Transact-SQL
Reinforcement Learning
R Programming
React Native
Python Design Patterns
Python Pillow
Python Turtle
Verbal Ability
Interview Questions
Company Questions
Cloud Computing
Data Science
Machine Learning
Data Structures
Operating System
Computer Network
Compiler Design
Computer Organization
Discrete Mathematics
Ethical Hacking
Computer Graphics
Software Engineering
Web Technology
Cyber Security
C Programming
Data Mining
Data Warehouse
How to write as if you aim to trick others into believing that you are generative ai.
Here is what you need to know about distinguishing generative AI content versus human-written ... [+] content, and the fakery involved at times.
In today’s column, I am going to mindfully tackle a controversial topic about generative AI that will somewhat extraordinarily turn the heated matter on its head. The result is nearly comical in some ways but avidly reveals sobering considerations of quite useful value. Hang in there, this will all be worth it. For my ongoing readers, this discussion is an ongoing extension of my coverage of the latest AI trends especially with an undercurrent entailing AI Ethics and AI Law, see the link here .
The crux of this analysis has to do with the now classic question of distinguishing human-written text from text that was produced by generative AI.
It’s a big issue.
A really, really, big issue.
Let’s unpack it.
The Trouble With Generative AI As Your Ghostwriter
We normally hear about the problem of using generative as your ghostwriter in the context of schools and coursework. A student opts to use generative AI to write their essay for them. The student sneakily turns in the paper, hoping the teacher will not be the wiser. Sometimes the student manages to get away with it, scot-free. Sometimes the student gets caught. It all depends.
This is not solely a problem within a school-related context. A person at work is assigned the task of writing an important report by their boss. Rather than writing the report by hand, the person instead has generative AI write the entire report for them. They give the report to their boss. If the person is lucky, they don’t get caught and avoid potentially being fired for their transgression.
‘house of the dragon’—silverwing, vermithor, bastard riders, explained, napa valley family releases their first wines made from top cabernet sauvignon vineyards.
The work setting does seem a bit different than the school setting. In school, the student is supposed to be learning and ergo the presumption is that the student is undercutting their own education by using generative AI to do their schoolwork for them. We don’t want students to essentially cheat on their education.
As an aside, there is rampant confusion on this matter, namely, you can, in fact, have students suitably use generative AI in the right way at the right time, and therefore exhortations about potentially banning generative AI outrightly from schools are short-sighted, untenable, and a regrettably foolhardy approach though it seems to keep coming up, again and again. See my discussion on this whole conundrum and the sensible way to do things, at the link here and the link here , among others of my postings.
Back to the use case in the workplace. The work-related transgression is perhaps more so that the worker failed to alert their boss that they were seeking to write the report via generative AI. The boss might have approved of such an approach. Furthermore, another issue could be that the worker is unaware of the need to double-check the generative AI after the report has been generated. This is an essential act.
Here's why double-checking is crucial. I’ve covered the now famous or infamous case of the two lawyers who used generative AI to write a legal briefing, which got them in hot water by a judge when the briefing turned out to have falsehoods produced by the generative AI (the lawyers had not dutifully double-checked). See my analysis of the headline-grabbing incident at the link here and the link here .
Generative AI readily produces fictitious output that appears to be factual, a facet commonly referred to as AI hallucinations or said-to-be falsehoods, see the link here . Please note that I disfavor the catchphrase of AI hallucinations since it anthropomorphizes AI. In any case, the phrase has gained popularity, and the essence is apt that you must be wary of all essays and outputs arising from generative AI.
All in all, people of any age and in all walks of life might readily opt to use generative AI to do their writing for them. In the wrong circumstances, this is considered a no-no. In other circumstances, the act of using generative AI as your authoring tool might be acceptable, assuming that you are aboveboard about it. For example, many news outlets and journals require that works authored by generative AI are to be explicitly labeled as such.
The ethics and legal blurring of this is underway such that not everyone is upfront and at times hides the authorship to prevent you from realizing what is afoot. A news outlet might for example assign a fake name that represents their use of generative AI. If you later ask whether the item was written by generative AI, they might say that yes, of course, everyone knows that the fake name is an obvious reference to the AI.
Wink-wink attempts at plausible deniability ensue.
Many legal and ethics twists and turns arise, which, in case you are interested, I cover in-depth at the link here .
Modes Of Writing And The Unreliable Eyeball Detection Method
Because of the concerns that people sometimes opt to have generative AI be their ghostwriter, there is an entire cottage industry now of automated detection tools that purportedly detect whether content has been devised by the human hand versus via generative AI. There are a lot of myths about the capabilities of these detection tools, oftentimes pushed as a silver bullet by the industrial complex and others that want to make you believe that the tools are the height of perfection.
I will be sharing with you the ugly underbelly of these at-times scams or charades.
Before we can get into that aspect, we need to explore what I refer to as the eyeball detection method. I’ll first unpack the role of human eyeballing of content to assess whether the content was human-written or devised by generative AI. No automated tools are invoked, and instead, a person examines an essay and decides who or what the author might be.
Let’s start at the beginning, namely that content can be written in these two ways:
Some proclaim they can easily determine which is which by purely manual inspection alone.
For example, sadly, there are teachers who brazenly believe they can discern whether an essay was written by a student versus via the use of generative AI. The teacher will look for clues such as vocabulary words that seem outside the range of the student. After reading the essay, such teachers will insist that they can identify who or what wrote the content.
Unfortunately, those emboldened teachers are often wrong in their guesses.
A research study entitled “Do Teachers Spot AI? Evaluation The Detectability Of AI-Generated Texts Among Student Essays” by Johanna Fleckenstein, Jennifer Meyer, Thorben Jansen, Stefan D. Keller, Olaf Köller, and Jens Möller, Computers And Education: Artificial Intelligence , June 2024, made these important points (excerpts):
Teachers can readily be fooled.
The crux usually to the teacher being fooled is that an astute student using generative AI can simply instruct the AI to write in the same manner as they do. The resulting essay appears as if it is student-written and can trick unknowing teachers into falsely believing the student wrote the content. This would be considered a false negative in experimental lingo (the teacher failed to discern that an essay was written by generative AI and ergo gives credit undeservedly to a student).
I’d like to augment the two ways of writing content by noting that the AIGC has an additional two subcomponents (labeled here as 2a and 2b):
If the person using generative AI to do their ghostwriting uses a conventional prompt, the AI will use whatever default wording is customarily being used in that AI app. On the other hand, if the person realizes they want the content to seem less likely AI-written, they can merely indicate to the AI to mimic the style of say a high school student, a college student, a middle school student, or whichever form of a person or persona they wish to invoke.
Now that I’ve got you in the mental sphere that it is readily possible and extraordinarily easy to instruct generative AI to write in various fashions or styles, let’s keep going on that tack.
Suppose that we opt to use generative AI as a rewriting tool. The idea is that we will take content and ask the AI to rewrite the material. This could be a conniving means of disguising the original writer or source of the content.
Consider these two possibilities:
The first of those two would be the use case of someone who has produced a generative AI essay, realizes they will get nabbed if they turn it in, and decides to just go ahead and have the AI rewrite it to appear as if human-written. This is similar to the use case of having the AI write content from scratch and telling the AI to make it look human-written. This is a step afterward if you perchance forgot to tell the AI to write it that way at the get-go.
The second of those two use cases is a bit strange to conceive of. The mind-bending idea is that someone might tell generative AI to rewrite a human-written essay and make it look as though it was written by generative AI. I’ll get more into this momentarily, so go ahead and ponder this intriguing notion for a bit, thanks.
There are more twists and turns to be had in all of this.
One twist entails the act of making modifications to written content.
We ought to include these distinct possibilities:
In essence, the content becomes a blend of both human-written and generative AI material.
A lot of people don’t seem to realize that this is yet another easy variation within this whole concoction. You might be inspecting content that is partially written by human hand and partially written by generative AI.
The first instance entails the act of having generative AI polish or otherwise modify a human-written essay. To some degree, everyday word processing packages tend to do this already. A person can have word processing provide advice on spelling, replacement of words for other words, and so on.
A full-on use of generative AI is somewhat larger in scope and can essentially demonstrably modify the initial content. A teacher can find themselves accusing a student of using generative AI to write the essay due to seemingly telltale clues that generative AI was involved in the process. Do you think it is fair that if the student merely had generative AI do some minor polishing, akin to what happens in common word processing packages, the student is to be dinged for doing so?
That’s why teachers and schools need to establish clear rules about what is permissible and what is considered out of bounds, see my discussion at the link here .
The second instance is the sneakier of the two, maybe. For example, a student produces an essay via generative AI. The student examines the essay and discerns that despite the AI mimicking a human writing style, there are portions that still might stick out like a sore thumb. The student proceeds to change those portions by hand, perhaps using simpler words or making the sentences less semantically complex. Voila, the otherwise generative AI essay now looks even more so as though it was student-written.
I want to make sure you can plainly see how confounding all of this is.
Let’s add two more subcomponents to the structure of things, doing so as 1a and 1b:
I’ve given you above an example of the 1a instance, namely that a person might rewrite a generative AI essay so that the essay appears to have been human-written.
The second instance or 1b seems oddly curious. A person decides to write an essay that looks as though it is written by generative AI. Why would anyone do this? The overarching assumption is that no one wants their writing to look as though it was done by AI. You want credit for your own writing. If you write as though the writing smacks of AI, people are going to start pointing fingers at you and denounce you for cheating.
Buckle up for a surprise about the matter.
Are you ready?
I’d like to reword 1b as follows: “Human writes a human-written essay that looks like it was written by generative AI.” I removed the intent.
Here’s the rub.
Suppose that someone writes in their normal way of writing, and they aren’t intending to appear to write as generative AI does. So, I am saying they are just writing as they tend to write. Period. But someone else who reads the content falsely believes the essay was written by AI. They are triggered by what they believe to be AI-powered writing, even though this didn’t at all occur in this instance.
This is the dreaded “false positive” that students and others are often faced with. I get emails and responses from readers that they have been falsely accused of using generative AI. Their scholastic pursuits can be totally smeared by this. It is very hard to fight such an accusation, assuming that it is indeed a false case. How do you “prove” that you didn’t use generative AI?
The problem is muddled since there are real instances of people who cheat and use generative AI, while there are false-claimed instances of people having used generative AI when they did not. Society right now assumes that a student or person couldn’t write as well as generative AI, thus, any good writing that seems out of place is instantly labeled as a generative AI essay.
Let’s not be so quick to judge.
Part of the reason why someone might write like generative AI is that more and more people are using generative AI on a regular basis. The thing is that your writing can start to mimic the writing of generative AI.
Yes, after you see a slew of generative AI essays, it is possible to start picking up the patterns associated with generative AI writing.
I realize this seems upside down, or maybe it is the completion of a somewhat bendy cycle. We have generative AI that is based on patterns of how humans write. Next thing you know, we have humans patterning how generative AI writes. But that is sensible in the manner of admiring the writing of generative AI that in a gestalt way has mathematically and computationally figured out the “best” way to write as per a massive examination of how humans write.
Detection Tools Tend To Be Misleadingly Portrayed
If eyeballing an essay won’t necessarily be reliable in identifying human writing versus generative AI, headlines boldly urge you to use automated detection tools.
Turns out a lot of these detection tools are either hogwash or principally smoke-and-mirrors, but most people don’t realize they need to be careful and understand what the tools can attain. You should scrutinize the likelihood of false positives and false negatives. At times, the vendors treat those factors as low-key and bury the details in their licensing agreements.
A common and upsetting approach is that a school or entity will decide to adopt a detection tool and act as though it is a perfect detection device. Little if any qualms are expressed. The users such as teachers are not given any heads-up about the downsides. Just go ahead and run essays through the tool and proceed to believe without hesitation whatever the tool spits out.
It is tempting to fall into that trap. The detection tools are presumed to be a labor savings mechanism. No need to eyeball an essay. Run it through the tool. See what the tool says and proceed accordingly. No fuss, no muss.
Why aren’t the detection tools able to be fully accurate?
For the very same reasons that I cited above about eyeballing of essays, which, you saw is not an ironclad means of making these assessments.
The tools tend to rely upon spotting the use of words and the phrasing of sentences that are frequently used by generative AI apps. A computational and mathematical analysis of an essay is undertaken. The writing is compared to what generative AI tends to produce.
In a recent research paper on trying to use AI to detect AI-based writing, entitled “Monitoring AI-Modified Content at Scale: A Case Study on the Impact of ChatGPT on AI Conference Peer Reviews” by Weixin Liang, Zachary Izzo, Yaohui Zhang, Haley Lepp, Hancheng Cao, Xuandong Zhao, Lingjiao Chen, Haotian Ye, Sheng Liu, Zhi Huang, Daniel A. McFarland, James Y. Zou, arXiv, March 11, 2024, the researchers made these points (excerpts):
I bring up this study to highlight how there are various words more likely to be found in generative AI writing and therefore are presumed to be a means to detect that an essay might be AI-generated.
This study, akin to others of a similar ilk, identified for example various adjectives and adverbs that tend to be used with a heightened frequency by generative AI apps. The paper lists some Top 100 sets of words, which I thought you might like to see a sample of them.
A sampling of the identified Top 100 adjectives disproportionately used by AI:
A sampling of the identified Top 100 adverbs disproportionately used by AI:
Ponder those words, just for a few seconds or so.
Remember that I mentioned that you can tell generative AI to write as though a human is writing an essay?
If you give that kind of a prompt, one thing the generative AI can potentially do is aim to not use those Top 100 words or at least use them rather judiciously. By doing so, any detection tool that assumes those words are a likely basis for determining whether the content is written by generative AI will be possibly fooled. The counts of words that are in the Top 100 will be low, thus the program will likely give a lowered estimate of the content being written by AI.
All of this is a continual cat-and-mouse gambit.
Generative AI gets better at writing as though something is human-written. Imagine that the detection tools are behind the times and the essays are rated as more likely to be human-written than AI written. The vendors of the detection tools upgrade their wares. The odds of detecting AI writing go up. Meanwhile, the generative AI gets better at downplaying the AI writing style. The tools are again less able to make detections.
Round and round this goes.
A recent analysis of detection methods called out the difficulty associated with this ongoing cat-and-mouse gambit, as noted in a research paper entitled “Hiding the Ghostwriters: An Adversarial Evaluation of AI-Generated Student Essay Detection” by Xinlin Peng, Ying Zhou†, Ben He, Le Sun, and Yingfei Sun, arXiv , February 1, 2024, per these salient points (excerpts):
One other final point before we move on.
You might be scratching your head and wondering if we could watermark essays that are produced by generative AI.
Those of you old enough to remember would recall that a classic technique for pictures, images, and the like would be to embed a watermark into the material. Then, if there was a dispute about the owner or author, it was possible to dig into the content to see if there was a watermark there.
The use of watermarks for digital content involving graphics or images is somewhat tenable because it is relatively easier to embed a watermark. They are generally hard for others to find and remove or find and change. Once again, this is a continual cat-and-mouse gambit. New methods of digital watermarking are figured out, and those who wish to defeat them are able to figure out ways to do so. On and on this loop occurs.
Watermarking of text is a lot tougher proposition.
Note that when you watermark a graphic item, this is usually done in a hidden manner, deep within the digital guts and bits and bytes of the item. There is essentially no difference from the normal eye of what the image or picture seems to be. Text is a different beast.
To watermark text, you typically need to seed certain words or phrases into the essay. A person reading the essay will presumably read the text in an ordinary fashion and expect that the text is entirely about whatever topic is at hand. To keep them from figuring out that there is a watermark here or there, the text must be carefully calculated to fit within the context of the text. You can’t just drop into the middle of an essay some bizarre sentence or plop a few oddball words as an attempt at watermarking the content. If you do so, the gig is up. I’ve covered these tactics at the link here.
Suppose that a watermarking scheme for text-based content was nearly invisible in the sense that the essay was almost semantically the same as if it didn’t contain the watermark. Yay, watermark success.
The biggest downfall is that a person can take an essay and opt to modify or rewrite the text, and doing so could readily mess up the watermarks. They might not have any idea what portion of the text a watermark is. Nonetheless, by moving around words, substituting words, adding words, and so on, they are likely to at some point be scrambling the watermark.
Digital watermarking and generative AI are a trending and rapidly evolving topic.
There are pressures underway to legally require AI makers to include watermarking as part and parcel of their generative AI apps, see my coverage at the link here . The belief is that this will aid in distinguishing what is produced by generative AI versus by the human hand. Regrettably, oftentimes lawmakers and regulators aren’t aware of how readily these can be defeated, especially in the case of text-based outputs. The result is that a strident push for watermarks can occur that is based on an outsized belief of what this will actually accomplish.
The viewpoint of some is that even if watermarks can be tricked, at least it is better to have them legally required versus not having them. For my in-depth analysis of the upsides and downsides, see the link here .
When You Want To Make Your Writing Look Like Generative AI
I have walked you through the whole kit and kaboodle about generative AI content versus by-hand content. My hope is that the quick tour was instructive and engaging. As your reward for having gone on that arduous journey, let’s do something that very few talk about.
First, grab a glass of fine wine and ready yourself for some useful fun.
Suppose you wanted to compose an essay that seemed to be written by generative AI.
Could you do so?
Yes, I would say you most certainly can.
All you need to do is mimic generative AI. I’m not saying that you need to somehow dive into the mathematical and computational formulations of generative AI. Nope, no need to do so. What you need to do is study the essays generated by AI. Mimic those essays and the writing style being used.
We are turning the world on its head. The rest of the world is focused on not writing like generative AI. You are supposed to write as a human would write. If you write like generative AI, you will be accused of having used generative AI.
Furthermore, if you go onto social media and online forums, and if you write in the style of generative AI, the odds are that you will be accused of being generative AI. Tons of AI chatbots are spewing content onto online arenas. Nobody can be sure if posted content is by a human being versus AI.
Be cautious because once you are labeled by the wisdom of the crowd as being generative AI, you might have a devil of a time proving otherwise. No matter what pictures or videos you provide to show proof of humanness will seemingly further convince others that you aren’t an entirely AI-contrived avatar. You will live your life cursed as being a generative AI.
Just wanted to give you a heads-up and sufficient warning.
You’ve been warned, proceed with caution.
Let’s put our heads together and figure out what to do if wanting to write like generative AI. I will ask that you play fair. An unfair method would be to have generative AI merely produce an essay for you. Thus, you didn’t lift a finger to write the essay. Cheater. Either go big or go home.
We are going to write from scratch an essay that would seem to be written by generative AI. The goal is that a person reading the essay will get their Spidey-sense tingling that the essay is written by AI. On top of this, the aim is that any of the detection tools would be fooled into scoring the essay as likely written by generative AI.
Here’s how we will proceed in terms of steps to undertake:
Let’s briefly try this.
First, I will share with you one of the most outrageous ways to accomplish this that in the right situation can be nearly surefire for fooling a human. It is done with immense ease.
Some of the generative AI apps such as ChatGPT tend to use this now classic line: “Please note that as an AI language model, I am …”. For example, you might enter a prompt into generative AI telling the AI to start crying, and the response often is something like this: “Please note that as an AI language model, I am unable to cry.”
You can write an essay by hand and include that line.
The person reading the essay is essentially clobbered over the head that the essay is written by generative AI. They would have to be entirely out to lunch to otherwise not assume that the essay was written by AI. Boom, drop the mic.
If you want to keep their suspicion somewhat low that they are maybe being tricked, make sure to include the line further into the essay rather than at the opening. Your rookie mistake would be to maybe have it be the first line and tip your hand too soon. This seems like a dead giveaway. On the other hand, you might have someone reading the essay that thinks they are a genius at noticing that the opening sentence tells that it is written by AI. Make them happy at their cleverness and good fortune at being smarter than an average bear.
You must gauge your audience accordingly.
Believe it or not, there have been research papers that used generative AI to either write the paper or rewrite the paper and opted to include the line about being an AI language model when they submitted the paper to journals for review. They weren’t trying to be tricky. They were caught with their hands in the cookie jar and asleep at the wheel. Worse still, sometimes these have managed to be published that way. A huge embarrassment to the journal, as you can imagine.
The key is that at this time the line about being an AI language model is assumed by the population at large as a sure sign that the essay was composed by AI. Until or if we ever see enough people using the same line as a joke or to fake out others, the line will remain a handy means of pulling the wool over the eyes of the unsuspecting.
There are numerous viable variations of the line in case you want to mix things up: "As an AI", "As an AI assistant", "As an AI developed by <AI maker>", whereby the AI maker could be OpenAI, Anthropic, Google, Meta, and so on.
You cannot stop though with that one trick pony.
Few of the detection tools will fall hook, line, and sinker for one line. They are going to proceed to calculate the rest of the essay. Even if the one line is there, the rest of the essay might not meet the expectations of an AI-written piece of content.
You should study the words that are commonly used by generative AI and include those throughout your essay. I gave you some of the words above. Use them to your heart’s content.
Try to write the essay in a bland manner. Be straight ahead. The usual default for generative AI is that the AI makers have data trained and refined the algorithms to write conventional sentences. Remember that in this tomfoolery we are mimicking the day-to-day use of generative AI.
Seek to be polite and formal. This ought to include phrases such as “In conclusion”, “One should consider”, “It is important to note”, etc. Toss in a lot of those.
Transition words are popular in generative AI. Use for example these transitions throughout the essay: "Furthermore,”, "Moreover,", "However,", "In addition," and others that come to mind.
Beauties to sprinkle include: "I hope this helps!", "If you have any more questions, feel free to ask.", "You're welcome", "If you have any questions”, and similar phrases. It is okay that those might seem out of place in the essay. A human will assume that the generative AI was too dumb to realize that those wordings ought to not have been included in the middle of things.
An important consideration is that you are trying to put yourself into the mind of a human reader and aiming to fool them into thinking the essay is written by generative AI. Placing yourself into the shoes and the mind of another person is generally referred to as Theory of Mind (ToM), see my discussion at the link here .
So far, we are doing this solely by mimicking generative AI. The other angle is to bet on the basic assumptions people have about how generative AI essays are written. They might not be right in those assumptions, but we don’t care. Triggering their rash or unsupported assumptions is the key here. Adopt their way of thinking. Exploit ToM.
My favorites are these lines: "Based on my training data", "From what I've been trained on", "According to the data I have access to", and "My training includes information up to <date>". I like those lines because you aren’t being as obvious as if you said, “I am AI” and instead you are implying that you must be AI due to the notion of being data trained. Nice touch.
Whatever you do, avoid curse words like the plague. I am not saying that generative AI won’t ever use curse words. In fact, you can instruct generative AI to do so, and in some mild ways, the AI will make use of foul words, though the AI makers have done a lot of filtering to try and stop this from happening, see my coverage at the link here . The gist is that most people assume that generative AI would rather fall apart than employ curse words, so go ahead and play into that as a helpful ploy.
I believe this will get you started on the path toward writing essays that seem to be written by generative AI.
Good luck, I’m sure you can pull this off.
Congratulations, overall, you now know a lot about generative AI and the ongoing controversy of using generative AI as a ghostwriter.
A final comment for the time being.
You might be familiar with an adage about being the real McCoy. There is debate over how the catchphrase came into existence. One version is that Kid McCoy, a boxing champion, was in a bar and got challenged to a fight. Supposedly, as the story goes, he knocked the other person down with one blow. To which, the person on the floor remarked, that’s the real McCoy.
When you are assessing an essay, please be careful and mindful about declaring who or what is the real McCoy.
Accusing a human of using generative AI to write their content is currently a rather societally disagreeable offense and the person could end up being unduly tarnished. Consider the risks and costs of falsely accusing someone of such a transgression.
Keep in mind too that astute users of generative AI can easily get the AI to appear to be the real McCoy, in the sense of mimicking a human writing style. This means that there are bound to be essays that were written by generative AI but that can be extremely hard if not impossible to pin down as being written in that manner. The essay will seem to be handwritten and there aren’t enough telltale clues to faithfully state otherwise.
I’m not saying that so-called cheaters should go scot-free. I am saying that you need to be careful of abiding by one of those most important principles that we dearly hold sacred in our society and culture. Namely, innocent until proven guilty. Be cautious in making accusations, since an innocent person accused of something is still going to be tainted.
Try your darndest to first identify the real McCoy, before finger-pointing takes place.
One Community. Many Voices. Create a free account to share your thoughts.
Our community is about connecting people through open and thoughtful conversations. We want our readers to share their views and exchange ideas and facts in a safe space.
In order to do so, please follow the posting rules in our site's Terms of Service. We've summarized some of those key rules below. Simply put, keep it civil.
Your post will be rejected if we notice that it seems to contain:
User accounts will be blocked if we notice or believe that users are engaged in:
So, how can you be a power user?
Thanks for reading our community guidelines. Please read the full list of posting rules found in our site's Terms of Service.
July 17, 2024
It’s important that we use accurate terminology when discussing how AI chatbots make up information
By Joe Slater , James Humphries & Michael Townsen Hicks
Malte Mueller/Getty Images
Right now artificial intelligence is everywhere. When you write a document, you’ll probably be asked whether you need your “AI assistant.” Open a PDF and you might be asked whether you want an AI to provide you with a summary. But if you have used ChatGPT or similar programs, you’re probably familiar with a certain problem— it makes stuff up , causing people to view things it says with suspicion.
It has become common to describe these errors as “ hallucinations .” But talking about ChatGPT this way is misleading and potentially damaging. Instead call it bullshit.
We don’t say this lightly. Among philosophers, “bullshit” has a specialist meaning , one popularized by the late American philosopher Harry Frankfurt . When someone bullshits, they’re not telling the truth, but they’re also not really lying. What characterizes the bullshitter, Frankfurt said, is that they just don’t care whether what they say is true. ChatGPT and its peers cannot care, and they are instead, in a technical sense, bullshit machines.
If you're enjoying this article, consider supporting our award-winning journalism by subscribing . By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.
We can easily see why this is true and why it matters. Last year, for example, one lawyer found himself in hot water when he used ChatGPT in his research while writing a legal brief . Unfortunately, ChatGPT had included fictitious case citations. The cases it cited simply did not exist.
This isn’t rare or anomalous . To understand why, it’s worth thinking a bit about how these programs work . OpenAI’s ChatGPT, Google’s Gemini chatbot and Meta’s Llama all work in structurally similar ways. At their core is an LLM—a large language model. These models all make predictions about language. Given some input, ChatGPT will make some prediction about what should come next or what is an appropriate response. It does so through an analysis of enormous amounts of text (its “training data”). In ChatGPT’s case, the initial training data included billions of pages of text from the Internet.
From those training data, the LLM predicts, from some text fragment or prompt, what should come next. It will arrive at a list of the most likely words (technically, linguistic tokens ) to come next, then select one of the leading candidates. Allowing for it not to choose the most likely word each time allows for more creative (and more human-sounding) language. The parameter that sets how much deviation is permitted is known as the “temperature.” Later in the process, human trainers refine predictions by judging whether the outputs constitute sensible speech. Extra restrictions may also be placed on the program to avoid problems (such as ChatGPT saying racist things ), but this token-by-token prediction is the idea that underlies all of this technology.
Now, we can see from this description that nothing about the modeling ensures that the outputs accurately depict anything in the world. There is not much reason to think that the outputs are connected to any sort of internal representation at all. A well-trained chatbot will produce humanlike text, but nothing about the process checks that the text is true, which is why we strongly doubt an LLM really understands what it says.
So sometimes ChatGPT says false things. In recent years, as we have been becoming accustomed to AI, people have started to refer to these falsehoods as “ AI hallucinations .” While this language is metaphorical, we think it’s not a good metaphor.
Consider Shakespeare’s paradigmatic hallucination in which Macbeth sees a dagger floating toward him. What’s going on here? Macbeth is trying to use his perceptual capacities in his normal way, but something has gone wrong. And his perceptual capacities are almost always reliable—he doesn’t usually see daggers randomly floating about! Normally his vision is useful in representing the world, and it is good at this because of its connection to the world.
Now think about ChatGPT. Whenever it says anything, it is simply trying to produce humanlike text. The goal is simply to make something that sounds good. This is never directly tied to the world. When it goes wrong, it isn’t because it hasn’t succeeded in representing the world this time; it never tries to represent the world! Calling its falsehoods “hallucinations” doesn’t capture this feature.
Instead we suggest, in a June report in Ethics and Information Technology , that a better term is “bullshit.” As mentioned, a bullshitter just doesn’t care whether what they say is true.
So if we do regard ChatGPT as engaging in a conversation with us—though even this might be a bit of a pretense —then it seems to fit the bill. As much as it intends to do anything, it intends to produce convincing humanlike text. It isn’t trying to say things about the world. It’s just bullshitting. And crucially, it’s bullshitting even when it says true things!
Why does this matter? Isn’t “hallucination” just a nice metaphor here? Does it really matter if it’s not apt? We think it does matter for at least three reasons:
First, the terminology we use affects public understanding of technology, which is important in itself. If we use misleading terms, people are more likely to misconstrue how the technology works. We think this in itself is a bad thing.
Second, how we describe technology affects our relationship with that technology and how we think about it. And this can be harmful. Consider people who have been lulled into a false of security by “self-driving” cars . We worry that talking of AI “hallucinating”—a term usually used for human psychology—risks anthropomorphizing the chatbots. The ELIZA effect (named after a chatbot from the 1960s) occurs when people attribute human features to computer programs. We saw this in extremis in the case of the Google employee who came to believe that one of the company’s chatbots was sentient . Describing ChatGPT as a bullshit machine (even if it’s a very impressive one) helps mitigate this risk.
Third, if we attribute agency to the programs, this may shift blame away from those using ChatGPT, or its programmers, when things go wrong. If, as appears to be the case, this kind of technology will increasingly be used in important matters such as health care , it is crucial that we know who is responsible when things go wrong.
So next time you see someone describing an AI making something up as a “hallucination,” call bullshit!
This is an opinion and analysis article, and the views expressed by the author or authors are not necessarily those of Scientific American.
Imagine meeting with the daughter of a critically ill patient. The patient (her mother) had a cardiac arrest, is in multiorgan failure, and cannot communicate. The daughter is uncomfortable making decisions because they are estranged and never discussed what her mother would want in this type of situation. The patient has no advance directive or alternative surrogate.
Now imagine this meeting taking place in a future where the mother’s medical visits have been audio recorded. Furthermore, you have access to an artificial intelligence (AI) algorithm that can identify and play excerpts of the mother talking about what mattered most to her. You and the daughter listen to these recordings together. Then you share that another algorithm, trained on 7 million patient records, predicts that the mother’s chance of walking again is less than 5%. It also predicts that 3 of 4 people like her—a registered Sierra Club member and volunteer dog walker who regularly purchased gardening equipment—would want to pursue comfort-focused treatment in this situation. Acknowledging that predictions cannot remove all uncertainty, you ask the daughter what she is thinking and feeling about this news.
Brender TD , Smith AK , Block BL. Can Artificial Intelligence Speak for Incapacitated Patients at the End of Life? JAMA Intern Med. Published online July 22, 2024. doi:10.1001/jamainternmed.2024.2676
© 2024
Artificial Intelligence Resource Center
Best of JAMA Network 2022
Browse and subscribe to JAMA Network podcasts!
Select your interests.
Customize your JAMA Network experience by selecting one or more topics from the list below.
IMAGES
VIDEO
COMMENTS
An essay hook generator can help you come up with creative ideas for hooks that you can use to make your essay stand out. By using a hook generator, you can save time and effort while creating a hook that will make your essay more engaging. An essay hook generator is a tool that generates creative ideas for hooks for essays. It is designed to ...
Artificial intelligence (AI) is a kind of technology that makes the devices smart as human beings to develop the human's life by using these devices in... Artificial Intelligence Advantages of Technology. Topics: Expert system, Fuzzy logic, Intelligent agent, Knowledge engineering, Problem solving. 2.
Here are examples of six hooks you could use to start a persuasive essay about artificial intelligence, plus three hooks you could use to start a sci-fi story. Example 1: Question Hook. Will artificial intelligence someday become smarter than humans? Example 2: Statement Hook. Artificial intelligence could become smarter than humans by 2050.
There are different types of essay hooks that you can use to capture the reader's attention. Here are some common types: Anecdote: Start your essay with a short and interesting story that relates to the topic. This personal touch can immediately engage the reader. Question: Begin your essay with a thought-provoking question that the reader can ...
2. Bold claim hook. When working on an argumentative essay, I always write with the mindset that nobody has the time to read my thoughts from start to finish.So, I have to get to the point quickly and make a solid argument worth people's time.. That's when opening with a bold claim works best. Condense all your views on the topic into a few thought-provoking lines that would make readers go ...
If you don't have a great essay hook, the chances are you won't capture the attention of the professor and this will impact the grade you receive on the essay. ... Let's imagine you are writing an essay about using artificial intelligence to create articles. An intriguing rhetorical question you could use for the hook of the essay could ...
An AI hook generator is a tool that uses artificial intelligence to generate a sentence or phrase that is intended to grab the reader's attention and entice them to read further. The purpose of a hook generator is to help writers come up with ideas for opening sentences that will be interesting and engaging to their readers.
Here are seven ideas to choose from: 1. Story. Everyone likes a good story. If an interesting story or anecdote relates to your essay topic, the hook is a great place to include it. For example: In January 2023, two children were playing outside in a Los Angeles neighborhood.
107 Artificial Intelligence Essay Topic Ideas & Examples. Artificial Intelligence (AI) is a rapidly advancing field that has the potential to revolutionize various aspects of our lives. From self-driving cars to virtual assistants, AI is already making a significant impact on society. If you are tasked with writing an essay on artificial ...
Here's an example of the steps you can follow that help you outline your essay. First paragraph: Establish the thesis. Body paragraphs: Supporting evidence. Last paragraph: Conclusion with a restatement of the thesis. Revisit the first paragraph: Find the best hook. Obviously, the first step is to determine your thesis.
Intelligence is the ability to carry on abstract thinking. It is the capacity for flexible adjustment. Intelligence is the degree of availability of one's experiences for the solution of immediate problems and the anticipation of the future ones. Intelligence is the capacity for constructive thinking, which involves a discovery of appropriate ...
Select Hook Type: Choose the type of hook you want (e.g., anecdotal, question, quotation, statistic, or statement—see above for explanation of the differences). Think about what type of essay you're writing, and select the appropriate hook type. Generate Hook: Click the generate button. Our tool processes your input and selected hook type ...
Once you understand your audience, the next step is to clarify the purpose of your essay to craft an effective hook. Knowing your essay's purpose helps you tailor your hook to align with the overall goal, making it more impactful. ... A Hook Generator leverages artificial intelligence to craft these attention-grabbing openings. Users provide ...
Here are six tips to help make the essay writing process easier. Write better essays with AI >. . 1. Create a stronger thesis statement using AI. A thesis statement helps you stay on-track with your essay's overall aim. It's a brief statement that summarizes what your essay is about and what readers can expect.
Artificial Intelligence Essay: Hook Examples. Stephen Hawking's Vision: "Everything that civilization has to offer is a product of human intelligence; we cannot predict what we might achieve when this intelligence is magnified by the tools that AI may provide, but the eradication of war, disease, and poverty would be high on anyone's list. . Success in creating AI would be the biggest ...
88 essay samples found. A college essay on AI allows one to delve into the intriguing world where machines and algorithms shape the future. It demonstrates how exciting the field of advanced technology and its impact on human society can be. When preparing a persuasive and argumentative essay on artificial intelligence, it is essential to ...
106 Artificial Intelligence Essay Topics & Samples. 7 min. In a research paper or any other assignment about AI, there are many topics and questions to consider. To help you out, our experts have provided a list of 76 titles, along with artificial intelligence essay examples, for your consideration. Table of Contents.
Type 3: Theory of mind - This refers to understand others. Above all, this means to understand that others have their beliefs, intentions, desires, and opinions. However, this type of AI does not exist yet. Type 4: Self-awareness - This is the highest and most sophisticated level of Artificial Intelligence. Such systems have a sense of self.
Ahrefs' Hook Generator uses a language model that learns patterns, grammar, and vocabulary from large amounts of text data - then uses that knowledge to generate human-like text based on a given prompt or input. The generated text combines both the model's learned information and its understanding of the input.
Essay Topics. Artificial intelligence (AI) is a rapidly growing field, and writing an essay on it can be an excellent way to explore its various aspects. An artificial intelligence essay can be a great opportunity to discuss the advantages and disadvantages of AI and how it can impact society. It can also explore the technical details of AI ...
Today, some of the events and impact of AI on the education sector are concentrated in the fields of online learning, task automation, and personalization learning (Chen, Chen and Lin, 2020). The COVID-19 pandemic is a recent news event that has drawn attention to AI and its role in facilitating online learning among other virtual educational ...
AI. Artificial Intelligence, at its core, is the simulation of human intelligence processes by machines, especially computer systems. These processes include learning (the acquisition of information and rules for using the information), reasoning (using rules to reach approximate or definite conclusions), and self-correction.
Examples of AI-Artificial Intelligence. In this topic, we are going to provide an essay on Artificial Intelligence. This long essay on Artificial Intelligence will cover more than 1000 words, including Introduction of AI, History of AI, Advantages and disadvantages, Types of AI, Applications of AI, Challenges with AI, and Conclusion.
A student opts to use generative AI to write their essay for them. ... Köller, and Jens Möller, Computers And Education: Artificial Intelligence, June ... tools will fall hook, line, and sinker ...
Right now artificial intelligence is everywhere. When you write a document, you'll probably be asked whether you need your "AI assistant." Open a PDF and you might be asked whether you want ...
Our mission is to ensure that artificial general intelligence—AI systems that are generally smarter than humans— benefits all of humanity. If AGI is successfully created, this technology could help us elevate humanity by increasing abundance, turbocharging the global economy, and aiding in the discovery of new scientific knowledge that changes the limits of possibility.
W e have reached a moment of reckoning about what artificial intelligence means for the human experience. This is a moment of reckoning, too, for higher education. This is a moment of reckoning ...
The rapid progress being made on artificial intelligence means that we face a strategic choice about what kind of world we are going to live in: Will it be one in which the United States and ...
Furthermore, you have access to an artificial intelligence (AI) algorithm that can identify and play excerpts of the mother talking about what mattered most to her. You and the daughter listen to these recordings together. Then you share that another algorithm, trained on 7 million patient records, predicts that the mother's chance of walking ...