AI Slop on Feeds: Expert Henry Ajder on Sora, Deepfakes, and Digital Trust

The Rise of AI-Generated Content

• 💡 The internet is seeing an uptick in AI-generated content, from videos of cats riding motorcycles to images that look slightly off and text that's clearly not human-written.
• 🚀 OpenAI's Sora 2 has significantly changed the landscape of generative video, making it easier to create and circulate AI content.
• 🗺️ The generative AI landscape has expanded dramatically since 2018, with tools now available for text-to-speech voice clones, voice skinning, realistic image generation (e.g., Midjourney, Imagen), and video manipulation.
• ⚠️ A danger exists in underestimating AI content; even content that doesn't appear as "slop" might still be AI-generated.

Realism and Consistency in AI Video

• 🎯 The appeal of AI content isn't solely based on photorealism; consistency in physics and temporal behavior is crucial, even for absurd content.
• 🔊 The ability to generate semantically consistent audio alongside video makes AI content more visceral and engaging.
• 🖼️ Early AI video tools were clunky, often focusing on face swapping, which was initially associated with non-consensual abuse. Modern tools are far more accessible.
• 💬 Creating AI videos now often involves simple text prompting, similar to using ChatGPT, with outputs generated in minutes.

Challenges in AI Representation and Bias

• ⚠️ AI models can struggle with representing diverse individuals accurately, particularly concerning skin tones and accents, leading to concerns about bias.
• ⚖️ If AI models are not trained with representative data, their outputs may not reflect the diversity of the real world, potentially creating a "second-class citizen" category for underrepresented groups in AI avatars.
• 🚀 While Meta's AI video app "Vibes" didn't gain traction, Sora 2's success is attributed to its higher quality, ability to clone known individuals (like Sam Altman), and broader functionality.

Ethical Dilemmas and Consent

• 🕊️ The use of AI to generate depictions of deceased individuals presents significant ethical challenges, termed "techcromancy" when done without consent or respect.
• 📜 While some individuals may specify wishes regarding their likeness after death (e.g., Robin Williams), many have not, making informed consent impossible in numerous cases.
• ⚖️ Companies face pressure to balance freedom of speech with ethical considerations, leading to debates on what content should be permissible and how it should be moderated.

The Attention Economy and Future of AI Content

• 💰 Companies like OpenAI are investing heavily in AI content generation, driven by the attention economy; gaining attention translates to power and currency.
• 📈 Algorithms on platforms like TikTok are designed to maximize engagement, leading to the proliferation of AI content that users interact with.
• 🤔 A key challenge is determining the staying power of AI content versus its novelty, and whether it will lead to backlash or apathy.
• 🌐 The widespread presence of sophisticated synthetic media is leading to "reality apathy", where distinguishing real from fake becomes difficult, potentially allowing biases to influence perception.

Navigating the Synthetic Media Landscape

• 🔍 While tools exist for detecting deepfakes, their reliability is often overestimated, and they can be inaccurate, leading to false confidence.
• 🏷️ Concepts like "digital nutrition labels" aim to provide provenance for media, but adoption is currently low, and metadata support is inconsistent across platforms.
• 🚫 It is unrealistic to expect everyday users to become "digital Sherlocks"; the responsibility for digital trust infrastructure needs to be shared by companies, governments, and civil society.
• 💡 The goal is not to eliminate AI content but to provide users with informed positions to make judgments about the media they consume, moving beyond reliance on gut feelings.