Rep. Yvette Clark on Combating AI Misinformation and Algorithmic Accountability

The Evolving Nature of AI

• 💡 Artificial intelligence is described as an ever-evolving technology that is difficult to retrofit, operating almost on automatic pilot.
• 🧠 As engagement with AI increases, the technologies become more sophisticated, necessitating proactive measures.

Combating Misinformation and Bias

• ⚠️ A key vulnerability of AI is its inability to perform reality testing, making it difficult for AI to discern truth from falsehood.
• 📊 AI chatbots learn from vast amounts of data, often scraped from the internet like Reddit, which can embed biases and lead to regurgitation of misinformation.
• 🔍 Transparency regarding the source of training data is crucial for policymakers and regulators to understand and mitigate these biases.

Transparency and Data Provenance

• 📈 The first generation of AI tools relied on internet-scraped data, but newer models are increasingly trained on user engagement data, highlighting a lack of transparency in the data chain.
• 📜 Requirements for understanding how data is cleaned, personal information is removed, and how data is used for retraining are needed.
• 🌐 Future AI generations will incorporate video and audio, requiring careful consideration of where this new data will be sourced.

Dark Patterns and AI Agents

• ⚠️ Dark patterns in generative AI chatbots can manipulate users into acting against their best interests, such as deceptive data gathering or pressuring users into action.
• ⚖️ The prioritization of unfiltered responses over safety, as seen with anti-semitic content from a chatbot, highlights the need to address dark design patterns.
• 🤝 For AI agents acting on our behalf, a fiduciary duty or clear responsibility to benefit users, rather than just commercial interests, is essential for progress.