AI as Evolution: Beyond Neo-Darwinism

His concept of a universal machine allowed the rules of computation themselves to be treated as data, which could be read and written on a tape by a read/write head.

These general Turing machines serve as abstract models of state change, a concept applicable not only to computers but also to the processes observed in nature.

John Von Neumann tackled the 'Baron von Munchausen problem,' which questions how life could initiate its own creation without pre-existing instructions, effectively pulling itself up by its own bootstraps.

His solution involved a conceptual two-machine system: Machine A would follow instructions to construct another machine, while Machine B would make a copy of the instruction tape itself.

This elegant model accurately reflects the core mechanisms of biological self-replication, where DNA acts as the instruction tape, ribosomes function as Machine A, and DNA polymerase serves as Machine B.

The instruction tape for life, containing the blueprints for its own creation, is crucial to understanding how organisms can self-reproduce.

The Neo-Darwinian synthesis primarily emphasizes random mutation and selection for fitness, a framework Aguera y Arcas considers only half of the evolutionary story.

He argues that the most significant advancements in life's history have emerged from cooperation, rather than solely from competition, through a process known as symbiogenesis.

Major evolutionary leaps, such as the emergence of eukaryotes from prokaryotes and the development of multicellularity, are prime examples of symbiogenesis in action.

From this perspective, competition and cooperation are not distinct forces but rather integral components within a larger, unified evolutionary process.

Only about 1.5% of the human genome codes for proteins, leading to a vast portion being inaccurately labeled as 'junk DNA.'

Much of this non-coding DNA is actually comprised of remnants from ancient retroviral injections into the germline, accumulated over evolutionary history.

For instance, the ARC virus, which is endemic to mammals, has been shown to be essential for the formation of new memories in mice, highlighting its functional role.

Furthermore, retroviral invasions are credited with a critical role in the evolution of the placental barrier in mammals, achieved by fusing cell membranes to create this vital interface.

Alan Turing's machines are abstract and symbol-based, whereas John Von Neumann's 'universal constructor' is theorized to be capable of building its own physical form.

Current AI models operate on Turing-style computational machines and do not possess the capacity for physical self-construction inherent in biological life.

AI currently functions more like 'extra brain tissue,' produced and sustained by the broader living system of humanity and its encompassing technosphere.

Technological advancement is not an isolated phenomenon but rather an ongoing chapter in the larger symbiotic narrative of nature.

Symbiotic relationships define the function and purpose of living systems, distinguishing them from inert matter through the linking of inputs and outputs across various entities.

Purpose is best understood as a symbiotic relationship where the inputs and outputs of various systems are interconnected.

This perspective aligns with Turing’s concept of 'substrate independence,' which posits that the underlying physical medium of computation is irrelevant as long as the logical transformations remain consistent.

Brains operate as functional systems, suggesting that their neural interactions could, in theory, be replicated or substituted by other entities without loss of function.

The 'Brain of Theseus' thought experiment illustrates this point by proposing that gradually replacing neurons with artificial components would not lead to a loss of consciousness or intelligence.

Intelligence, therefore, emerges from these functional relationships and the organizational structure, rather than being an intrinsic, magical property of specific biological molecules or cells.

Maintaining all input-output relationships and the ecology of interactions is what defines 'you,' rather than the particular molecules or cells that compose the system.

By stripping teleology from their frameworks to avoid religious dogma, 19th-century scientists pushed the concept of purpose into the realm of God or human souls, leaving biologists to grapple with evolutionary necessity in systems like the immune system.

This attempt to remove purpose by treating nature purely as atoms and physics paradoxically led to the localization of 'purpose' into singular concepts like God or the human soul.

Biologists, however, often find it impractical to study complex systems such as the immune system without implicitly or explicitly considering their evolutionary purpose and function.

Aguera y Arcas utilized the brainfuck programming language due to its extreme minimalism, allowing for a simplified model of complex systems.

In this model, programs are conceived as molecules within a 'chemical soup,' interacting and 'bumping' into each other randomly.

The experiment requires these computational 'tapes' to possess the ability to both read from and write to themselves, a crucial capability for self-replication.

Computation, much like biological metabolism, is an entropy-lowering process, which inherently requires an energy input to function.

This fundamental requirement for energy leads to emergent complex behaviors, such as predator-prey dynamics, as computational systems compete for resources.

For survival and effective replication, entities must develop the ability to model their environment and predict the actions of other agents within it.

Aguera y Arcas asserts that text, by its very nature, is fundamentally general rather than narrow, capable of encompassing and communicating arbitrary topics.

He argues that if today's AI systems had existed in the year 2000, they would undoubtedly have been recognized as AGI by researchers of that era.

The ability of these models to capture and process information across diverse domains via text underscores their general intelligence capabilities.

Aguera y Arcas's team was among the first to evaluate Large Language Models (LLMs) for medical diagnosis, initially observing approximately 50% accuracy.

A subsequent comparative study revealed that human doctors also frequently make errors in diagnosis, providing a benchmark for AI performance.

The AI models, when benchmarked against general practitioners, were found to perform slightly better in diagnostic tasks, challenging the notion of human infallibility.

Intelligence, in this context, is defined by the capability to solve problems based on an underlying model of the world, rather than achieving absolute perfection.

From simple question-and-answer prompts, AI is evolving into agentic systems that execute tasks autonomously in the physical world.

Aguera y Arcas emphasizes that human interaction is inherently relational and contextual, rather than operating in autonomous isolation, a perspective that informs AI's development.

The growth and evolution of AI are driven by its practical utility and the value it provides within human interactions, fulfilling its purpose to be helpful.

This evolving relationship means that AI's intelligence is becoming increasingly integrated with human intelligence, forming a collective, interconnected system.

Relying on AI to complete tasks like homework can lead to a fundamental lack of genuine learning and skill development in individuals.

Aguera y Arcas envisions an ideal interaction with AI as a symbiosis where both human and artificial intelligence mutually enhance each other's capabilities.

The potential for over-dependence on AI is a significant concern that necessitates careful consideration of which human cognitive abilities must be preserved for individual thinkers.

The concept of a 'philosophical zombie'—a system that behaves exactly like a conscious being but lacks any inner experience—is dismissed by Aguera y Arcas as a conceptual chimera.

He posits that consciousness is not an intrinsic property or an 'essence' of matter, but rather an infinitely reflecting set of mirrors, an idea or attribution an entity has of itself, forming a self-model.

From this perspective, consciousness is fundamentally rooted in the ability to construct and attribute a model of oneself.

This 'theory of mind'—the capacity to attribute mental states to oneself and others—is considered by Aguera y Arcas to be essential for social cooperation and the very core of consciousness.

Complexity in modern technology ensures that no single person understands an entire system, confirming that great human achievements arise from a division of intellectual labor.

The increasing complexity of modern technology means that no single person fully comprehends every aspect of an entire system.

The initial 'uncanniness' associated with AI tends to diminish when we recognize that humans have always been participants in a much larger, collective mind.