toplogo
Sign In

Objective Limits of Intelligent Behaviour: Avoiding Computational Dualism


Core Concepts
The core message of this article is that the concept of intelligent software is flawed due to the problem of "computational dualism", where software is treated as separate from the hardware that interprets it. The author proposes an alternative formulation based on enactivism, pancomputationalism, and weak constraint optimization, which allows for objective claims regarding the upper bounds of intelligent behaviour.
Abstract
The article starts by discussing the problem of computational dualism, where the behavior of software is dependent on the hardware that interprets it, undermining claims about the behavior of theorized software superintelligence. The author argues that this problem has a broader significance, echoing Descartes' interactionist substance dualism between mental and physical substances. The author then proposes an alternative formulation based on enactivism, which holds that mind and body are inseparable and embedded in time and place. The author formalizes this by using a pancomputational model of the environment, where everything is a computational system. This allows the author to describe artificial minds in a purely behaviorist manner, focusing on inputs and outputs rather than the mechanism that maps one to the other. The author then formalizes the concepts of abstraction layers, tasks, inference, and learning, using a proxy called "weakness" to estimate the sample efficiency of policies. This allows the author to define the objective upper bound of intelligent behavior, which is attained by using the weakness proxy to maximize the utility of an uninstantiated task across all possible vocabularies. The article concludes by discussing the implications of these results for understanding problems in AI safety and general intelligence.
Stats
"AIXI is the most intelligent policy if it uses the same UTM." [4, p.10] "This undermines all existing optimality properties for AIXI." [4, p.1]
Quotes
"The best model of the world is the world itself." - Rodney Brooks [28]

Key Insights Distilled From

by Michael Timo... at arxiv.org 04-12-2024

https://arxiv.org/pdf/2302.00843.pdf
Computational Dualism and Objective Superintelligence

Deeper Inquiries

How might the proposed formalism be applied to develop more robust and reliable AI systems that avoid the pitfalls of computational dualism?

The proposed formalism offers a shift away from traditional computational dualism by emphasizing enactivism and pancomputationalism. By treating the mind as an inseparable part of the environment and focusing on behavior and causality rather than a strict separation between software and hardware, AI systems can be designed to operate more cohesively with their surroundings. This approach can lead to the development of AI systems that are more adaptive, context-aware, and integrated with their environment. By formalizing tasks, policies, and learning processes within the framework of enactivism and pancomputationalism, AI systems can be designed to learn and generalize more efficiently, avoiding the subjective interpretations and limitations associated with computational dualism.

What are the potential limitations or challenges in implementing the objective upper bound of intelligent behavior in practice, and how might they be addressed?

One potential challenge in implementing the objective upper bound of intelligent behavior is the complexity and computational resources required to search an infinite space of possible vocabularies to maximize utility. This exhaustive search process may be impractical in real-world applications where efficiency and scalability are crucial. To address this challenge, techniques from optimization theory and machine learning, such as meta-learning and reinforcement learning, can be employed to efficiently explore and exploit the space of vocabularies. Additionally, leveraging parallel computing and distributed systems can help speed up the search process and make it more feasible for practical implementation.

Given the emphasis on embodied and enacted cognition, how might this framework inform the design of physical robotic systems that can adapt and learn in complex, dynamic environments?

The framework of embodied and enacted cognition provides valuable insights for designing physical robotic systems that can adapt and learn in complex, dynamic environments. By considering the robot as an integral part of its environment and emphasizing the interaction between the robot and its surroundings, designers can create robots that are more contextually aware and responsive. This approach can lead to the development of robots that learn from their interactions with the environment, adapt their behavior based on feedback, and effectively navigate unpredictable and changing conditions. By formalizing tasks, policies, and learning mechanisms within the framework of embodied cognition, robotic systems can be designed to exhibit more intelligent and autonomous behavior in real-world scenarios.
0