[06/2023] On getting our hands dirty
Strategies to avoid drowning in the flood while still going with the flow
Welcome back to the issue nr. 6. If you were wondering what was the nature of the datasets scientists are using to train the algorithms we are talking about a “data dump so enormous we could never review or scan it all.” A bit like everything, AI is based on a squishy elusive entity we must learn not to fear but to use creatively, learning about it and teaching about it. We also provide you some experts’ opinions about how to learn to live with it and what needs to be controlled about it.
Then we will be talking about Amazon’s offensive against the flooding of AI-generated books on its channels. We will also see how some Gen Z war-games hackers are trying to protect their right to plagiarise AI-generated content while some others are creating tools that aim to detect content written by humans by content written by machines.
Last but not least, Some encouraging news in one of the hottest fields for AI in these months, healthcare sector, regarding breast cancer screening. We know we need good news there so, read on.
Midjourney/Prompt: "gen Z squishy hacker --s 750"
The world is changing, and Gen Z is coming to the rescue
AI and self-publishing: Amazon to address the issue of AI generated content in Kindle Direct Publishing at last
According to Publishers Weekly, Amazon has begun to address the issue of AI-generated self-published e-books through Kindle Direct Publishing. The company has announced a limit on the number of books that can be self-published to their store in an effort to combat the number of titles created by artificial intelligence. Users can now publish three books per day.
As we can read in the new content guidelines we have now two definitions, AI-generated and AI-assisted content. The definition of AI-generated content includes “text, images, or translations created by an AI-based tool”. If an author uses an AI-based tool to create the actual content (whether text, images, or translations), Amazon considers it "AI-generated," even if substantial edits were applied afterwards. If authors create the content themselves, and use AI-based tools to edit, refine, error-check, or otherwise improve that content (whether text or images), then Amazon considers it "AI-assisted". Similarly, if authors use an AI-based tool to brainstorm and generate ideas, but ultimately create the text or images themselves, this is also considered "AI-assisted" and not “AI-generated.” If authors publish AI-generated books, they need to disclose this to Amazon.
Why these new rules? Because there’s a problem to solve. Already in February of this year Amazon was flooded by AI-written self-published books and in the last weeks it was forced to remove books published under an author’s name, Jane Friedman, that she had never written but were nevertheless bearing her name as the author.
The Guardian interviewed Nicola Solomon, the chief executive of the UK’s leading industry body for writers, the Society of Authors, who said:
“Readers deserve transparency when they purchase a copy of a creative work, so we have regularly highlighted the need for clear labelling of AI-generated content in our discussions with industry.” Solomon said she and her colleagues welcome Amazon’s announcement, as it “will ultimately benefit human authors and their readers”. However, she added: “While this will apply to new and updated KDP publications, we would like to see Amazon ask the same question of all books published on KDP.” She continued: “The past year alone has seen a huge influx of poor-quality, rapidly generated titles in the KDP store alongside human-created works. We also look forward to understanding how readers will be able to filter out AI-generated content from search results when they browse for books.”
AI-detection tools are ramping up, but students are defending their right to copy
In other news but on the same topic of the previous story, the moment has come when we all would like to know if the content that we are reading, or listening to, or looking at, was produced by an AI agent or a human person. We know that ChatGpt can elude a Captcha to prove it’s a human, since it hired a worker from Taskrabbit to solve it in its place. So we will be soon also wondering what, instead of who, performed a certain action.
It’s true that AI-generated art seems to have sometimes an extra finger, and that AI-generated prose is always a bit too formulaic, but until when will we be able to tell the difference? Our days are numbered, we have to embrace what’s coming. And we are ready to witness the emerging of the Gen Z hackers to the rescue, or to shuffle things up.
Like some of their predecessors, they have access to VC capital, they learned coding at the same time as they were learning the alphabet, and while some of them are defending us against a flood of AI-generated text, others are defending their right to copy, or better, to plagiarise the content written by an AI agent.
Read excerpts of the story published by Wired about the creator of GPT Zero and of the creator if its nemesis, WorkNinja. The first mentioned aims to detect ChatGPT generated text, the second aims to make this detection close to impossible. Fun times, 🚫🧢.
GPTZero, the AI-detector
As the world lost its mind over this new, radically improved chatbot, Tian was already familiar with the underlying GPT-3 technology. And as a journalist who’d worked on rooting out disinformation campaigns, he understood the implications of AI-generated content for the industry.
While home in Toronto for winter break, Tian started playing around with a new program: a ChatGPT detector. He posted up at his favorite café, slamming jasmine tea, and stayed up late coding in his bedroom. His idea was simple.
The software would scan a piece of text for two factors: “perplexity,” the randomness of word choice; and “burstiness,” the complexity or variation of sentences. Human writing tends to rate higher than AI writing on both metrics, which allowed Tian to guess how a piece of text had been created. Tian called the tool GPTZero—the “zero” signaled truth, a return to basics—and he put it online the evening of January 2.
He posted a link on Twitter with a brief introduction. The goal was to combat “increasing AI plagiarism,” he wrote. “Are high school teachers going to want students using ChatGPT to write their history essays? Likely not.” Then he went to bed.
WorkNinja, the AI “humanizer”
Semrai pulled out his laptop and ginned up a script that would write an essay based on a prompt, run the text through GPTZero, then keep tweaking the phrasing until the AI was no longer detectable—essentially using GPTZero against itself.
Semrai introduced his program a few days later at Friends and Family Demo Day, a kind of show-and-tell for Stanford’s undergraduate developer community. Standing before a roomful of classmates, he asked the audience for an essay topic—someone suggested “fine dining” in California—and typed it into the prompt box. After a few seconds, the program spat out an eight-paragraph essay, unoriginal but coherent, with works cited. “Not saying I’d ever submit this paper,” Semrai said, to chuckles. “But there you go. I dunno, it saves time.” He named the tool WorkNinja and put it on the app store two months later. With the help of a promotional campaign featuring the Gen Z influencer David Dobrik and a giveaway of 10 Teslas to users who signed up, it received more than 350,000 downloads in the first week; sign-ups have slowed since then to a few hundred a day, according to Semrai. (Semrai wouldn’t say who funded the campaign, only that it was a major Silicon Valley angel investor.)
Breast cancer screening and AI: “AI plus a Radiologist was non inferior to Double Radiologist”
Whomever is eligible for breast cancer preventative screening knows how difficult is to get an appointment due to radiologists’ shortages, yet we all know that early detection is the only way to reduce mortality. The Lancet just published a prospective study where they take into account early cancer detection policies and the objective problems that AI can help solve. The study suggests that the work of two radiologists was non inferior to the work of one radiologist that was also using an AI tool. Read the full article on The Lancet.
Still Trying to Understand Where AI Will Bring Us
Squish Meets Structure: Designing with Language Models
Maggie Appleton gave a speech at the Freiburg SmashingConf 2023 on designing products with language models. In doing so, she explains very clearly and fundamentally how large language models work, their strengths, and weaknesses. She tells us that they are “squishy”: «more like a biological creature we’ve grown, rather than a tool we invented. A little evolutionary: learns by trial and error. Emergent, surprising behaviour we need to study.»
She further adds that «that most of the training data we used to train these models is a huge data dump so enormous we could never review or scan it all. So it’s like we have a huge grotesque monster, and we’re just putting a surface layer of pleasantries on top of it. Like polite chat interfaces.» This leads us to the fundamental contradiction: we are attempting to apply the rigid operational structure of computers to an unpredictable and opaque system. What could possibly go wrong?
AI and Creative Learning: Concerns, Opportunities, and Choices
Mitchel Resnick writes a thorough article on opportunities and choices in the relationship between artificial intelligence and creative learning. I read it, acknowledging that the choice is partly in our hands, as we know that much is beyond our control. Nevertheless, it's worth reading his perspective.
The choice is up to us. The choice is more educational and political than technological. What type of learning and education do we want for our children, our schools, and our society? All of us — as teachers, parents, school administrators, designers, developers, researchers, policy-makers — need to consider our values and visions for learning and education, and make choices that align with our values and visions. It’s up to us.
AI and our future with Yuval Noah Harari and Mustafa Suleyman
The Economist brought together Yuval Noah Harari and Mustafa Suleyman to grapple with the biggest technological revolution of our times. They debate the impact of AI on our immediate future, how the technology can be controlled and whether it could ever have agency.