第七章_AI启示
Chapter 7 Implications for Artificial Intelligence: Why "Pure Software" is Not Enough
If we accept that "consciousness = the structural expression of life's agency," then current AI (large language models), although they have passed the Turing test in symbolic processing, do not possess consciousness because they lack:
- Metabolic structure (an internal mechanism to resist entropy increase)
- Survival pressure (the real risk of dying if not recharged)
- Embodiment (sustained interaction with the physical environment)
The implications of SATC for AI can be compressed into the distinction among three routes:
- Representation and Language Route: Acquiring strong representation capabilities through large-scale data and model structures, excelling in language, reasoning, and generation
- Control and Embodiment Route: Acquiring closed-loop capabilities with the environment through robotics, sensors, and control systems, excelling in action and adaptation
- Self-maintenance and Synthesis Route: Acquiring a sustained closed loop closer to that of life through energy/resource constraints and long-term self-maintenance mechanisms
The claim of SATC is not to deny the first route, but to remind us: even if it is extremely powerful, it does not automatically lead to Level III agency. What agency requires is structured closed-loop conditions, rather than merely stronger representation capabilities.
The path to AGI (Artificial General Intelligence) or artificial consciousness must be the path of "artificial life": empowering systems with structural survival pressure to let "agency" emerge from the process of resisting entropy increase, rather than merely writing more complex algorithms.