The idea that machines could become conscious has been widely discussed in artificial intelligence and transhumanist literature, often based on the assumption that sufficient computational complexity might produce phenomenal consciousness. However, this paper argues that such literature lacks scientific rigor because the hypothesis that machines are not conscious cannot be falsified. The authors provide a list of arguments showing that every approach in machine consciousness research relies on philosophical assumptions unprovable by the scientific method. They specifically demonstrate that phenomenal consciousness is not computable, regardless of algorithmic complexity, cannot be objectively measured or quantitatively defined, and is inherently subjective and internal to the observer. The work concludes that the notion of conscious machines is currently a myth of transhumanism and science fiction culture.
Integrated information theory (IIT) proposes a quantitative measure, Φ, to estimate whether a physical system is conscious, its degree of consciousness, and the complexity of its experienced qualia. The theory models a physical system as a probabilistic causal graph of interconnected elements with input-output functions. This paper presents a random search algorithm that optimizes Φ to investigate how graph structure changes with increasing numbers of nodes to achieve higher Φ. The authors also discuss why more complex black-box search methods like Bayesian optimization or metaheuristics face difficulties for this problem and suggest future research directions to improve the search for maximal Φ.
An autonomous agent can benefit from a cognitive architecture inspired by conscious beings, using a global workspace that integrates information from subsystems like attention, memory, and inner feelings. In a large experiment set, agents with this architecture navigated environments with multiple independent magnitudes, adapting to find positions matching their preferences. The model incorporates mechanisms for selecting which magnitude to attend to, storing beliefs and past experiences, and controlling information flow. Results suggest that such a design improves the agent's ability to adapt and perform in complex environments.