Maybe in some hypothetical future where the AI actually understands the text and can create an intelligent analysis. Until then, it’s just going to spew out probability-generated text which may likely be plagiarized and could easily be flat-out wrong.
I imagine the folks at Cliff’s Notes are getting their lawyers ready to sue the operators of any AI that supplies their content without a license.
Absolutely, but the conversation has been about a situation where the ultimate product is something written. In that situation, writing is thinking.
The issue is that a teacher doesn’t just walk into a classroom and say “okay, here’s critical thinking” and have the students pick it up, you have to do actual things that teach them to think critically. Those activities are almost always some form of knowledge acquisition, analysis of that knowledge, and communication of the analysis. Or to put it in the form of an actual exercise: I’ll have my students read the Bill of Rights, and analyze it through discussion and writing. They’ll read each other’s ideas and react to them, and write their reaction. The reading and writing is where most of the insight comes, not in some abstract period of thinking.
Actually, I think this question about identifying incentives is the wrong question. To me, the absence of desire to communicate at least intelligibly, if not skillfully, is one regrettable result of the dumbing down of the population imho; and this problem (if it exists, which I most certainly believe it does) isn’t just about the educational system or social media—it’s a symptom of the deterioration of our whole society.
No, it’s precisely the question I meant to ask, inspired by your suggestion that incentives play a role. If you don’t want to answer it, that’s cool. 
I didn’t ask a question about incentives. I replied to someone else’s question about what incentive might encourage people to develop better communication skills. But when I thought about this question further, I realized that—in my opinion—it is the wrong question.
I was highly intrigued by this post, where a guy fed the briefs submitted to the Supreme Court this year and fed them to Claude 3 Opus. He found that Claude consistently decided the cases “correctly,” meaning that it either agreed with the Supreme Court decision or agreed with the dissenting justices in reasonable ways. This is an interesting testing approach because it compares the results against the results generated by humans. (And folks, please don’t take this in a political direction that will force me to delete posts—you have better things to do with your time than write posts that I will delete, and I have better things to do with my time than delete them.)
Fair point. Perhaps a more generalized statement of what I’m trying to tease out is that the need is to organize one’s thoughts, and there are multiple ways of doing that. Once one’s thoughts have been organized, they need to be communicated, and writing and speaking are the predominant forms of that.
Is there a role for these AI-driven writing tools in that? I’d argue that there could be, particularly if they’re used along the lines of a Socratic dialog, where the person keeps testing ideas and seeing how they work and how they line up with one another.
Totally agreed. My junior year at Cornell was my most formative. I took Greek Composition with Matt Neuburg, where one other student and I translated 25 sentences from English into Ancient Greek each week, and then defended our choices verbally in class. (Greek and English are rather different languages, so you had to determine the meaning of the English sentence before translating the words.) Then I took Aristotle with Gail Fine, who is one of the preeminent classical philosophers. She was very nice, but you learned early in the class not to open your mouth unless you could back up what you were going to say. And I took Carl Sagan’s Seminar in Critical Thinking, which was primarily a weekly discussion with an entrance-by-application class he picked every other year. It was absolutely brilliant, and the capstone assignment was one where everyone picked a topic they were passionate about—abortion, the death penalty, world hunger, etc. Then he paired us up with others in the class who felt equally passionate about the other side of the issue. Our challenge was to make the best possible argument for the side we hated, and we could use any resources, including the other person. All great stuff, but relatively little of it involved actual writing. (There were papers in the Aristotle and Critical Thinking classes, to be sure, but I don’t remember them at all. I remember the discussions being where the ideas emerged.)
To tie this into the incentives discussion, was any of this easy? No. Did I expect to write Ancient Greek or become a philosopher? No. But in each case, I was motivated by the desire to get things “right” (harder in the humanities than in STEM fields), to please the intellectual giants teaching me, and to get a good grade.
All that said, and I realize that I’m flip-flopping a little here, I don’t think AI tools would have done much for my work at this time. I remember the classes being where ideas and arguments took form, and to this day, I like to talk new ideas out with Tonya before I write. I haven’t really tried doing that with a chatbot, but it might be an interesting exercise.
I agree that the article and the interactions it quotes are very impressive. But since I’m not a legal scholar (and I’d assume that the person writing the article isn’t a computer scientist), it’s hard to draw real conclusions from it. In particular:
But it is an impressive demonstration, nonetheless.
Having had to read a lengthy sample of legalese recently, I am not particularly impressed. Legalese is a distinct language in itself, intended to be clear to those who understand it, without much consideration for the general public. This is not unique to the law. Scientists, engineers, physicians and other professions have their own specialized versions of English. These specialized languages evolve because they let specialists communicate effectively within their own community. I have written for laser industry publications for many years, and I understand the special “laser geek speak” it uses.
Although these specialized languages are useful within their specialties, they don’t work well outside it. Most of us don’t understand legalize, medical jargon, the language of quantum physics or the more arcane rules of baseball.
Large language models rely on training the AI on as complete a sample of the language as possible. Limiting the AI to a specialized language lets it master the specialized language well. But that does not train it write a different specialized language or generalized English, which would require training on vast amounts of material. One size in AI does not fit all needs.
Having seen the hullabaloo over Adobe’s recent change to terms and conditions it’s fair to say many weren’t impressed (although they were understood).
One wonders how possible it will be for AI to convert typical licensing legalese into ‘layperson’ speak - potentially a far more useful purpose than getting it to write the legalese (or any other overly specialised language model).
My 2¢: While I agree with Adam’s initial reply (and I remain very much in favor of the incorporation of AI Tools in Apple products)…I also agree with the notion that it will likely…inevitably…degrade people’s overall writing competence. The question is: Is that okay?
I am reminded back (decades ago) when I taught Statistics…and was confronted with the decline in students’ math skills related to the use of hand-held calculators. As one extreme example, I had a conversation (late in the semester) with an A student where it became clear that she had no idea what a square root was. I was stunned. I asked he how she was doing so well despite this lack of basic understanding. She said that she simply pressed the square root button on the calculator when required…and proceeded with the rest of the problem…even though she didn’t have any idea what the button actually did.
Maybe you could attribute this to a failure in my instruction (although I clearly went over it in class…and these were college juniors who should have learned this before anyway).
But I viewed it as an example of how the availability of calculators was allowing (perhaps even encouraging) students to forego competence in basic information. Still, the student was successfully navigating my class.
Other examples abound. I once requested a 1/3 of a pound of whole beans at a coffee store. It took three workers to confer to figure out how to convert the one pound price to a 1/3 pound price (because their cash register system did not have a button to do it automatically). On another occasion, a cashier at a fast food restaurant had to seek out the calculator in her purse to figure out how to deduct a mistaken charge of 50¢ on my bill (for some reason, the cash register was not up to the task). She claimed she wouldn’t know how to subtract the amount without it.
I think there could be an intriguing “philosophical debate” here: At what point does a technology become so useful and so ubiquitous that we should accept it as a replacement for previous “manual” processes? For example, returning to the world of statistics, there was a time when — to calculate an Analysis of Variance — you needed to recall (or look up) several formulas and do a series of calculations in your head (or with a basic calculator). Now, you just enter the raw data and press one button — and the entire results appear on a screen. Is that a bad thing? Should we still require that students learn how to do an ANOVA without the benefit of any digital tools? Or should we accept that the world has moved on…and that (save for a few people who develop tests like ANOVA) we no longer need to know how the work is done? Should we even teach it in class anymore? If we can correctly interpret the results, isn’t that sufficient in almost all cases?
And so it is with AI writing tools. At one point do we say that it no longer matters whether a person is capable of writing well without help from AI? If they use AI and get the desired results, shouldn’t that be good enough?
Personally, I’m not sure. If this were a debate, I believe I could make a good case for either side.
No, it’s not in the hands of somebody who knows how to do it by hand and understands the process. For this person that frees up their time to do something that cannot be done by the press of one button. OTOH in the hands of somebody who lacks proficiency and doesn’t know what that button press does, I would argue it is a bad thing. Rather than pressing that button, they should first learn why they’re pressing that button and what it is that button does. That would be true growth, unlike just posing as somebody who has a clue.
I see no added value from rewarding people for being ignorant, unskilled, or just flat out lazy. People should strive to overcome their weaknesses. And yes, that takes effort and willpower — but I’d claim it’s as true as always: no pain, no gain. If those who aren’t skilled can’t be bothered to grow, why should I take any interest in their inertia? Why should I cheer on new tools that help them mask their incompetence?
I agree. But, as I believe you are saying, you can teach the conceptual underpinning without requiring that they know how to do the actual calculations.
It is quite possible that in the non-jargon realm, AI will soon produce better writing than many humans do. When browsing the fiction section of my local library and pulling random books off the shelf, one of the first things I look at is the length and structure of sentences and paragraphs. Too many short sentences with “subject - verb - object” form and it goes back on the shelf. Lazy writing, even by humans, is just not an enjoyable read.
I think AI writing is just at the beginning. As it is trained on better sources, will it improve enough to win a Booker or Pulitzer Prize? Even contemplating that possibility makes me shudder. Will it eventually founder on the shoals of plot construction and character development? I sincerely hope so.
I’m not sure that’s true. Rather we might say that we’re not going to require people to actually understand things that they produce. Where the line is on that, I don’t know, but I also hear people talking about teaching “critical thinking” without any sense how the two might be related.
Shorter version: in the future, when AI hallucinates, will anyone know it’s doing so?
To know what something is, you don’t necessarily need to know an efficient method to determine it. You can understand that the nth root of a number is the number that you multiply by itself n times (I.e., raise it to the nth power) to get the original number. You may have been taught a nice algorithm for computing square roots and have a general understanding of more complex methods that converge quickly. But do you remember that algorithm? If you don’t have a calculator or table, you would probably try to make a reasonable guess and refine it until you get something close enough.
So, it is important to know what a term or idea represents. Being able to derive it efficiently is secondary if you can access a reliable source.
In this regard, I think the most interesting question is: Will it have anything to say? I have a very hard time believing that AI will ever have a soul, or truly care about anything. A mash-up of the work of famous writers is not “creative writing”. We live in an age when almost anyone can call themselves an “artist”, but this is merely a social aberration, not an explosion of true creativity.
The equation of intelligence with the ability to speak English is … troubling.
I have not read anything in this discussion that would suggest this “equation”. However, many (if not most) of us are American-born English speakers; so it is only natural for this discussion about using language to focus on English. Every discussion doesn’t have to be all-inclusive.
I don’t think anybody mentioned ‘efficient’. If somebody properly understands what the nth root is, they can calculate it, if only in an iterative manner. Or at least get a pretty good approximation. And that’s perfectly adequate.
And also vastly superior to (not to say worlds apart from) somebody who just blindly taps a button according to a set of instructions delivered to them. That person doesn’t need AI or a calculator, that person needs a math tutor.
Once they have understood what the nth root is and how to approximate/calculate it, then we can talk about calculators and code and efficient numerical recipes.
AI doesn’t do anything on its own. Restating your question: Could a person use AI-driven writing tools, perhaps including chunks of text that were generated entirely by the AI, to write something that would be of high enough quality that it would win a Booker or Pulitzer prize?
To that, I see no problem in answering yes, but I wouldn’t be in the slightest bit perturbed by it. Nearly every modern book is affected by a wide variety of human and automated inputs, ranging from spell check to dedicated editors. AI writing tools just provide another input, but only under the direction of a writer or editor, who then retains control over whether to accept that input.
Personally, I don’t think it even makes sense to ask the question. We’re talking about software that outputs results when operated by people. It’s like asking how Photoshop feels when you fill an area with color.
It seems to me that this is merely another in a long series of questions about whether assistive technologies are problematic. Is relying on GPS for directions problematic because people can’t read maps? Is relying on maps problematic because people can’t navigate using the stars? Everyone has some belief about what knowledge or skill is necessary, probably based on what they learned as a teenager (which was the best decade ever).
So yeah, I think I’m on the side of agreeing that if the results are as desired, how you get there may not be that important in most cases. That’s not to say there isn’t a great deal of value in understanding everything en route, but not everyone has to do so to lead a happy, productive life. And while it’s hard to excuse not being able to do basic math, it is also true that the world continually becomes far more complex than in the past, which in essence forces everyone to specialize more and ignore more outside their specializations.