Humane-AI Asia

EU AI Act, "AI Systems" and ChatGPT Custom Conversations

Index
    EU AI Act, "AI Systems" and ChatGPT Custom Conversations

    "AI System" A Comparative Analysis
    The upcoming EU AI Act and its definition of "AI System" may be based on the recently adopted OECD definition. This will provide much-needed consistency in AI law across different countries.

    Although the final text of the Act is not yet available, we have been able to analyze the leaked definition, comparing it with earlier drafts and the OECD version. We observed a shift from focusing on specific techniques to the nature of the system itself and from "human-defined" goals to "implicit" goals to account for AI's learning capabilities. Additionally, the scope of the Act is very broad, applying to both providers and users within and outside the EU.

    Planning for compliance with the AI Act is encouraged even before it is finalized. Understanding the potential scope and definitions can help organizations prepare for future requirements.

    President Biden's Executive Order on AI
    The EO's definition, like the OECD definition, uses the term "machine-based system." However, unlike the OECD and EAIA, it emphasizes "human-defined goals" for AI.

    The EO definition highlights:

    • AI predicts, suggests, or decides based on these goals.
    • It uses inputs from both machines and humans.
    • It perceives, abstracts, and reasons with models to generate choices.

    The EO definition:

    • Does not explore how AI interprets inputs for outputs.
    • Does not include concepts of autonomy or adaptability.
    • Focuses on processing and generating outputs.

    Conclusion
    The OECD definition is considered the most expansive, but these three definitions are similar enough to initiate a catalog of AI Systems regardless of which definition you apply. Including the capabilities of AI Systems to infer goals from inputs, as well as varying levels of autonomy and adaptability, is indeed important. As emphasized above, this helps ensure that control measures are implemented proportionately to the risks posed by a given system.

    ChatGPT and What Helped Me This Week
    ChatGPT's "Custom Instructions" feature allows you to set personalized context for interactions, saving you time and effort. This feature, akin to a teacher's guidance or an effective coding option, enables you to define parameters such as:

    • Target Audience: ChatGPT can tailor responses based on whom you are talking to (e.g., a third-grade student in a science class).
    • Personal Preferences: Specify language choices, family size for a grocery list, or tone (formal, casual, etc.).
    • Knowledge Level: Clarify your expertise to avoid overly basic or complex explanations.
    • Opinion Filtering: Choose whether you want ChatGPT to provide its own perspective or stick to factual information.

    Benefits:

    • Efficiency: Avoid repetitive contextual explanations.
    • Personalization: Tailor responses to your needs and preferences.
    • Flexibility: Define different parameters for various use cases.

    Once set up, this feature allows you to interact with ChatGPT more naturally and effectively, making it a valuable tool for various tasks such as lesson planning, grocery shopping, and even scheduling workouts.