Generative AI carries serious online risks

Significance

These models are fed huge datasets of media content and trained to generate new content following a short thematic prompt. The most hyped -- and feared -- models include OpenAI’s ChatGPT, a conversational agent (or ‘chatbot’) which can generate human-like answers to text prompts, and GPT-4 which extends these capabilities by allowing both text and images to be used as input.

Impacts

• Privacy and safety of AI tools are sensitive policy issues but most democracies will not take Italy’s lead in banning ChatGPT temporarily.
• Smaller extremist groups and low-level cyber criminals are likely to use AI tools for their activities as access barriers are lowered.
• Distrust in mainstream media will increase as AI-generated fake content erodes trust in the entire public information ecosystem.
• Public pressure for AI regulations will rise worldwide; currently, only China and the EU have clear frameworks.