The Power and the Peril
The same technology that lets a small business owner generate professional marketing imagery also enables the creation of non-consensual intimate imagery and political disinformation. Generative AI is not inherently good or bad — but the concentration of creative power without commensurate accountability structures creates real risks that demand our attention.
Deepfakes and Synthetic Media
Deepfakes — AI-generated videos that replace a person's face or voice with another's — have advanced from obvious artifacts to near-indistinguishable forgeries. The harms are real: revenge porn, political manipulation, corporate fraud via CEO voice cloning, and evidence fabrication.
Detection tools lag behind generation: current detectors achieve 70–85% accuracy at best, and sophisticated generators specifically optimise to fool detectors. Technical solutions alone cannot solve this problem.
Copyright and Intellectual Property
Training diffusion models on copyrighted artwork without consent has triggered class-action lawsuits against Stability AI, Midjourney, and DeviantArt. The legal questions are genuinely novel:
- Does training on copyrighted works constitute infringement?
- Who owns the copyright to AI-generated images — the user, the model developer, or no one?
- Is an AI image that closely resembles a specific artist's style a derivative work?
US courts have so far held that pure AI-generated images without human creative input are not copyrightable. The EU AI Act and various national laws are introducing disclosure requirements for AI-generated content.
Misinformation at Scale
LLMs can generate thousands of convincing news articles, forum posts, and social media comments per minute. The cost of disinformation campaigns has collapsed. Research from Stanford shows AI-generated misinformation is rated as equally credible as human-written misinformation by most readers.
The misinformation problem is not primarily technical. It is an epistemological challenge: how do we build societies that can reason about truth when the cost of creating falsehood is near zero?
What Should We Do?
- Content authentication: C2PA (Coalition for Content Provenance and Authenticity) standards embed cryptographic provenance data in images. Major platforms and camera makers are adopting it.
- Consent and opt-out: Artists should be able to opt out of training datasets. Some platforms are moving in this direction.
- Disclosure requirements: AI-generated content in political advertising should be labelled — several US states have already legislated this.
- Platform responsibility: Platforms distributing AI-generated content should implement detection and labelling at scale.
Other Articles
