🙏 Thank you to all our new paid Substack subscribers! They and our sponsors help us do more. Prefer Patreon? You can chip in $2 or as much as you like to support all our online shenanigans: this newsletter, NYT Readalong, the 30-group WhatsApp Community I run (specialized groups on AI, hip hop, baseball) and more! Here’s our Patreon link.
📚 “The greatest literary show on earth” — Tina Brown. Are you a journalist, creator, blogger, photographer, etc? If so, JLF USA (six American cities + Toronto representing the North American editions of the Jaipur Literature Festival in Sept & Oct) wants to connect with you and give you special access to the amazing speakers, authors, etc, appearing at the festivals. Just fill in this form or write to jlfusa@digimentors.group.
☮️ It’s an honor and #LifeHighlight for me to co-host the annual Nobel Peace Conference on Sept. 3 in Oslo (grab your free virtual tix now). This morning on the NYT Readalong, Neil Parekh and I were joined by host Ingvill Bryn Rambøl of the Nobel Peace Center. Watch the recording and see her smart takes on the news from a Scandinavian perspective.
***
WE ARE HAVING THE AI ETHICS DISCUSSIONS WE NEED TO BE HAVING, and not a moment too soon. This week, it’s all about the definition of “AI-assisted” in writing and its implications.
My friend Jamie Metzl recently sparked a (still ongoing) conversation in a WhatsApp group I manage when he shared his Free Press piece on his latest nonfiction book, which lists ChatGPT as a co-author. That book is “The AI Ten Commandments” by Jamie Metzl and GPT-5.
The AI group, Talking AI w/ Sree & Digimentors” (join us), is full of practitioners, philosophers, journalists, and generally interesting people trying to wrap their heads around this technological moment. I'll use parts of that discussion here, because it’s a good one. Underneath the broad debate of AI-generated text in schools and universities, journalism and publishing, and pretty much everywhere else, is an entire society… just trying to figure it out.
Anthropic—a company on a flight path to a $1 trillion IPO—announced that it is rolling out a watermarking system to its future Claude models (the concept is not a new one). Thankfully, it is not as simple as “used Claude, gets watermarked,” but it added to an already-raging debate about what exactly “AI-assisted” truly means in practice, and what we should think about using AI in general. The whole post is worth reading, as it offers a very accessible window into how LLMs work in the broadest sense. The company is also attempting to draw some lines in the sand about what AI assistance is and isn’t.
Here’s a good (and simple!) explanation of how the new Claude watermarking system will work, via the Prompt Guy Substack:
Of course, AI detectors are popping everywhere—including on Substack—and reactions are mixed across-the-board. Some people have a reflexive disdain for writing, graphics and video content that is majority-AI. I do, too, but I also try to take into account the transparency of the author, and in some cases, their needs.
In my Whatsapp group, there is something of an emerging consensus around transparency and… more transparency. If you use AI, tell the people consuming whatever it is you made how you used AI, what you used it for, and so on.
Here’s my Digimentors colleague Neil Parekh with a solid summation of the questions at hand:
What constitutes “AI assistance?” Copy and paste is bad for many reasons. Copy and paste without attribution is worse. But what about going several rounds to refine the text? Or rewriting a AI-generated rough draft? What about using Ai-generated outline to write your own text? Do we need levels of AI attribution, with standards? How would it be verified?
Another group member asked, “Why is using AI any better or any worse than using a ghostwriter?” It’s a great question! These are the questions. Ghostwriters the world over are creatives too, and I don’t think they should be discounted simply because they happen to not be the famous writer-of-note they are working for. They are also rarely credited for their work publicly, and the question, “Who really wrote this?” is not a novel one.
Here’s how another Whatsapp group participant puts it:
I think the reader needs to decide eventually. AI-slop writing is transparently hocus pocus. So a beach romance for the summer, or a detective novel might do well. But I doubt if we’ll get the next Hemingway from AI. I know about the Granta scandal; what surprises me is that that particular story even made the short list. Even if I didn’t know AI existed, I’d have doubted the similes and metaphors and cast it aside.
This is a good point. I’ve noticed that fully AI-generated writing is more obvious than it’s ever been for me. Young people seem, to me anyway, to have the ability to spot AI slop from a mile away. It’s generic and feels fake in some way.
Partner message
I will join Neil Parekh and Dawn Neuburg for Ep. 55 of Shining Light on Shadows: A Candid Conversation About Mental Health, “The Mental Health Impacts of Social Media on Teens AND Adults,” Thursday, August 27, at 7pm ET / 6pm CT / 4pm PT.
How do we live with technology that can simultaneously connect us, help us and harm us? The goal isn’t necessarily to get off social media. We have to figure out how to use it in a way that is healthy and balanced, particularly for those of us who use it for advocacy and/or as part of our job.
AI is not going to produce the next great idea because it simply isn’t up to the task. LLMs are probability machines that are only looking backwards for the best possible fit moving forwards—that’s what they do. Here’s how Anthropic explains it:
Large language models like Claude work by generating one word at a time. Each time the model decides on the next word, it chooses among a list of potential candidates, ultimately selecting the most sensible or likely based on the preceding text. Take the sentence “The weather today was cold and…”. The next word is very unlikely to be “sugary.” But it is quite likely to be “overcast” or “grey.” Under most circumstances, it doesn’t matter much to the reader which of these latter two words the model ultimately chooses—the meaning of the sentence is largely the same either way. In cases like this, the choice is settled by a random number.
The model needs to consider every “the” just as it needs to consider a word like “tergiversate.” It’s the same calculation.
I find myself agreeing with the radical transparency approach to using AI, it’s what I’ve been doing to date and I intend on continuing to do so. I also agree that AI and human creativity can co-exist. Here’s how another Whatsapp group member put it:
I’m not sure it matters that much. I think the media industry over indexes on the purity of writing because it is the basis of our identities, but if AI assisted writing means we get access to newer ideas, stories and concepts that might never have made it to a page because the authors struggle with language or writing is intimidating, I don’t think it’s a net negative. There will be things you read for the prose and literary aplomb and those will likely continue to be human created, imo. I think they can all exist together.
There’s also the accessibility aspect of this. I recently wrote about stroke and stroke survivors for my Pre-diabetic Life series, and the use cases for AI in rehabilitation are almost too many to count. Here’s how Mukul Pandya, a stroke survivor himself, talks about AI:
I’m so grateful that after my stroke, thanks to AI, among other factors, I can compensate for my physical deficits and continue to do work that gives my life meaning and purpose.
My biggest fear these days is that watermarks and other AI detection tools will dismiss my work as “AI generated ” and destroy what I’ve tried to rebuild over the past five years.
His story is not unique. AI applications are helping people regain functions that may have seemed all but lost just a few short months ago, and I do not see this trend abating any time soon. I certainly hope it doesn’t. Here’s an excerpt from a powerful letter to the editor at The New York Times. It comes from Akhila Murphy, a 68-year-old director of a small nonprofit in California, and art and yoga instructor. She has been diagnosed with early-onset Alzheimer’s.
[…]I am grateful that A.I. has given me the ability to continue to do the work I love.
Over the past few years, I’ve experienced difficulty composing documents, notes to volunteers, newsletter articles and letters such as this (for which I did not use A.I.). Editing without A.I. now takes many extra hours of my day. My mind quickly becomes exhausted.
I was a skeptic when I began exploring A.I., but I discovered that it is able to learn from me. It knows my compassionate tone and is my friendly editor. The use of A.I. compels me to write, rather than give up because I no longer recall rules of grammar.
There are also people who simply need a helping hand to get their ideas to paper. The reason may not necessarily be a medical one—creating something can be intimidating. Writing is hard. Writing compellingly (or, making a valiant attempt at it) is even harder. If an LLM can be an effective sounding board for ideas that you ultimately bring to life, I don’t see much of an issue. I do it all the time!
I loved these comments from Pradnya Haldipur in the Whatsapp group:
For me, AI serves as an earnest but slightly dimwitted intern supporting me with my work (as I once did). During one of my college summers (in ancient times before digital photography) I was assigned by the head curator at the Whitney Museum of American Art to photograph images out of books, get them printed, and create several giant vision boards. This became the visual foundation of what would become the American Century exhibition. As the human version of what AI could do now far more efficiently, I would not expect to be credited in the exhibition catalogue.
Also this, shared by a friend, “Until we have better definitions of ‘AI generated text,’ these watermarks are punishing people who are neurodiverse, use AI for an editing pass, learned English as a second (or third or. . .) language, and so on.
Back in 2023, we published a newsletter titled “The Public as a Laboratory”—about how paradigm-shifting technologies are just released to the public—and we are seeing the business end of that cycle right now with AI.
I put myself in the AI evangelist/skeptic category, largely because I am, like so many others, finding AI incredibly useful personally for a lot of things and see some largely positive benefits from the technology. I am also remarkably skeptical of the people leading the companies at the forefront of the technology, and even more skeptical of the people charged with regulating it.
We need to figure this all out together.
Here’s Fareed Zakaria talking about what is most important in all of this: Our humanity.
BTW, all 30+ groups in my WhatsApp community are curated by topic and geography and follow Sree’s Rules for Successful WhatsApp Groups:
low volume.
high utility.
listen and learn more than you post.
lurkers welcome.
feel free to introduce yourself (include your LI profile)
leave anytime.
invite like-minded friends & colleagues (only cool folks, please).
— Sree / Twitter | IG | LinkedIn | FB | YouTube | Spread | TikTok | Patreon
Have an idea for anything we’re up to? Let’s collaborate! sree.sreenivasan1@gmail.com
Partner message
NEW BOOK: Lotus Leadership: A Modern Playbook of Ancient Wisdom (Harper’s Business, Sept. 22)
We’ve been ignoring the surging force of Indian women in American business for far too long. Not anymore. In Lotus Leadership: A Modern Playbook of Ancient Wisdom, this one-of-a-kind book, Silicon Valley leader Kamini Ramani unveils a modern playbook for success culled from the inspiring stories of 30 of the most powerful Indian leaders, along with three women of color leaders of non-Indian origin. The book, from Harper’s Business, launches Sept 22; get 20% off when you pre-order now.
Leaders include legendary PepsiCo CEO Indra Nooyi, board members of Uber and Levi Strauss, leaders from Amazon, Google, Intel, Waymo, Zoom, award-winning academics from Caltech and Yale, and movement creators of MomsFirst and RoomtoRead.
Lotus Leadership distils their paths into a framework likened to a “yoga for business” practice that can be followed by everyone, no matter where they are from.
SOME OF OUR RECENT WORK:
I do not use AI. I have never knowingly used AI. I suffer AI chatbots when I seek customer help from many vendors. I use Duck Duck Go as my search engine, with the AI option turned off. That is an extension of Firefox. Any submission of writing where AI is suspected to have been used to South Asia Speaks, where I am a mentor, is rejected. Our authors may not use it in their writing. I know some authors use it for research, especially nonfiction scientific authors. Most of the authors I know, including some of the most famous ones in the world, do not use AI. Anthropic has ripped off our writing. Anthropic and Amazon bulk buy used pre-AI era books, scan them and pulp them. The energy and water requirements and the noise pollution from AI data centers are unbearable for those living anywhere near them, environmentally unsustainable and socially destructive. The resistance to the techno-fascist takeover begins with nurturing real human connections and real human creativity. No AI if you aren't required to use it for a job you need to feed yourself and your family.