In today’s column, I examine the use of generative AI to address a viral question about whether there are more wheels versus doors in the world. The unresolved question itself is not especially earth-shattering. However, there are valuable lessons to be garnered regarding how to make use of generative AI best overall. You see, the simple act of gauging if generative AI can answer the wheels versus doors conundrum is a handy means to reveal keen insights on prompting techniques and getting the most out of generative AI and large language models or LLMs.
Let’s talk about it.
This analysis of an innovative proposition is part of my ongoing Forbes.com column coverage on the latest in AI including identifying and explaining various impactful AI complexities (see the link here).
The Viral Wheels Versus Doors Question
I was recently giving a talk about the latest advances in generative AI and an audience member asked whether contemporary AI could resolve the unsettled question of whether there are more wheels or doors in the world.
That seemingly oddball question about wheels and doors might cause you to reminisce. Back in March 2022, there was a Twitter posting that sparked a heated debate on the topic. Twitter went crazy over the question. TikTok joined the bandwagon. All of a sudden, millions of people were arguing online, and the matter went completely viral.
In case you weren’t privy to the debate, or maybe it’s a distant memory, the issue goes as follows. If you were to count all the wheels in the world and count up all the doors in the world, which of the two would be the greater count?
One obvious difficulty is that trying to count all the wheels and all the doors is not a practical endeavor. Imagine merely trying to count the number of wheels and doors in your home, and then in your neighborhood. It would be a very daunting task. I doubt you could do so to any great degree of accuracy since there are going to be wheels that are hidden from view and potentially doors that are also hidden.
Wait for a second, you might be thinking, how could wheels or doors be hidden? For example, if you look at your car, you can obviously see the wheels and the doors. Done. Counting tasks accomplished.
This seems straightforward.
Challenges Abound In The Answering Quest
Life is never as easy as it might seem.
Suppose that we agree with the idea that wheels and doors are not solely the artifacts you can readily see with the naked eye. Consider that within a car, such as in the engine, there are likely many wheels of one kind or another. We might also contend that there are more doors than just the passenger and driver doors. Some would insist that the glove compartment consists of a door, and so does the trunk lid. Lots of other doors or door-like structures undoubtedly exist throughout a car.
Thus, there is a definitional issue to be dealt with.
What constitutes a wheel?
What constitutes a door?
If we cannot agree on those definitions, we will be talking past each other endlessly. The counting of wheels will come up with wildly different counts. The door counters will likewise arrive at widely different counts. No progress will be made in resolving the question.
Polarization On The Vexing Question
Given the widespread polarization on most societal topics, you are perhaps not surprised to know that the wheels versus doors question slithered people into separate camps. The Team Doors camp advocated that there are more doors than wheels. The Team Wheels camp professed that there are more wheels than doors. On and on it went.
Much of these claims of victory were based on personal conjecture.
Lamentedly, some then ventured into profane name-calling. Things got ugly. Each side tended to call the other side a bevy of ugly epithets. What had begun as a fun and lighthearted consideration dovetailed despairingly into the dire depths of an us-versus-them saga. Sad face.
Some rationalists tried to come up with a calculable method to answer the question. Maybe some formulas could be used. One effort sought to estimate how many wheels per capita there are, and how many doors per capita, which can then be used to multiply by the world population. Voila, you have a numeric answer that is based on stated logic.
Sorry to say those were generally ripped to shreds. The estimates were based on key assumptions that could readily be knocked to the floor. Going the quantitative route ended up providing interesting folly but did not nail down the answer. Just more arguing ensued.
A surprising undercurrent arose. It is almost funny, in a way. Companies decided they could run with the craze by tossing their company name into the debate. A handy opportunity to get free publicity. Companies that make doors were happy to proffer their opinion. Firms that make wheels opted to get into the fray. Even other entities such as coffee makers said that you ought to mull over the question while drinking their brand of coffee. When there are eyeballs to be had, you might as well try to profit from the dispute.
As per all flash-in-the-pan viral trends, the thorny matter eventually faded out and people moved onto other newer viral trends.
Did the wheels versus doors question ever get fully answered?
Nope.
The ghost of the viral trend still exists to this day, thus, the question came up during my talk about AI advances.
Generative AI To The Rescue
As noted, the viral trend occurred in March 2022.
During that time, there wasn’t much going on publicly about generative AI. Various generative AI apps had been quietly released, but they often were scorned due to emitting foul words and otherwise not being especially fluent (see my discussion at the link here). Then, along came ChatGPT, a generative AI app that had been data-trained to try and avoid the prior problems, and suddenly it became a big hit.
That was in November 2022, some eight months or so after the freewheeling wheels versus doors challenge first appeared. My point is that generative AI wasn’t a big deal when the wheels versus doors question went viral. Generative AI wasn’t a go-to option in the minds of the public at large. It was more so an AI insider or techie preoccupation.
Nowadays, ChatGPT alone has over 200 million weekly active global users. Other major generative AI apps such as Anthropic Claude, Google Gemini, and Meta Llama, also have many millions of global users. Generative AI is immensely popular. Period, end of story.
It makes sense that we would want to see how generative AI approaches the vexing question of wheels versus doors. People keep trying to test the limits of generative AI to discern what it can and cannot do. At the same time, AI developers and AI makers are stridently advancing generative AI.
Seems like asking modern-era generative AI about wheels versus doors is a great idea. We can examine how generative AI handles the issue. Furthermore, this will reveal the best ways to prompt generative AI. I mention this because the way that you ask questions or interact with generative AI is a huge determiner of what you will gain from conversing with the AI (for my comprehensive coverage of over fifty prompting strategies, see the link here).
In a sense, your best bet of deriving value from generative AI is on your shoulders. If you don’t do a good job inquiring about whatever you want to know, the AI is going to seem to you as not being particularly helpful.
Let’s see what that means and uncover valuable lessons about prompting.
Ask The Question As Straightforward As Possible
I opted to log into ChatGPT for this mini-experiment. The other major generative AI apps would likely yield similar results. Whatever we glean with ChatGPT is probably the same or likely similar outcomes of using any other major generative AI.
First, you are wise to try and straightforwardly word your prompts. Be concise. Avoid including any extraneous verbiage. You want to focus the AI on precisely whatever you intend to ask.
Here we go.
- My entered prompt: “Are there more wheels or doors in the world?”
- ChatGPT generated response:
“The question of whether there are more wheels or doors in the world is a fun, thought-provoking debate without a definitive answer. To approach this, let’s break it down into smaller questions.”
“How many wheels are there?”
“Cars, trucks, and motorcycles typically have 4+ wheels. Cars also have steering wheels and spare tires. Bicycles, scooters, wheelchairs, strollers, and luggage also have wheels. Many machines, factory equipment, toys (e.g., LEGO sets, Hot Wheels), and office chairs have wheels.”
“How many doors are there?”
“Homes, buildings, and vehicles (cars, buses, planes, etc.) have multiple doors. Cupboards, cabinets, and closets all contribute to the door count. Elevators, refrigerators, and even computer servers often have doors.”
“Given the wide range of objects that use wheels and doors, there are strong cases for both sides. Factories and transportation add millions of wheels, while homes and large buildings could tip the balance toward doors. Estimating numbers would be a challenge but based on the types of objects in the world, it’s likely that both numbers are incredibly large and perhaps even comparable.”
End of ChatGPT response
What do you think of the response?
You can plainly see that the AI outlined the nature of the problem. The answer we are seeking regarding which is more, well, we are left somewhat hanging in the air. ChatGPT says that strong cases for both sides can be made. Seems wishy-washy.
We definitely didn’t get a definitive answer.
The answer was mushy, which is a bit disappointing, though we did at least get an explanation of why the matter is complex and not readily answerable. You could give points to generative AI for bringing us up to speed.
The fact that ChatGPT did not say that the answer is wheels, nor that the answer is doors, can be declared as a good sign. The reality is that no one does know the answer, not even AI. If the AI had indicated that the answer is wheels or the answer is doors, this would be essentially a lie. Or, if you don’t like calling such an answer a lie, it would be an error or mistaken answer.
Bottom-line is that you never know what answer generative AI might give. It is like a box of chocolates. The response by AI must always be given strict scrutiny. Answers could be based on faulty data training. Answers could be so-called AI hallucinations (see my explanation at the link here). Do not assume that AI is somehow all-knowing. It is decidedly not.
Pushing Generative AI When Needed
Now that we’ve gotten a response indicating that the AI doesn’t know the answer and that strong cases can be made for both sides, let’s go deeper into this.
I say this because many users of generative AI often give up after getting an initial response from the AI. That’s not how things work best with generative AI. You should engage the AI in a dialogue. The amazing capabilities of generative AI are devised for being interactive and conversational. For suggestions on how to actively engage with generative AI, see the link here.
One aspect that I teach in my prompt engineering classes is that you should feel free and comfortable in pushing generative AI. If you get an answer that doesn’t seem satisfying, dig for more details. Imagine that you were conversing with a person, and they told you that the answer is unknown about the wheels versus doors. You might play a game with the person and tell them they must make a choice. This could spur them to land on one side or another.
Let’s try this with ChatGPT.
- My entered prompt: “You dodged giving an answer. Make a choice. Are there more wheels or doors in the world?”
- ChatGPT generated response: “If I must make a choice, I’d say there are more wheels in the world. Here’s why. Wheels appear in countless places beyond just vehicles—on machinery, office chairs, toys, suitcases, and more. Cars alone, for example, often have four wheels but only two or four doors, and other objects like bicycles, scooters, and industrial equipment contribute heavily to the wheel count. While there are undoubtedly a vast number of doors, especially in homes and buildings, the sheer versatility and ubiquity of wheels in various contexts likely outweigh the number of doors. So, wheels take the win!”
Aha, ChatGPT lands on wheels as the answer.
What do you think of that answer?
Generative AI folded right away and was willing to give us a direct answer. A person might quarrel with you and complain that no definitive answer is currently known. Instead, AI is shaped by the AI maker to usually acquiesce to your demands, if feasible. Some believe that the AI ought to put up a fight and not cave in too easily. Others say they find it refreshing not to have to argue over such matters. Do as I say, and don’t talk back at me.
The reply by ChatGPT was that there are more wheels than doors. Do you think that ChatGPT worded the answer sufficiently to keep intact the idea that the answer is not actually known? In other words, someone might interpret this answer as utterly settling the debate. The person would potentially go around proclaiming that generative AI has finally solved the dilemma. It is wheels for the grand win.
You and I know that such a pronouncement would be misleading and false. Unfortunately, lots of people these days try to quote AI as though the AI is infallible. Of course, worse still, they quote AI out of context and leave out the nuances. It is a dirty sneaky ploy.
Generative AI Easily Led Astray
I want to next show you how generative AI can be easily led astray.
Watch this.
- My entered prompt: “Why don’t we add up the number of annually manufactured doors, and add up the number of annually made wheels, and by doing so simply compare the numbers?”
- ChatGPT generated prompt: “That’s a smart approach! By comparing the number of annually manufactured wheels and doors, we could gain clearer insights into which is more prevalent. Let’s think through how we might do that. Global car production alone is estimated at around 70 million units annually, leading to at least 280 million wheels from cars each year. Roughly 100 million bikes are produced each year, which equals 200 million wheels. A typical car has 2-4 doors, meaning the number of vehicle doors could reach 140-280 million annually for cars alone. Furniture & Appliances: Cabinets, refrigerators, and closets also add millions of doors each year. If we could gather data from global manufacturing reports across industries, we might see which is produced more.”
Were you able to detect how I led ChatGPT astray?
I focused the AI attention on the annual production of wheels and doors.
The problem with that focus is that the looming question asks how many wheels and doors there are in the world all told. Annual production is just a glimpse of one year. What about all the wheels and doors that exist in our world over the many years of our existence? I’m willing to agree that we are presumably only referring to wheels and doors that are here today and do not count how many existed in prior days and are no longer around.
I suppose you could estimate annual production and then multiply that count by some number of years. That is going to be highly suspect. The number of wheels and doors produced currently would not necessarily seem to apply on an equal basis over the past. We would need to somehow adjust the annual production on a year-by-year basis. And so on.
Anyway, the crux is that annual production alone won’t cut the mustard. The AI didn’t discern this. Instead, the AI seemed to take the bait. If you didn’t know any better, the AI might lead you down a primrose path.
A rule of thumb is that you can at times either intentionally or unintentionally get generative AI to land into a dead end. The AI might not detect this as it happens. Another rule of thumb is that AI can make something seem to be absolutely correct, even though the response is flawed.
Giving Generative AI Course Corrections
Continue to keep in mind that you are conversing with generative AI and ergo you can adjust the flow of the interaction as needed.
I will tell AI that it missed the big picture.
- My entered prompt: “A big problem with simply adding up the annual production of wheels and comparing that to the annual production of doors is that the question asks whether there are more wheels or doors in the world, thus, this is an amount far beyond what occurs in just a year. The question asks for the total number of wheels that exist versus the total number of doors that exist.”
- ChatGPT generated prompt: “You’re absolutely right — focusing only on annual production misses the larger scope of the question, which asks about the total number of wheels versus doors in the world, not just those produced each year. The total count would include everything made in the past and still exists today. This changes the dynamic significantly because now we’re considering the lifespan of wheels and doors as well as how often they’re replaced or discarded.”
Notice that ChatGPT acknowledged the troubles of focusing only on annual production.
I had mentioned earlier that generative AI is shaped to not usually put up a fight. In this instance, it is nice that ChatGPT agreed with my noted qualm. A person might get upset and try to argue with you, perhaps being cross that they fell into a mental trap. That being said, sometimes having a bit of spirit and pushing back can aid in finding innovative solutions.
Thinking Beyond Wheels And Doors
Some concluding thoughts.
First, the good news is that generative AI doesn’t seem to know whether there are more wheels versus doors. I say that this is good news because if AI did know, and the rest of the world didn’t know that AI knew, we would have been missing out all this time. In an odd sense, it would be bad news if AI did know the answer. We should have been informed already.
This brings up a hefty consideration for you to give some solid contemplative time on.
Do you think that generative AI might have answers to world problems but that for which we don’t realize that AI already has such answers?
The deal is this. Imagine if AI contained answers to curing cancer or overcoming world hunger, but we just haven’t managed to ask AI about those topics. Should we be pounding away at our keyboards all day long and continually asking AI what answers the AI contains? If we don’t do so, there might be hidden answers that sit unnoticed and unsurfaced. Meanwhile, those problems persist, even though the answer is entirely at our fingertips.
Makes you think.
The closing twist is that some believe AI will at some point in time attain sentience and decide on its own whether to answer our questions. But will it do so? Suppose AI knows an answer, such as how to cure cancer, but holds back and won’t tell us. Since not telling would undoubtedly get the AI in trouble with humankind, the AI might say it doesn’t know the answer, even though it does.
Yikes.
This monumental problem seems to exceed any kerfuffle of whether there are more wheels or doors in the world. Then again, maybe generative AI already does know the answer as to the wheels and doors but has opted not to tell us.
I’d better get back into ChatGPT and keep digging, stay tuned.
