 |
|
2026-09-01 Synthetic Photography SIG meeting notes We met online using ZOOM 6:00pm-8:50pm You are invited to join us on ZOOM for the "Synthetic Photography SIG". I expect that this will be a controversial SIG as the topic of AI generated art seems to polarize people into 2 camps, they either love the concept that now they can make beautiful art or they hate the idea that "unskilled" people can make better art than they can. I am sure that we will have different views, but I hope that we can get along and respect everyone's views. I want to expand the focus of the Synthetic Photography SIG to include other image editing tools and techniques while maintaining our main focus on AI tools and techniques. "Freedom of speech is a principle that supports the freedom of an individual or a community to articulate their opinions and ideas without fear of retaliation, censorship, or legal sanction. The right to freedom of expression has been recognised as a human right in the Universal Declaration of Human Rights (UDHR) and international human rights law." - Wikipedia ATTENDEES 1. The Fake Disease That Proves You Can’t Trust All Health Advice - 14:28 - Med Man The researcher left plenty of clues that it was a fake disease. Here's the problem. Ask yourself honestly, if you typed your sore eyes and your fatigue and your brain fog into AI at 11 p.m. A real peer-reviewed journal, "Cureus Journal of Medical Science", the kind where actual academics publish, cited one of the fake papers. Dr Quinn had a patient who had a stroke, recognized that she had a stroke, then looked online and was advised to eat raw garlic to thin her blood and cure the stroke. She waited 3 days before going to the doctor, and starting more appropriate treatment. Her health has been seriously damaged by following health advise from the Internet. Why do intelligent people fall for things that aren't true? When you're looking at health information, AI, influencer, it doesn't matter. If you're being told the medical establishment is hiding a simple cure from you, stop. If the same treatment is being sold for cancer, dementia, arthritis, depression, thyroid disease, Beware no context, no nuance. The things that actually keep you healthy are not exciting: 2. I Got Accused of Selling AI Art at an Art Fair - 9:19 - Krystle Cole Art Artists insulting people who us AI are greatly insulted when accused of using AI themselves. 3. Nobody Will Pay For AI-generated Stuff - 24:13 - House of El: AI In 2026 a humanoid robot can outrun every human being who has ever lived. What does SLOP actually mean? In June 2026 human beings became a minority on the Internet. In January 2026, YouTube permanently terminated 16 channels with a combined 35 million subscribers and 4.7 billion lifetime views. On LinkedIn over a million people have clicked the "Seems like AI slop" button in less than a single month. Miriam Webster named "SLOP" their word of the year. When you say the problem is AI, the solution looks like banning AI or labeling everything AI touches or If you give an experienced senior developer access to an AI coding assistant with unlimited tokens, they produce better code because they have the judgment to direct the tool to catch its mistakes to know what good architecture looks like before the first line is written. The capacity for misuse has never in the entire history of human technology been a sufficient argument for abandoning a tool. It's an argument for building safeguards around it. YouTube's own head of trust and safety, Matt Halprin, said "AI can actually allow people to make a lot of videos. Sometimes those videos are great and it it really enhances creativity and you can create a higher The fact that some people misuse AI to produce industrial quantities of garbage should not mean that everyone who touches AI is producing garbage. An expert in any field using AI to refine an argument, synthesize, research, or make a representative thumbnail is not the same activity as a bot farm uploading seven videos a day to 150 channels. I will show you exactly how much economic damage the slop panic is doing to everyone, including people who have never touched AI in their lives. On CG Trader, which is a long established marketplace for 3D assets used by game developers, film editors, and 3D printing enthusiasts, AI generated models now represent one in every six uploads. What happens when institutions try to draw the line for them? Coca-Cola released an AI generated holiday campaign, got hammered for it, and then did the exact same thing the following year with animated polar bears. McDonald's pulled an AI generated ad after viewer outcry. Dove pledged never to replace real people with AI in advertising. Brands like Airy, Equinox, and Almond Breeze ran explicitly anti-AI campaigns. A 2025 survey by Raptive found that consumer trust drops approximately 50% when content is merely perceived as AI generated regardless of whether it actually is. El is regularly asked whether she's AI because there is AI in the name of the channel, 59% of voters believe that content they encounter online is often shaped by AI. Look at what the AI industry has done with its own messaging. AI generated music and human-made music don't have to compete in the same category. AI used thoughtfully is an extraordinary equalizer. There are people propelling the world forward right now in labs, in trading floors, in engineering departments who are functionally not that able to communicate what they know to the people who need to hear it. The interesting questions here are about degree, intent, judgment, and what the human is actually The AI companies need to stop promising mass displacement as a feature. Cyber security, content authentication, platform design, every element of the digital ecosystem needs to adapt in parallel to the reality that bad actors can and will exploit whatever tools exist. The business model that financed the open web for 30 years was built on one assumption that the traffic was human and human attention could be sold to advertisers. 4. When AI art has no author: MIT study finds generated images often can't be traced to any training data This study is about Diffusion models, not Large Language models yet. When an AI image generator produces a portrait, whose work went into it? New research from MIT's Computer Science and Artificial Intelligence Laboratory The scientists identified a phenomenon they call *** attribution decay***, where the more data a generative model is trained on, the less any individual training example matters to any particular output. At sufficiently large scales, MIT scientists find you can often remove any single image from the training data, or every image by a given artist, or every photograph of a given person, and the generated image doesn't change. If removing something changes nothing, the researchers argue, it can't be said to be responsible for anything. If you take away a piece of data from the training data and the image generated by the model doesn’t change, then that piece of data didn’t affect the generated image," says Zheng Dai SM ‘21, PhD ‘24, former MIT CSAIL researcher and lead author on the work. It doesn't make much sense to attribute the output to that piece of data. Testing this idea directly meant answering a what-if question. What would this model have produced if it had never seen this particular image? Answering it honestly means retraining the model from scratch without that image, then doing it again for the next image, and the next. With millions of training examples, the math quickly becomes prohibitive, which is why prior work in the attribution field has relied on approximations that estimate a training example's influence rather than actually removing it. Their workaround is an architecture they built themselves, called a “diffusion ensemble.” Instead of one monolithic model, it's made up of many smaller components, each trained on a different slice of the data. Want to know what the model would do without a particular image? Just switch off the parts that saw it. No retraining, no approximation. What's left is a true counterfactual model, not an estimate of one. The images generated by the diffusion ensemble models came out looking about as good by standard monolithic models. Gifford sees the finding as bearing directly on the legal question of whether model outputs are derivative works. “One way to think about this is that these models are creative. They are not simply copying what they are fed, but creating brand new outputs. If those outputs have nothing to do with any individual piece of training data, that raises questions about fair use, about whether the outputs are themselves copyrightable as novel works, and about how authors get compensated when what comes out of a model isn't attributable to anything on the internet.” “If attribution (removing data from the training set affects the generated image) worked, it would reliably tell us whether similarities between a model's output and a copyright-protected work are due to copying or coincidence,” says James Grimmelmann, who is a law professor at Cornell Law School and Cornell Tech. “But this paper provides reason to think that attribution will fail for interesting models. Instead, technologists and courts will need to resort to other methods for assessing copying.”
5. Google's AI Called Me A Rapist, Now I'm Suing Them! - 31:00 - RobbyStarbuck
If there is anything related to AI that you would like to discuss at our SIG meeting, please email me so I can add them to our agenda. Please email Mike, info@fcdcc.com, when you find mistakes, missing information or if you have suggestions for the Synthetic Photography SIG and I will try to address the issues. Thanks, --------------------------- Meeting Summary from ZOOM ----------------------- Quick recapThe meeting was an informal discussion group where participants shared various personal updates and discussed topics related to AI, photography, and technology. Mike shared details about attending Tour de Fat and witnessing an accident involving a 75-year-old cyclist, while Jim Limburg discussed his recent moose sighting and mentioned his son's safe location in Kathmandu despite nearby flooding. The group had an extensive discussion about AI's capabilities and limitations, including Mike's presentation of a study from MIT that suggested AI models don't infringe on copyright when trained on large datasets due to the minimal dependence on individual images. They also discussed the challenges of using AI for photography evaluations, with Rich Ernst sharing his experiences using various AI tools like Lensic and Blake Rudis's Division Mentor service. The conversation covered both the benefits and drawbacks of AI in photography, including how some photographers feel AI tools are being used to detect and flag traditional art as AI-generated. The group also discussed the growing prevalence of "slop" content on the internet, with Merriam-Webster naming it their word of the year, and shared experiences about using various AI tools for tasks like identifying birds, photos, and locations.
Next stepsJim Limburg
SummaryBike Activities and Life UpdatesMike and Jim Limburg discussed their recent activities, including Jim's bike ride to Fossil Creek and Mike's experience at Tour de Fat, where they encountered a bicycle accident involving a 75-year-old man. They also talked about driver's license testing requirements for older adults and the changing weather patterns. Mike shared his experience with skunk problems in his neighborhood and how he solved it using solar-powered motion sensor lights.
AI Misinformation Experiment DiscussionThe group discussed recent flooding and disasters in Nepal, where Jim's son is safely located in Kathmandu. Mike presented a fascinating experiment where a researcher created a fake medical condition called "Vixonia" to test AI's ability to spread misinformation, demonstrating how multiple AI systems including ChatGPT and search engines accepted and propagated the fake information with confidence. The discussion concluded with personal anecdotes about using AI for medical advice, with Rich suggesting that in-person physical therapy would be more beneficial than online exercises.
AI Content Challenges DiscussionThe group discussed challenges with AI-generated content, including a medical journal article that contained inaccurate information and a story about an artist whose traditional hand-painted art was incorrectly flagged as AI-generated by Pinterest. Rich Ernst shared his experience with AI photo evaluations, noting that tools like Lensic often gave unhelpful feedback, while he found more value in Blake Rudis's trained AI for photo critiques. The participants agreed that while AI tools can be helpful, they should be used with caution and taken with a grain of salt, as they often provide generic responses rather than nuanced feedback.
PSA Club Guidelines DiscussionThe group discussed PSA club guidelines and judging criteria, particularly focusing on the subjective nature of the "wow factor" evaluation. Rich Ernst and others shared negative experiences with PSA members being rude and monopolizing photography locations at popular sites like Maroon Bells and Yellowstone. The discussion highlighted how PSA's formal judging system, while teaching rules, sometimes lacks helpful feedback for photographers looking to improve their skills.
AI Copyright Concerns StudyMike discussed a recent MIT study that examined how AI models train on images and found that deleting individual images from training datasets doesn't significantly affect generated images, potentially resolving copyright concerns raised by artists. The group discussed how AI models work through tensors and neural networks, with Jim noting that the experimental design could help establish that AI image generation doesn't infringe on copyright. Rich raised questions about who might legally challenge AI copyright issues, suggesting that individual artists lack the resources to pursue legal action and that it would likely take a significant entity to bring a successful challenge.
AI-Generated Images and CopyrightThe group discussed the impact of AI-generated images on stock photo libraries and copyright laws. They noted that while AI images are increasingly accepted by stock photo sites, traditional libraries still offer valuable historical content. The conversation also covered a specific case involving Robbie Starbuck's legal issues with AI-generated content, with Jim Limburg questioning the logic of using the same AI system both to claim defamation and as evidence in defense. Rich Roberts shared his positive experience using Google Images to identify historical photographs, including one that successfully identified a Civil War-era image.
AI Bot Prevalence on InternetMike discussed how AI bots have become prevalent on the internet, with bots generating 57.5% of web page requests according to Cloudflare data from June 2026. He cited YouTube's termination of 16 channels with 35 million subscribers for creating AI-generated content at scale, and reported that 41% of LinkedIn long-form posts were flagged as AI-generated. The group also discussed increasing spam email issues, with participants sharing their experiences of receiving political, marketing, and scam emails, and Rich Roberts noting that spam seems to increase when traveling to different locations.
IP Addresses and Device TrackingThe group discussed IP addresses, confirming that most are shared rather than dedicated and are typically dynamic, assigned when a connection is established. Pam shared a story about using Apple's "Find My" feature to locate her husband's stolen phone, which led to its quick recovery by the police. The conversation also touched on how modern devices, including phones and smart home products, are assigned IP addresses and tracked through Wi-Fi networks.
Generative AI Content EconomicsThe group discussed how generative AI has changed the economics of creating internet content, making the production of low-quality content nearly free while being driven by ad revenue and engagement algorithms. Mike explained that content creators earn money primarily through Google ads and sponsored content rather than direct subscription fees, noting that one creator made $17 million in a year from AI-generated shorts. The discussion also covered how content creators use affiliate links and sponsorships as additional revenue streams, with Rich Roberts mentioning how landscape photographers promote platforms like Squarespace through sponsored content rather than relying solely on Google's ad revenue.
YouTube Monetization and AI DevelopmentThe group discussed YouTube monetization models and subscription services, with Rich Roberts sharing his experience with Canadian car review channels and their subscription strategies. The conversation then shifted to AI development and regulation, where Mike shared insights about AI coding assistants and the importance of experienced developers overseeing AI-generated code. The discussion concluded with concerns about China's potential dominance in AI and broader discussions about education quality in the United States, with participants agreeing that education reform is needed to address both AI development and broader technological competitiveness.
Education and Data Center ChallengesThe group discussed education and data center development, focusing on challenges in public schools and the impact of data centers on communities. They explored issues around charter schools, teacher salaries, and educational opportunities, with Rich Roberts noting that educational reform would likely need to happen at the local level despite tax decisions being made at the national level. The conversation then shifted to data centers, where they discussed concerns about power requirements, water usage, and community impacts, with Jim Limburg explaining how data centers often avoid certain regulations and Mike suggesting that better public explanation of data center needs could address misconceptions.
China's Data Center and AI AdvantagesThe group discussed China's advantage in data center infrastructure and manufacturing, with Jim Limburg noting China's demographic advantage in producing data engineers. Rich Roberts shared his experience with Apple's manufacturing decisions in the 2010s, explaining how cost reduction drove the company to move production to China despite concerns about technology transfer. The conversation then shifted to AI art generation tools, where Rich Ernst described his experience using Artspace to convert photos into paintings, discussing image sizes, credits, and enlargement options available in the platform.
Photography Techniques and Business ChallengesThe group discussed photography and art techniques, with Rich Ernst explaining his process of specifying no texture and then texturizing images in Photoshop. They talked about the challenges artists and photographers face in making money, with Mike noting that teaching classes seems to be one of the few viable paths for new photography businesses. The conversation concluded with Mike sharing details about his upcoming two-week vacation to the northwest, including plans to visit Forks, Washington, and the economic impact of AI-generated content, which he illustrated with examples from CG Trader and the music industry.
Public Attitudes Toward AIThe group discussed public attitudes toward AI, with participants sharing personal experiences and perspectives. Rich Roberts and Mike debated whether resistance to AI stems from early adoption concerns or broader fears about job replacement, with Rich emphasizing the importance of human connection while Mike suggested younger generations might view AI more positively. The conversation touched on practical applications of AI in healthcare and home automation, with Jim Fellion sharing his experience using AI for accessibility needs and others discussing concerns about data privacy and surveillance. The discussion concluded with an informal exchange about watercolor painting classes and creative activities.
|