You talk about people being “punished” for disclosing their use of AI. But is it really punishment if someone simply decides not to read material that was made using AI? Isn’t that just people deciding for themselves what they want to consume? I sometimes get the impression that authors and artists feel entitled to have people consume their work, regardless of how it was made, and perceive it as an attack when someone simply opts out.
Thoroughly found this piece interesting. Being a writer myself pre and after the rise of AI, I can see the difference whether on the tone and style I use to write (which isn't necessarily bad). The point is just to have boundaries even while working with these technologies. That way you don't eventually get lost.
This was a fantastic read! I’m so glad I stumbled upon it. If anything, I trust you *more*, BUT I may be very biased, because I don’t think it’s thanks to the skills you shared, but more so because you’re citing papers I’ve been reading lately, some of them I’m quoting in my next essay (not published yet). So I’d say that trust was based on common ground and a shared framework, which may or may not answer your question 😉
Looking forward to engaging with more of your work!
PS: editing after reading your reply rules to add that I hope this does not qualify as pitching or self-promotion. If it does, please let me know and I’ll edit my reply accordingly 😊
On the topic of Google Search being borked and using AI platforms for research, does it not go along the lines of AIs being “seductive” and bring hidden biases?
Google Search is indeed borked, but a direction of querying a database is theoretically more objective than what an AIs deems as relevant and decides what is a result or not.
Since AIs use tools, they might very well be indexing the same content as Google, but with extra layers.
I don’t disagree. You can see my research prompt. I also have profile level instructions to guard against it and I reinforce that I prefer an argumentative AI in my “memory” but it is definitely not straightforward.
And some people regretted paying a locksmith for opening their locked door in seconds, though they have waited outside their locked home for hours! People don’t appreciate expertise when things appeared “so effortless” to an expert!
Thankyou for sharing your detailed and instructive practices and prompts; greatly appreciated. In terms of your comment regarding epistemic injustice, long before AI-users were diminished or made invisible by such injustice, women, minorities and marginalised others have experienced this.
You make an extremely important point that is often missed -- if you use AI for research, you must follow the sources and vet them. LLMs are notorious for hallucinating and creating authors and other kinds of references out of nothing. They also confuse similar titles and conflate sources. A huge problem with AI tools is the unquestioning trust people place in them.
Hi, Rachael. I like this piece and your take on the effort-conundrum. I openly admit to being biased by--and reactant to my own--perceived effort. The discrepancy between valued effort and our work to avoid it is also well-taken. I see the two as related: I know how hard I would have to work to accomplish this outcome, that's exactly why I value the results of others who have committed to it. Expertise is when accumulated effort leads to high-value production with diminished effort. The "dues have been paid", and the expert now reaps and shares the reward.
My concerns with AI-proliferation have been captured in other comments but I have my own slant to them: a) Very nearly all the training has been stolen without credit or recompense, and is being sold back to users; b) AI hallucinates. Humans can be wrong or biased, but they cannot be wrong as quickly or effortlessly. The sheer speed and volume of AI-generated research and public discourse makes fact-checking nearly impossible. What chance does the casual consumer have to evaluate accuracy? Even responsible AI-research use tales time and will be over-run by AI misuse; and most damning I think, c) Reliance on AI instead of hard-won, collaborative and sometimes contentious expertise is robbing its users of critical-thinking and evaluation skills. These take time and real effort to train and hone. Their absence forces further reliance on AI to summarise and evaluate for you, and worse, leave people vulnerable to propaganda, memetic manipulation, and identity-political polarisation beyond any hope of real independent self-driven fact checking.
To your point about cited AI-use, I recognize the value of agentic research for canonical sources, but am concerned that the agent could directly guide and misdirect your use of classification of sources as "canonical". Using AI for editing support is also valuable and reasonable, if supported and backed up by real historical familiarity with grammar to avoid exactly the drift and memetic inclusion you cite. I have never see the SEO use you describe and have nonway to evaluate it. I am mildly suspicious that it looks like marketing language adjustment to support distribution by the very AI you are so careful to limit elsewhere. These look like reasonable tool-use, rather than a value-reducing crutch, and thus do not devalue your work. They raise minor concerns about contamination for me, though.
Holy heck--how did this comment grow a soap-box under my own feet?
The SEO stuff is definitely a point of discomfort. It is essentially a process you do at the end of the post to make sure you are giving yourself the best chance of being indexed and surfaced when people search. As I say in the piece I don’t just accept all its recommendations but where a subheading or SEO headline change feels appropriate as well as improving searchability I will take that advice. So it is more about the furniture around the publication rather than substantive.
Thanks Rachael. After hitting post, I was momentarily concerned that I had TLDR'd myself.
Fair call regarding SEO; If you seek to expand readability regarding AI use and the effect of self-nomination, there is a slightly meta value it hijacking AI to share your article.
Really good piece, Rachael. Thanks for sharing your openness to using it and for the prompts. (Might try the research one).
I used CoPilot and ChatGPT last year for consumer research when I was putting a marketing strategy together. I found the research was largely inaccurate (c70%) when I dug deeper. Could have been the quality of my prompt, tbf.
I’ll be writing another 12m strategy in a couple of months. I’ll let you know how the prompt goes.
I like the way you use AI - you fact check, you do the reading - you, as the writer, understand the source material and that's heavy listing. And I realize that sentence actually sounds like I'm a subject in the study on why human's like things to be harder for other folks. (I could read so many articles on that - fascinating), but my dislike of AI is: 1) distrust (as you pointed out, it can mischaracterize or straight up be wrong) 2) negative environmental impact 3) negative impact on cognition and 4) stealing of copyrighted materials.
To point 1, the majority of users aren't as dedicated or trained on AI as you and won't go the extra lengths. We're even seeing kids skip over assignments in schools in favor of using AI.
I would LOVE this article to be required reading for high schools, particularly the amount of effort you put in to prompts.
Sorry. Maybe I didn't explain well myself. It was a compliment to your prompts at the end of the essay. In my view, a good way of using these tools. And yes, they require thinking to design to obtain effective outputs.
100% - I meant that it feels strangely intimate to expose your prompt architecture to scrutiny. Like people rifling through your mind a little bit hahahaha
Such an interesting read. It reminded me of arguing the merits of being “innocent until proven guilty” versus “guilty until you prove you’re innocent”. Very different standards and how you feel about it typically depends on whether you’re a victim of crime or a falsely accused. I also remember it being very gendered especially with particular classes of crime. Prima facie anyone? I wonder what the AI skew looks like? So interesting. Thanks R.
People penalise disclosure because they read it as effort avoided, and that 2016 finding says what they were actually valuing was the implied care, regardless of outcome.
So, the penalty isn't about AI. It's a care judgment wearing a costume.
Which might mean the way through is showing the effort that didn't go away. Producing got cheap. Answering for the thing did not. I keep a log of everywhere a machine touched my work and I sign it, and the signature is the part that still costs something, because it's what I can't take back when the work turns out wrong.
Interesting article Rachael. I suppose a couple of things spring to mind. The first is an obvious one, in that not all papers are publically available. You say you check all of the sources AI produces for you. Do you have licenses for the various journals to allow you to do that?
Secondly, there appear to be two ways of doing things. One is to read widely and draw conclusions because that's where the evidence is pointing. The second is to have a conclusion and then try and find evidence to support it. It doesn't appear that you asked for any studies to counter your point in your article, so you appear to be doing the second. Is that fair?
Now you are correct that I couldn’t find a publicly accessible full version of the effort heuristic. That is a seminal work in its field and well known, and I did read it when I was at university. You can check my interpretation against the abstract I provided if you think I’m wrong about its findings and conclusions.
On the second point, I actually love being challenged and do ask … everyone and anything I can for rebuttal or different points of view. The problem you’re identifying here is real but it’s not an AI problem. It’s a matter of human judgement that could manifest just as easily if you were doing your research on microfiche at the library.
I’m curious whether you disagree with any of the actual points made? I am always happy to hear your alternative point of view :)
I've read numerous studies over the past few years on AI in creativity (and our perception of it), and a common thread is that people don't mind AI-generated work until they're told that it is produced by AI, then they flip. That could be a time thing, in that we want people to invest time in producing something, 100%. My career has straddled academia, consulting, and journalism, so I'll add a couple more penneth to the mix in the sense of authenticity and credibility.
Authenticity is important because writing is an inherently human endeavour, and I think we want to know that the person putting their name to something is the author of that something (I’m not a fan of ghostwriting for the same reason). It’s like if you go to a music concert, you want the performers to actually perform, not lip synch.
On the credibility side, this is more aimed at using AI for research. The various examples from the consulting world showcase this, where citations were included that the authors clearly hadn’t read because most of them didn’t exist. You can also see this in the Jason Arday situation at the moment, which while not AI-related, does underline what we expect of people. If you’re writing something and being in a position to write something because of your expertise, then don’t fake it by using AI.
Those are big things. I should probably add a third but I’ve never regarded myself as a good writer so it’s easy to discount, but being able to write well and put thoughts together in a readable way is also important. I’d hate to think that skill will be eased out simply because getting AI to write is good enough.
And I fully get that there is perhaps good use of AI, such as finding papers that are then read in full and digested, but I wonder how many people would take that path. I read a study this week on, essentially, AI and laziness, and when AI does the first run at something, people were much less inclined to try and improve on it. It was deemed good enough.
Which leads into a final point, in that various studies have shown that it's struggling with things that underpins how we learn. Maybe in a subtle way that underpins why people feel so strongly about effort. To achieve mastery it should be hard. There should be friction. AI is removing that, and that worries me in terms of what it's doing to us.
Hopefully that adds a bit to the discussion. I enjoyed reading your article.
The disclosure dilemma is so real. We say we want honesty about AI, then still punish people for admitting to it.
We have been treating visible struggle as evidence of care. When AI removes some of that struggle, we instinctively assume the value has disappeared too. But perhaps the effort has simply moved elsewhere, into deciding what to ask, what to accept, and what to reject.
Publishing all the prompts was a smart move. Curious to see how people react to that level of transparency.
Ohhhhh my mistake! You’re just a troll! Hahahah whaaaatevs. If being rude to internet strangers makes you feel good about yourself then you do you, I guess?
I do apologise if you were insulted, I meant it as agreement. I was trying to say I feel like you can tell when something has a person behind it. I was agreeing with you.
I appreciate the radical transparency.
You talk about people being “punished” for disclosing their use of AI. But is it really punishment if someone simply decides not to read material that was made using AI? Isn’t that just people deciding for themselves what they want to consume? I sometimes get the impression that authors and artists feel entitled to have people consume their work, regardless of how it was made, and perceive it as an attack when someone simply opts out.
Thoroughly found this piece interesting. Being a writer myself pre and after the rise of AI, I can see the difference whether on the tone and style I use to write (which isn't necessarily bad). The point is just to have boundaries even while working with these technologies. That way you don't eventually get lost.
This was a fantastic read! I’m so glad I stumbled upon it. If anything, I trust you *more*, BUT I may be very biased, because I don’t think it’s thanks to the skills you shared, but more so because you’re citing papers I’ve been reading lately, some of them I’m quoting in my next essay (not published yet). So I’d say that trust was based on common ground and a shared framework, which may or may not answer your question 😉
Looking forward to engaging with more of your work!
PS: editing after reading your reply rules to add that I hope this does not qualify as pitching or self-promotion. If it does, please let me know and I’ll edit my reply accordingly 😊
Not at all - and I really appreciate your feedback!!
Wonderful! Thanks again for this great article.
Great article!
On the topic of Google Search being borked and using AI platforms for research, does it not go along the lines of AIs being “seductive” and bring hidden biases?
Google Search is indeed borked, but a direction of querying a database is theoretically more objective than what an AIs deems as relevant and decides what is a result or not.
Since AIs use tools, they might very well be indexing the same content as Google, but with extra layers.
I wish this was simpler…
I don’t disagree. You can see my research prompt. I also have profile level instructions to guard against it and I reinforce that I prefer an argumentative AI in my “memory” but it is definitely not straightforward.
And some people regretted paying a locksmith for opening their locked door in seconds, though they have waited outside their locked home for hours! People don’t appreciate expertise when things appeared “so effortless” to an expert!
Thankyou for sharing your detailed and instructive practices and prompts; greatly appreciated. In terms of your comment regarding epistemic injustice, long before AI-users were diminished or made invisible by such injustice, women, minorities and marginalised others have experienced this.
You make an extremely important point that is often missed -- if you use AI for research, you must follow the sources and vet them. LLMs are notorious for hallucinating and creating authors and other kinds of references out of nothing. They also confuse similar titles and conflate sources. A huge problem with AI tools is the unquestioning trust people place in them.
Hi, Rachael. I like this piece and your take on the effort-conundrum. I openly admit to being biased by--and reactant to my own--perceived effort. The discrepancy between valued effort and our work to avoid it is also well-taken. I see the two as related: I know how hard I would have to work to accomplish this outcome, that's exactly why I value the results of others who have committed to it. Expertise is when accumulated effort leads to high-value production with diminished effort. The "dues have been paid", and the expert now reaps and shares the reward.
My concerns with AI-proliferation have been captured in other comments but I have my own slant to them: a) Very nearly all the training has been stolen without credit or recompense, and is being sold back to users; b) AI hallucinates. Humans can be wrong or biased, but they cannot be wrong as quickly or effortlessly. The sheer speed and volume of AI-generated research and public discourse makes fact-checking nearly impossible. What chance does the casual consumer have to evaluate accuracy? Even responsible AI-research use tales time and will be over-run by AI misuse; and most damning I think, c) Reliance on AI instead of hard-won, collaborative and sometimes contentious expertise is robbing its users of critical-thinking and evaluation skills. These take time and real effort to train and hone. Their absence forces further reliance on AI to summarise and evaluate for you, and worse, leave people vulnerable to propaganda, memetic manipulation, and identity-political polarisation beyond any hope of real independent self-driven fact checking.
To your point about cited AI-use, I recognize the value of agentic research for canonical sources, but am concerned that the agent could directly guide and misdirect your use of classification of sources as "canonical". Using AI for editing support is also valuable and reasonable, if supported and backed up by real historical familiarity with grammar to avoid exactly the drift and memetic inclusion you cite. I have never see the SEO use you describe and have nonway to evaluate it. I am mildly suspicious that it looks like marketing language adjustment to support distribution by the very AI you are so careful to limit elsewhere. These look like reasonable tool-use, rather than a value-reducing crutch, and thus do not devalue your work. They raise minor concerns about contamination for me, though.
Holy heck--how did this comment grow a soap-box under my own feet?
The SEO stuff is definitely a point of discomfort. It is essentially a process you do at the end of the post to make sure you are giving yourself the best chance of being indexed and surfaced when people search. As I say in the piece I don’t just accept all its recommendations but where a subheading or SEO headline change feels appropriate as well as improving searchability I will take that advice. So it is more about the furniture around the publication rather than substantive.
Thanks Rachael. After hitting post, I was momentarily concerned that I had TLDR'd myself.
Fair call regarding SEO; If you seek to expand readability regarding AI use and the effect of self-nomination, there is a slightly meta value it hijacking AI to share your article.
Really good piece, Rachael. Thanks for sharing your openness to using it and for the prompts. (Might try the research one).
I used CoPilot and ChatGPT last year for consumer research when I was putting a marketing strategy together. I found the research was largely inaccurate (c70%) when I dug deeper. Could have been the quality of my prompt, tbf.
I’ll be writing another 12m strategy in a couple of months. I’ll let you know how the prompt goes.
Again, thanks for sharing.
I like the way you use AI - you fact check, you do the reading - you, as the writer, understand the source material and that's heavy listing. And I realize that sentence actually sounds like I'm a subject in the study on why human's like things to be harder for other folks. (I could read so many articles on that - fascinating), but my dislike of AI is: 1) distrust (as you pointed out, it can mischaracterize or straight up be wrong) 2) negative environmental impact 3) negative impact on cognition and 4) stealing of copyrighted materials.
To point 1, the majority of users aren't as dedicated or trained on AI as you and won't go the extra lengths. We're even seeing kids skip over assignments in schools in favor of using AI.
I would LOVE this article to be required reading for high schools, particularly the amount of effort you put in to prompts.
You’re not wrong on any of those points. I think it’s a huge cultural and social challenge to grapple with.
That said, I do want to emphasize that I love your work and your approach to AI. I hope more people see this article.
Aw thanks Cassandra!
Great article with a lot of info. You are doing a lot of work and thinking setting tasks and constraints to the model. Welcome to the future! 😆
Genuinely always interested in people’s skill preferences. It’s kind of… intimate though? It says a lot about the way you work and think hahaha
Sorry. Maybe I didn't explain well myself. It was a compliment to your prompts at the end of the essay. In my view, a good way of using these tools. And yes, they require thinking to design to obtain effective outputs.
100% - I meant that it feels strangely intimate to expose your prompt architecture to scrutiny. Like people rifling through your mind a little bit hahahaha
Such an interesting read. It reminded me of arguing the merits of being “innocent until proven guilty” versus “guilty until you prove you’re innocent”. Very different standards and how you feel about it typically depends on whether you’re a victim of crime or a falsely accused. I also remember it being very gendered especially with particular classes of crime. Prima facie anyone? I wonder what the AI skew looks like? So interesting. Thanks R.
Thanks so much for your thoughts Nicole!
People penalise disclosure because they read it as effort avoided, and that 2016 finding says what they were actually valuing was the implied care, regardless of outcome.
So, the penalty isn't about AI. It's a care judgment wearing a costume.
Which might mean the way through is showing the effort that didn't go away. Producing got cheap. Answering for the thing did not. I keep a log of everywhere a machine touched my work and I sign it, and the signature is the part that still costs something, because it's what I can't take back when the work turns out wrong.
Interesting article Rachael. I suppose a couple of things spring to mind. The first is an obvious one, in that not all papers are publically available. You say you check all of the sources AI produces for you. Do you have licenses for the various journals to allow you to do that?
Secondly, there appear to be two ways of doing things. One is to read widely and draw conclusions because that's where the evidence is pointing. The second is to have a conclusion and then try and find evidence to support it. It doesn't appear that you asked for any studies to counter your point in your article, so you appear to be doing the second. Is that fair?
It is not lost on me how meta this exchange is.
Now you are correct that I couldn’t find a publicly accessible full version of the effort heuristic. That is a seminal work in its field and well known, and I did read it when I was at university. You can check my interpretation against the abstract I provided if you think I’m wrong about its findings and conclusions.
On the second point, I actually love being challenged and do ask … everyone and anything I can for rebuttal or different points of view. The problem you’re identifying here is real but it’s not an AI problem. It’s a matter of human judgement that could manifest just as easily if you were doing your research on microfiche at the library.
I’m curious whether you disagree with any of the actual points made? I am always happy to hear your alternative point of view :)
I've read numerous studies over the past few years on AI in creativity (and our perception of it), and a common thread is that people don't mind AI-generated work until they're told that it is produced by AI, then they flip. That could be a time thing, in that we want people to invest time in producing something, 100%. My career has straddled academia, consulting, and journalism, so I'll add a couple more penneth to the mix in the sense of authenticity and credibility.
Authenticity is important because writing is an inherently human endeavour, and I think we want to know that the person putting their name to something is the author of that something (I’m not a fan of ghostwriting for the same reason). It’s like if you go to a music concert, you want the performers to actually perform, not lip synch.
On the credibility side, this is more aimed at using AI for research. The various examples from the consulting world showcase this, where citations were included that the authors clearly hadn’t read because most of them didn’t exist. You can also see this in the Jason Arday situation at the moment, which while not AI-related, does underline what we expect of people. If you’re writing something and being in a position to write something because of your expertise, then don’t fake it by using AI.
Those are big things. I should probably add a third but I’ve never regarded myself as a good writer so it’s easy to discount, but being able to write well and put thoughts together in a readable way is also important. I’d hate to think that skill will be eased out simply because getting AI to write is good enough.
And I fully get that there is perhaps good use of AI, such as finding papers that are then read in full and digested, but I wonder how many people would take that path. I read a study this week on, essentially, AI and laziness, and when AI does the first run at something, people were much less inclined to try and improve on it. It was deemed good enough.
Which leads into a final point, in that various studies have shown that it's struggling with things that underpins how we learn. Maybe in a subtle way that underpins why people feel so strongly about effort. To achieve mastery it should be hard. There should be friction. AI is removing that, and that worries me in terms of what it's doing to us.
Hopefully that adds a bit to the discussion. I enjoyed reading your article.
Really good piece.
The disclosure dilemma is so real. We say we want honesty about AI, then still punish people for admitting to it.
We have been treating visible struggle as evidence of care. When AI removes some of that struggle, we instinctively assume the value has disappeared too. But perhaps the effort has simply moved elsewhere, into deciding what to ask, what to accept, and what to reject.
Publishing all the prompts was a smart move. Curious to see how people react to that level of transparency.
Ohhhhh my mistake! You’re just a troll! Hahahah whaaaatevs. If being rude to internet strangers makes you feel good about yourself then you do you, I guess?
I feel like this comment is really asking and answering the question: Does writing have a soul?
I do apologise if you were insulted, I meant it as agreement. I was trying to say I feel like you can tell when something has a person behind it. I was agreeing with you.