Generative AI

Generative AI refers to artificial intelligence systems designed to create or synthesize new content—text, images, audio, video, or multimodal combinations—that resembles human-created output. These systems learn patterns from training data and generate novel instances that reflect those patterns.

Key generative models

Large Language Models (LLMs): Neural networks trained on massive text corpora (GPT-4, Claude, Llama, Gemini) that generate coherent text sequences. Modern LLMs can perform reasoning, coding, and creative writing indistinguishable from human output.

Diffusion Models: Models trained to iteratively remove noise from random input to generate images (Stable Diffusion, DALL-E 3) or video. Capable of producing photorealistic imagery from text prompts.

Generative Adversarial Networks (GANs): Two competing neural networks—a generator and discriminator—that iteratively improve each other, generating realistic synthetic instances (images, video, audio).

Variational Autoencoders (VAEs): Probabilistic models that encode data into a latent space and decode it back, enabling generation of new instances matching the probability distribution of training data.

Dual-use nature

Generative AI has inherent dual-use characteristics: the same technology that enables legitimate applications (creative expression, accessibility, education) can be weaponized for misinformation, fraud, harassment, and automated propaganda campaigns.

• Detection vs. generation arms race: Models trained for detection are themselves vulnerable to adversarial attacks and model-specific evasion techniques.
• Speed and scale: Generative systems enable content creation at speeds and volumes humans cannot match, particularly relevant to information warfare and industrialized disinformation.
• Authenticity challenge: As synthetic content becomes indistinguishable from authentic, societies face what Ferrara (2026) terms the "Generative AI Paradox"—rational discount of all digital evidence when verification cost exceeds generation cost.

Key papers and articles

• The GenAI Catch-22: Use of Generative Artificial Intelligence in Norwegian — Longitudinal case study of GenAI adoption in four major Norwegian newsrooms during the 2025 election. Identifies the "GenAI Catch-22": newsrooms need human expertise to oversee GenAI systems, but extensive GenAI use leads to de-skilling that erodes that oversight capability, creating structural vulnerability to errors during high-stakes democratic events.
• Industrialized Deception: The Collateral Effects of LLM-Generated Misinformation — Updated perspective on generative AI and misinformation, introducing JudgeGPT and RogueGPT tools for measuring human perception and deceptive potential; documents agentic shift toward autonomous content generation and coordinated campaigns.
• Adapting Fake News Detection to the Era of Large Language Models — Demonstrates that fake news detectors trained on human-written content fail on machine-generated fakes; recommends balanced training data mixing human and AI-generated content.

Related topics

• Synthetic media (broader category encompassing all AI-generated content)
• LLM-Generated Misinformation (specific focus on language model-generated disinformation)
• Deepfakes (video/audio synthetic media and detection)
• Fake news detection (detection methods and countermeasures)
• Content moderation (platform approaches to managing synthetic content)