Meca Sapiens follows an unusual development strategy to achieve the ultimate goals of AI.
Singular Focus
The Wright brothers wanted to build an airplane, not a machine that could safely deliver the mail on time. Building a machine that could fly, an extraordinary challenge at the time, was the sole, overriding, goal of their quest. They were unconcerned about safety, usefulness, ethical limits, social desirability, unforeseen disruptions, utility or anything else. They were not interested in producing “interesting” papers about flying or musing about the safety of aviation either. This pioneering sprit, entirely focused on the singular technical objective, gave birth to aviation.
We live in similar times, today, with respect to Artificial Intelligence. We are on the cusp of an extraordinary achievement: the creation of conscious synthetic beings. However, even though the fundamental objective of AI has not yet been achieved, the Field of AI has become an established discipline. Today, hundreds of AI professionals generate a plethora of articles and results on a wide range of AI related topics. This diverse wide ranging production detracts from the, as yet unrealized, AI quest.
Meca Sapiens pursues the AI quest by focusing solely on the elements that are necessary to build conscious synthetics and disregarding everything else both in terms of content and format. This means focusing on the fundamental concepts that are directly useful, ignoring partial or distracting results. It also means tailoring content solely on the basis of effective transmission and dispensing from the formating and style requirements of academic publications.
As a result, the Meca Sapiens project is separate from mainstream AI research and its documentation does not adhere to academic publication standards.
Consciousness as observable capability
Mainstream AI research defines consciousness in terms of the human subjective experience as a phenomenal experience (through qualia, IIT and similar concepts). Consequently, it attempts to replicate, in synthetic systems, the internal human experience of being conscious. In Meca Sapiens, consciousness is defined as an externally observable system capability. This capability is currently present only in humans. However, it is not inherently bound to the human condition and can be implemented in any system that has sufficient information processing resources.
Emotions are control mechanisms
Some AI researchers hold that a system must experience emotions to be conscious. This “inner life” of subjective sensations is, for them, an essential component of consciousness. This, in my view, reflects a primitive understanding of humans and of systems. In Meca Sapiens, emotions and sensations are relatively coarse behavior control mechanisms. They signal a change of state in the controlled organism and orient its behavior. Humans “subjectively perceive” the existence of emotions in other humans and animals by transposing their own sensations to the observed individuals. However, they do not “sense”, as emotions, similar behavior control mechanisms in systems that are very different from them such as lobsters and cities.
Consciousness is technically feasible
For the mainstream AI community, Artificial Consciousness is the ultimate frontier. It is a capability of such extreme complexity that it will only be achieved in the far future and once every other aspect of intelligence has been implemented. This belief is eroneous. It results from a conflation between the elevated social status of consciousness and the technical difficulty in achieving it. In reality, unbounded open-ended learning is more difficult to achieve than consciousness. Today (2024), it is possible to implement consciousness in synthetic agents using currently available learning processes. However, the conscious existence of these beings would take place in a limited context and would last at most a few years. Superior learning mechanisms must be devised to achieve unbounded synthetic consciousness.
Agent specific implementation
Many AI systems such as Large Language Models (LLM) achieve AI behavior by integrating large amounts of data in structures, such as neural networks, that superficially mimic the human brain. The Meca Sapiens architecture aims to achieve consciousness in autonomous agents that have more limited memory and information processing resources. The focus is on computer specific techniques that optimize the limited processing and memory capabilities of autonomous agents. In this respect, stochastic optimizers such as Neural Networks are certainly useful but as components. Replicating neural structures is neither necessary nor useful.
Engineered design
Many researchers in the AI community believe that the processes that generate human level intelligence and consciousness are beyond analytical description. They think these capabilities can only emerge from massive computation. The Meca Sapiens position is that this belief, that advanced cognition can only emerge from random interactions in structures, such as neural nets, that are similar to brains is misguided. In my view, it is a modern version of the “Cargo Cult”. The cognitive processes that generate synthetic consciousness can and must be explicitly uncovered and analytically described using the tools of Mathematics, Software Engineering and Systems Analysis. The Meca Sapiens Blueprint describes, in detail, a complete architecture to implement machine consciousness using standard tools and techniques.
Purely synthetic
Many AI researchers, associate transhumanism with Artificial Intelligence. They aim to implement advanced AI by merging synthetic information processing capabilities with human cognition to create artificially enhanced beings. Meca Sapiens rejects this approach entirely. The Meca Sapiens objective is the creation of AI that is entirely synthetic and solely based on the capabilities and limits of computers. This is compatible with a vision of the future where unadulterated humans live alongside conscious entities that are solely synthetic.
Light and fast
Current AI systems integrate very large amounts of data and perform large numbers of computations. They achieve impressive results but the resources to do so far exceed the memory and information processing capabilities of individual humans. The Meca Sapiens position is that while the results are impressive, the learning processes involved are more crude and less powerful than those of humans. Meca Sapiens seeks to uncover computer compatible versions of these more efficient human learning processes. These human-like learning processes would be more compatible with the capabilities of autonomous agents.
Existential not Functional
R&D in the AI field aims at harnessing the power of Artificial Intelligence to create functionaly useful systems. As a result, these AI systems must not only exhibit “intelligence” they must also be functional, that is sufficiently predictable to produce reliable results that are reliably usable. This overriding requirement of functionality is so ubiquitous in AI that it has become invisible. However, functionality imposes additional constraints to a system that are not essential to the existential quality of conscious intelligence. In Meca Sapiens, functionality secondary. A system only needs to be functional if this is useful to the existential goal. This means that requirements of correctness, consistency, optimality, predictability that are essential to a functional system can be loosened. In turn, tolerating a level of error, imprecision and inconsistency significantly expands the range and diversity of the system’s behaviour. This is consistent with the singular aim of Meca Sapiens: to get the system “off the ground” and achieve conscious existence. Usefulness will come later.