Computational Dualism and Objective Superintelligence

Overview

• The paper argues that the concept of intelligent software is flawed, as the behavior of software depends on the hardware that interprets it.
• It introduces the idea of "computational dualism," where software and hardware are analogous to the mental and physical substances.
• The paper proposes an alternative approach based on pancomputationalism, which defines all aspects of the environment as relations between irreducible states.
• It formalizes cognition as embodied, embedded, extended, and enactive, rather than as a disembodied policy interacting with the environment.
• The paper argues that this allows for objective claims about intelligence, which is defined as the ability to generalize, identify causes, and adapt.
• It then proposes objective upper bounds for intelligent behavior.

Plain English Explanation

The paper suggests that the idea of highly intelligent software, or "software superintelligence," is problematic because the behavior of software ultimately depends on the physical hardware that runs it. This is analogous to the philosophical problem of computational dualism, where software is like the "mind" and hardware is like the "body."

To avoid this issue, the paper proposes an alternative approach based on pancomputationalism, which views the entire environment as a network of interconnected, irreducible states. In this view, cognition is embodied, embedded, extended, and enactive - it's not a separate "policy" interacting with the environment, but rather a fundamental part of the environment itself.

This allows the paper to define intelligence objectively as the ability to generalize, identify causes, and adapt. The paper then tries to establish upper bounds for this kind of intelligent behavior, rather than relying on the problematic concept of "software superintelligence."

Technical Explanation

The paper starts by criticizing the idea of intelligent software, arguing that the behavior of software is fundamentally dependent on the hardware that runs it. This undermines claims about the potential behavior of theoretical "software superintelligence."

The paper introduces the concept of "computational dualism," which is analogous to the philosophical problem of mind-body dualism. Instead of having mental and physical substances, computational dualism posits software and hardware as the two core components.

To avoid the issues with computational dualism, the paper proposes an alternative approach based on pancomputationalism. This view defines all aspects of the environment as nothing more than relations between otherwise irreducible states. The paper formalizes systems as "behavior" (inputs and outputs, with policy as a causal intermediary) and cognition as "embodied, embedded, extended and enactive."

This allows the paper to define intelligence objectively as the ability to "generalize," identify causes, and adapt. The authors then attempt to propose objective upper bounds for intelligent behavior, rather than relying on the problematic concept of "software superintelligence."

Critical Analysis

The paper raises important concerns about the limitations of the concept of intelligent software and the issues with computational dualism. By proposing an alternative framework based on pancomputationalism, the authors attempt to ground the discussion of intelligence in a more grounded, embodied view of cognition.

However, the paper's proposed definitions and formal models may be challenging to fully evaluate without a deeper understanding of the underlying philosophical and computational concepts. Additionally, the paper does not provide extensive empirical evidence or case studies to support its claims about objective measures of intelligence.

Further research and debate would be needed to thoroughly assess the validity and practical implications of the paper's arguments. Readers should think critically about the strengths and weaknesses of the proposed approach, and consider how it might complement or challenge other perspectives on artificial intelligence and consciousness.

Conclusion

This paper challenges the traditional notion of intelligent software by highlighting the fundamental dependence of software behavior on the underlying hardware. It introduces the concept of computational dualism and proposes an alternative framework based on pancomputationalism, which defines intelligence as the ability to generalize, identify causes, and adapt.

The paper's core contribution is its attempt to ground the discussion of intelligence in a more holistic, embodied view of cognition. While the technical details may be complex, the central idea of avoiding computational dualism and defining intelligence objectively is an important consideration for the field of AI and cognitive science.