The key arguments of the article “How Organisms Come to Know the World: Fundamental Limits on Artificial General Intelligence” are as follows:
Limitations of Current AI in Achieving AGI:
- The article argues that Artificial General Intelligence (AGI) cannot be fully achieved within the current algorithmic framework of AI research, which is based on universal Turing machines.
- Algorithms, including those running AI systems and robots, are limited by their predefined ontologies and cannot transcend these boundaries to identify and exploit new affordances in the way organisms can.
Concept of Affordances:
- Affordances are defined as opportunities or impediments in an agent’s environment relevant to achieving its goals.
- Organisms can dynamically identify and leverage new affordances, whereas AI systems cannot because they operate within a deductive framework constrained by predefined variables and properties.
Bio-agency vs. AI Agency:
- Organismic agency (bio-agency) involves self-determination through self-constraint and organizational closure, allowing organisms to initiate actions from within their own boundaries.
- In contrast, AI agents lack true agency as their outputs are entirely determined by inputs without any internal autonomy or goal-setting capabilities.
Inability to Handle Ambiguity and Novelty:
- Current AI systems fail to deal with ambiguity, perspective-taking, and exploiting ambiguities for innovation.
- Human creativity and problem-solving often rely on shifting frames and leveraging novel causal properties of objects, which algorithms cannot do since they require predefined rules and properties.
Implications for Robotics:
- Even embodied AI systems like robots face similar limitations due to the symbol grounding problem and the frame problem.
- Robots cannot generate new opportunities or engage in semiosis (the process of making meaning) as organisms do.
Open-Ended Evolution:
- Truly open-ended evolution, characterized by the continuous emergence of novel possibilities, requires organismic agency.
- Algorithmic systems and simulations lack this capability, constraining them to predefined spaces of possibilities and preventing genuine novelty or radical emergence.
Philosophical and Practical Implications:
- The article suggests that fears about AGI posing existential risks are exaggerated within the current algorithmic paradigm.
- However, it emphasizes the need for protocols and regulations for AI applications due to potential social changes and disenfranchisement of human agency caused by target-specific algorithms.
Future Directions:
- The authors call for a meta-mechanistic science that takes agency seriously and explores teleological explanations rooted in the self-referential closure of organization in living systems.
- Such a science would embrace a naturalistic but non-reductive worldview, acknowledging the radical emergence of goals, actions, and affordances.
These arguments collectively highlight the fundamental differences between biological organisms and algorithmic systems, emphasizing the limitations of the latter in achieving true AGI and engaging in open-ended evolutionary processes.