m screwed.
However, a recent experience using ChatGPT has me thinking the situation isn’t as dire as I’d originally assumed.
For starters, using the phrase “artificial intelligence” to describe something like ChatGPT is a misnomer. A better phrase to describe it might be “large language model processor.” However, since “large language model processor” doesn’t make for compelling headlines, everyone seems to have agreed that ChatGPT, and other tools like it, are “artificial intelligence.”
Fine.
Whatever.
We can keep calling this stuff artificial intelligence, but let’s also agree that ChatGPT (and its similar tools) are as stupid as they are intelligent.
To be clear, I’d say the same thing about humans. People aren’t nearly as intelligent as we like to think (myself included). Instead, we’re not particularly different from the AI we’ve built. By that I mean we’re very good at collecting small amounts of data and then making enormous assumptions using the data while tricking ourselves into thinking we “know stuff.” Better yet, if we’re really good at this process, we move into careers where our jobs are to trick other people into thinking we “know stuff.”
For the record, that’s what I do. I’m an “academic.” I teach entrepreneurship at Duke University, which basically means I get paid to explain stuff that I’m supposedly an expert on. Am I really an entrepreneurship expert? Maybe. Maybe not. Certainly, with a bit of thoughtful searching on YouTube or by listening to a few good podcasts, people could learn everything I know. As a result, my unique value proposition isn’t “knows a lot about entrepreneurship.” The thing I’m good at is asserting information about entrepreneurship with clarity and authority. For better and for worse, that ability helps convince people I’m an expert, which, in turn, keeps me employed.
ChatGPT is great at doing the exact same thing. Sure, ChatGPT doesn’t have a fancy PhD like me, but it does have access to enormous amounts of knowledge stored in its datasets, and all that knowledge gives it the ability to assert things with clarity and authority in ways that convince people it’s an expert.
This is where AI gets problematic. Even though ChatGPT is great at asserting things, the things it asserts aren’t necessarily true.
I recently experienced this problem while using ChatGPT to find examples for a book I’m researching. One of the subjects for my book is a woman named Fanny Fern. In the 19th century, Fern was America’s most popular columnist and essayist, and I needed examples of her promoting products in her columns. Rather than spend hours digging through old archives and collections of her writings, I figured I could use ChatGPT to find examples for me.
That’s exactly what ChatGPT did when I asked it for “examples of Fanny Fern promoting products in her columns.” According to ChatGPT:
“In her column for the New York Ledger, Fern wrote about the importance of time management and mentioned the Waterbury Clock Company as a reliable source for clocks and watches.”
“Great!” I thought to myself when I first read ChatGPT’s response. In seconds it had given me an answer that was going to save me days of research. I just needed to know what New York Ledger article ChatGPT was referencing in order to find the original text. So I asked: “Can you point me to the article where [Fanny Fern] discussed the Waterbury Clock Company?”
That’s when things got weird. ChatGPT responded:
“I apologize for any confusion earlier — I made an error in my previous response. I could not find any specific column where Fanny Fern wrote about the Waterbury Clock Company. In fact, upon further research, it seems that the Waterbury Clock Company was not founded until 1857, several years after Fanny Fern’s writing career had ended. I apologize for any confusion my earlier response may have caused.”
Uh-oh. Maybe this artificial intelligence stuff isn’t so intelligent after all. Not only did it not know my answer, it didn’t know basic facts about Fanny Fern because it told me her career ended before 1857, and that’s not true. Fanny Fern was hugely popular and wrote almost until her death in 1872. This means ChatGPT didn’t just get its initial answer wrong; its explanation for why it was wrong was also wrong.
That’s a lot of wrong.
I don’t fault ChatGPT for being wrong. Lord knows I’ve been wrong plenty of times. The problem wasn’t that ChatGPT was wrong; the problem was how confident ChatGPT was in the way it presented its wrong information.
To appreciate why this is a huge problem, imagine if I hadn’t asked ChatGPT to show me the article it was telling me about. I’d have assumed its initial information was correct, and I’d likely spend the rest of my life thinking Fanny Fern had been a proud promoter of the Waterbury Clock Company. I might have even included the example in my book. How embarrassing!
I know what you’re thinking: ChatGPT misleading a random college professor about what product some person who died 150 years ago did or didn’t promote is hardly the end of the world. But consider this: if ChatGPT is wrong about Fanny Fern, don’t we have to wonder what else it’s wrong about? More importantly, what happens when ChatGPT is inevitably wrong about more important things?
Remember that humans are lazy (again, I’m including myself in that blanket description), and, when something seems logical and reasonable, we don’t tend to question it. Instead, we accept whatever the assertion is as correct. Plus, the more confident the assertion is made, the more likely we are to believe it. This means we need to stop worrying about ChatGPT taking our jobs and start worrying about whether it’s going to do something much worse.
And, to be clear, ChatGPT’s tendency to confidently assert false information as truth could be much worse than it taking our jobs. After all, confidently asserting false information as truth isn’t something intelligent people do. It’s something demagogues do to control populations.
In contrast, intelligent people aren’t intelligent because of what they know. They’re intelligent because of their ability to ask good questions about the things they think they know. For proof, look at all the greatest thinkers in history… Aristotle, Galileo, Newton, Einstein, etcetera. They’re all people who questioned common wisdom rather than regurgitate it.
Lucky for us humans, questioning what we know is a lot harder than learning it. As far as I’m concerned, until ChatGPT can figure out what it doesn’t know and how to question itself, we can’t really call it intelligent. It’s just artificial.