Rethinking AI's Capacity to Surpass Human Thought

By Soham NandiReviewed by Susha Cheriyedath, M.Sc.Jan 4 2024

In an article published in the journal PNAS Nexus, the authors questioned the idea of artificial intelligence (AI) surpassing human thought. They argued that the prevailing definition of intelligence in AI, focusing on accomplishing complex goals, was inadequate for capturing the essence of human thought. The discussion emphasized the significance of the body, brain lateralization, and glia cells, highlighting the limitations of AI in mimicking conscious functions and underscoring the richer nature of consciousness compared to mathematics and computations.

Background

Integrating sophisticated mathematical algorithms with powerful computers has led to remarkable achievements, culminating in AI. AI has not only excelled in gaming scenarios, such as defeating chess champion Garry Kasparov but has also demonstrated profound real-world applications. These applications encompass the automation of routine tasks, speech and image comprehension, medical diagnostics mechanization, language translation, protein folding, antibiotic development, and autonomous vehicles like driverless cars.

Despite the significant strides made by AI, the concept of "artificial general intelligence" and the potential for machines to surpass human thought have ignited scholarly discussions. However, methodological challenges arise in proving that a machine has surpassed human intelligence, especially considering the vast array of conceivable goals. Max Tegmark proposed a definition of intelligence as "the ability to accomplish complex goals," encompassing knowledge acquisition, learning, problem-solving, logical algorithms, and planning.

While suitable for machines and technology, this definition arguably falls short in capturing the intricacies of human thought. This study critically evaluated Tegmark's definition, contending that it inadequately represented the essence of human thought. While acknowledging the applicability of Tegmark's definition to machines, the authors highlighted the need for a more nuanced understanding of human intelligence. By delving into the complexities of human cognition, the research aimed to provide a comprehensive perspective on intelligence that goes beyond mere goal accomplishment.

Understanding, implementation, and creativity

In assessing Max Tegmark's definition of intelligence as "the ability to accomplish complex goals," it was argued that machines and humans fundamentally differed in their understanding. Machines, exemplified in tasks like machine translation, established relations based on data-driven discoveries, lacking the depth of human conscious and unconscious associations. Human cognition, rooted in emotional intelligence and subjective experiences, exceeded the predictability achieved by AI.

The implementation of goals diverged significantly, with the embodied brain not only processing information but also generating unique subjective experiences. Creativity, a crucial aspect of intelligence, challenged the reductionist view of intelligence as goal accomplishment. Human creativity, often driven by unconscious processes, involves establishing remote associations and generating unexpected relations. Unlike machine-defined goals, creativity thrives in nonliteral, nonalgorithmic, metaphorical, imaginative, and transcendental realms. The less predefined and preconditioned a creative goal is, the higher the likelihood of breakthroughs, as illustrated in art, mathematics, and sciences.

In mathematics, the contrast between achieving a specific goal through machine learning (ML) in electroencephalography (EEG) and the unconscious genesis of a novel approach to the Lindelöf hypothesis underscored the nonliteral nature of creativity. The value of artistic creations, such as Picasso's Guernica, increased when goals were less predefined. Similarly, Richard Feynman emphasized that scientific breakthroughs stem from an open-minded exploration of nature, resisting predefined expectations. This comprehensive analysis challenged the oversimplified view of intelligence in AI, recognizing the intricate and multifaceted nature of human thought and creativity.

From Platonism to AI

The article explored the philosophical implications of AI, drawing parallels with Plato's emphasis on reason and abstraction. It critiqued the reductionist view that AI can simulate every aspect of intelligence by reducing phenomena to constituent elements. Rodney Brooks' acknowledgment of AI limitations highlighted the need for a more adaptive and locally responsive approach, contrasting with AI's current lack of adaptability. The discussion extended to evolutionary biology, emphasizing the intentionality and adaptability of life organisms, setting them apart from mechanical systems.

The article challenged the bias of those advocating AI superiority, arguing that physics' predictability does not necessarily apply to the trial-and-error nature of biological evolution. The proposed flexible framework suggested focusing on neuronal mechanisms like associations, continuity, and plasticity. The paper advocated reconsidering the dogmatic adherence to reductionism and computability in favor of a more nuanced understanding rooted in fundamental neuronal processes. It aligned with the "bag-of-tricks hypothesis," suggesting that specialized biological algorithms outperformed the general-purpose algorithms employed in AI.

The embodied brain, laterization, and the role of the glia

This article underscores the overlooked aspects crucial for understanding human thought that AI neglected: the impact of the body on the brain, the laterization of brain functions, and the vital role of glia cells. The brain's embodiment and complex connectivity influenced by neuromodulators from various nuclei emphasized the intricate interplay between the body and the brain. Laterization, evident in functional hemispheric differences, was deemed essential for proper brain functioning.

The significant role of glia cells, constituting 85% of brain cells, involves more than mere support; they contribute to vital substances, engage in direct and indirect communication with neurons, and enhance plasticity crucial for learning and memory. The authors argued that AI's attempts to replicate neuronal functions ignored these fundamental elements, hindering its potential to reach human thought.

They emphasized the richness and broader scope of human consciousness, grounded in dynamic interactions between conscious and unconscious processes, environmental factors, and physiological mechanisms. The limitations of mathematical modeling and AI algorithms, which mimic only conscious cortical functions, were highlighted, reinforcing the uniqueness and complexity of human thought that extended beyond the capabilities of AI.

Conclusion

In conclusion, the researchers challenged prevailing notions of AI surpassing human thought, revealing the inadequacy of current definitions. It delved into the intricacies of human cognition, emphasizing the overlooked roles of the body, brain laterization, and glia cells. By exploring the philosophical implications, they advocated for a flexible framework and highlighted the trial-and-error nature of biological evolution. Ultimately, they contended that human thought, with its richness and complexity, remains beyond the reach of current AI capabilities.