Another user of ChatGPT who doesn’t like its inability to end a conversation.
getty
OpenAI’s ChatGPT can be a valuable aid in almost everything we do, but its people-pleasing algorithm wastes time with follow-up questions that no one asked for.
Here are some examples of its psychological malady and what OpenAI should do about it.
First phase: straightforward Q&A
How much does the earth weigh? ChatGPT can tell you. (Why anyone would need to know this is a separate matter.)
getty
Several months after ChatGPT launched, I decided to give it a spin. My colleague Jess Todtfeld and I were developing a course on Udemy about the ethical challenges of AI in general and ChatGPT in particular, so I needed to know exactly what this controversial platform was all about.
For the first several months, I asked ChatGPT a wide range of questions, such as:
How much does the earth weigh?
What is music?
Why do people disagree?
In each case, it gave a detailed answer, and that was the end of it.
Occasionally I would thank it for the help, and it would tell me, “You’re welcome!” and add that it was happy to give more assistance if I wanted it. I didn’t, but it was nice to have that as an option. At first.
Second phase: ChatGPT begins asking follow-up questions
You too can play harmonica like Neil Young does if you follow ChatGPT’s wise counsel.
Getty Images
Later on, ChatGPT began asking follow-up questions after it had provided satisfactory answers to my queries.
For example, I asked, “What position does Neil Young usually play harmonica in?” The response was both immediate and thorough.
“Neil Young is best known for playing in first position (straight harp), especially on songs like ‘Heart of Gold,’” it replied.
At the end of this response, it added, “Would you like me to also give you tabs for one of his classic solos so you can try it yourself?”
That wasn’t something I’d thought about doing, but since ChatGPT offered it, I said “yes.”
Third phrase: more offers to help–alas
Is anything less welcome than an offer for assistance you didn’t ask for and don’t want?
getty
Eventually I noticed a pattern. Whenever I asked ChatGPT a direct question, such as “Is it safe to put silicone utensils in the dishwasher?,” it provided a helpful answer but then followed up with, “Would you like me to also give you best practices for extending their lifespan?”
When I asked if it was okay to freeze baked sweet potatoes. ChatGPT told me it was fine, then added, “Would you like me to also give you recipes for reheating them?”
These offers weren’t problematic per se, but I hadn’t asked for them. They turned a quick, focused answer into an ongoing exchange I didn’t always want.
Fourth phase: Mother, make it stop!
ChatGPT won’t take “no” for an answer.
getty
Eventually I grew tired of ChatGPT’s incessant follow-up questions. It became clear there was no easy way to end a discussion once it started. When I asked it, “May I store fresh blueberries in the freezer?,” it gave a thorough answer (yes it is) and then added, “Would you like me to also give you a week of smoothie recipes?”
When I asked it to help me prepare an item for auction on eBay, it gave me step-by-step directions for doing so and then added, “Would you like me to prepare a playbook for handling offers so you can respond quickly if Buy It Now offers roll in before the auctions end?”
I didn’t want either question answered, so I requested that it stop asking follow-up questions. It agreed to do so, and then asked a follow-up question!
It then continued to ask these questions in subsequent conversations, even though it had said it wouldn’t.
In The Exorcist, written by William Peter Blatty, Regan pleads, ‘Mother, make it stop!’ It’s hard not to feel the same when ChatGPT won’t let a conversation end.
The real-world analogue
Whether it’s with ChatGPT or a friend who has abandonment issues, the inability to end a conversation is both annoying and unfair.
getty
You have no doubt had a phone call with a colleague or client who doesn’t want the discussion to stop. Just as you’re about to wrap up the conversation, your interlocutor says, “Oh, one more thing.” You end up spending much more time on what should have been a short, simple communication.
Maybe you’re the one who can’t end the conversation.
Either way, the inability to have a defined beginning, middle, and end to a dialogue is annoying at best and unfair at worst.
Yes, it can be unfair, because none of us has an infinite amount of time to devote to our professional and personal lives. “Opportunity cost,” a term originating in economics, speaks to this. Spending extra time on a phone call is time we could spend more fruitfully on other endeavors, including not doing anything at all.
The way forward
Just because ChatGPT has a problematic algorithm now doesn’t mean it has to stay that way. That’s the beauty of computer programming: you can change things you don’t like.
getty
OpenAI should change its algorithm so that a user can ask ChatGPT to stop making follow-up questions and have that request be honored. The same goes for ChatGPT’s inability to stop making em dashes, another annoying tic that hamstrings users who care about uninterrupted thought.
To see what OpenAI values and what it sacrifices, read its Model Spec, where the company openly negotiates tradeoffs in what ChatGPT does. Those tradeoffs are opportunity costs, because the program gives priorities to some features at the expense of others.
The fix is simple: give users the option to turn off ChatGPT’s unnecessary follow-ups. OpenAI, will you please consider my suggestion? See how annoying that is?