a person playing chess

The problem of relevance realization for artificial general intelligence

Philosophy Psychology Science Technology

As artificial intelligence continues to advance, there is increasing interest and investment in developing machines that can achieve artificial general intelligence (AGI) – the ability to perform a wide range of intellectual tasks that are characteristic of human beings. However, one of the biggest challenges in developing AGI lies in the ability to enable machines to: 

  • understand the relevance of information in a given context.
  • negate a near-infinite (and thus combinatorially explosive) number of other, irrelevant inputs. 

That cognitive process is referred to as “relevance realization”. 

This essay argues that relevance realization is a critical problem that needs to be solved in order to achieve AGI. Relevance realization is an essential component of human cognition, and developing machines that can perform this function is essential for the development of true artificial intelligence. Given the evolutionary and embodied nature of relevance realization, the problem that it presents for AGI may well be insoluble. 

What is relevance realization?

John Vervaeke is a cognitive scientist and professor at the University of Toronto who has developed the relevance realization theory, which offers a new framework for understanding human consciousness, meaning, and purpose. According to Vervaeke, relevance realization is the process by which we create and maintain meaning in our lives (Vervaeke, 2017). This process involves actively seeking out and identifying meaningful connections between different pieces of information, rather than simply reacting to stimuli (Vervaeke, 2017).

At the core of relevance realization is the idea that meaning is not a fixed entity that exists in the external world waiting to be discovered. Instead, meaning is actively created and maintained by our cognitive systems (Vervaeke, 2017). This means that meaning is not static but rather is constantly evolving and adapting to new circumstances.

The process of relevance realization can be understood through concrete examples. For instance, imagine walking through a park and seeing a tree. The cognitive system is immediately engaged in pattern recognition, searching for meaningful connections between different sensory inputs. The color of the leaves, the texture of the bark, and the sound of the wind rustling the branches are all sensed and processed by the cognitive system.

As the cognitive system processes this information, it engages in abstraction and categorization, grouping together different pieces of information that are related to each other. For example, the color of the leaves, the texture of the bark, and the sound of the wind might all be grouped together as part of the sensory experience of the tree (Vervaeke, 2017).

The cognitive system is continually engaged in the process of relevance realization, seeking out meaningful connections between different pieces of information and integrating them into an overall sense of the world. This process of sense-making is not a passive or automatic process, but rather requires active engagement and attention (Vervaeke, 2017).

The relevance realization theory extends beyond simple sensory experiences and applies to all aspects of human cognition. For example, when reading a book or listening to a lecture, the cognitive system is continually seeking out meaningful connections between the information being presented and integrating it into an overall understanding of the topic at hand (Vervaeke, 2017).

The importance of relevance realization is not limited to the creation of meaning in the moment. It is also critical for long-term learning and memory formation (Vervaeke, 2017). By continually seeking out meaningful connections between different pieces of information, the cognitive system is better able to integrate new knowledge into existing knowledge structures and create more robust and accurate mental models of the world.

Furthermore, relevance realization plays a crucial role in our ability to set goals and pursue them. By seeking out meaningful connections between our present circumstances and our desired outcomes, we are better able to develop a sense of purpose and direction in life (Vervaeke, 2017).

The relevance realization theory has important implications for a variety of fields, including education, psychology, and philosophy. By better understanding the process of relevance realization, educators can develop more effective teaching strategies that help students build stronger connections between different pieces of knowledge.

Psychologists can also use the relevance realization theory to better understand the cognitive mechanisms underlying a range of mental health conditions, such as depression and anxiety. By understanding how individuals create and maintain meaning in their lives, psychologists can develop more effective treatments that help individuals regain a sense of purpose and direction (Vervaeke, Ferraro, and Standing, 2018).

Finally, the relevance realization theory has important implications for philosophy, particularly in the area of existentialism. According to Vervaeke, existentialism is based on the assumption that meaning is not inherent in the world, but rather must be created by individuals (Vervaeke, 2017; Vervaeke 2021). The relevance realization theory provides a more detailed account of how this process of meaning-making occurs, and offers new insights into how individuals can find purpose and meaning in their lives.

Overall, the relevance realization theory represents an important contribution to our understanding of human cognition, consciousness, and meaning-making. By emphasizing the active and dynamic nature of the process of sense-making, it offers new insights into how we create and maintain meaning in our lives, and has important implications for a wide range of fields, from education and psychology to philosophy and existentialism.

As we’ll see, understanding relevance realization is also critical to the project of artificial intelligence (AI) and to any prospects of artificial consciousness.

Defining consciousness

Phenomenal consciousness refers to the subjective experience of sensory and perceptual events. It is the “what it is like” to experience something, such as the color red or the taste of chocolate. According to Chalmers (1995), phenomenal consciousness is a fundamental aspect of the mind that cannot be reduced to physical or neural processes. It is a subjective and irreducible feature of experience that is often described as “qualia.”

Meta-consciousness, on the other hand, refers to the ability to reflect on and be aware of one’s own mental processes. It is sometimes called “access consciousness” because it involves the ability to access and control one’s own thoughts and perceptions. According to Baars (1988), meta-consciousness is a higher-order cognitive process that allows us to monitor and manipulate our own mental states.

The main difference between phenomenal consciousness and meta-consciousness is that the former refers to the subjective experience of sensory events, while the latter refers to the ability to reflect on and be aware of those experiences. Phenomenal consciousness is often described as a “first-person” perspective, while meta-consciousness involves a “second-person” perspective in which one is aware of one’s own mental states as objects of thought.

Another important difference between these two concepts is their relationship to neural activity. Phenomenal consciousness is often associated with activity in specific regions of the brain that are involved in sensory processing, such as the primary visual cortex or the gustatory cortex. In contrast, meta-consciousness is associated with activity in more widespread brain networks that are involved in higher-order cognitive processes, such as the prefrontal cortex or the parietal cortex (Baars & Franklin, 2003).

Despite these correlations, there exists no scientific theory of either phenomenal or meta-consciousness, as both have thus far proven to be irreducible to physical brain states. That result defies today’s mainstream neuroscientific paradigm, which assumes that consciousness supervenes on the physical. As we’ll explore later on, physics itself is moving in a direction that would also contradict such an assumption. 

What is the relationship between relevance realization and consciousness?

The relationship between relevance realization and consciousness is complex and multifaceted. According to Vervaeke (2017), consciousness is intimately tied to relevance realization because it involves the active construction of a model of the world that is constantly updated and modified based on incoming sensory information. This model is not a passive reflection of the world, but rather an active and dynamic representation that is shaped by our goals, expectations, and beliefs.

Moreover, Vervaeke (2017) argues that relevance realization is a necessary condition for consciousness, as it enables us to extract meaning from sensory input and construct a coherent representation of the world. Without relevance realization, sensory input would be meaningless and the world would appear as a chaotic and disordered collection of stimuli.

Recent research has also shown that the brain networks correlated with relevance realization are closely linked to those correlated with consciousness. For example, the default mode network (DMN) has been associated with both relevance realization and self-referential processing, which are key aspects of consciousness (Kleckner et al., 2017). The DMN is a network of brain regions that is active when the brain is at rest, and is thought to be correlated with a range of cognitive processes, including self-reflection, social cognition, and memory.

More specifically, while relevance realization is not the same as either phenomenal consciousness or meta-consciousness, it has important connections to both.

On the one hand, relevance realization can be seen as a key aspect of phenomenal consciousness, as it involves the perceptual and cognitive processes of subjective experience. According to Vervaeke (2019a), relevance realization is a process of “informational integration” that allows us to combine multiple streams of sensory information into a coherent and meaningful experience. This process involves both bottom-up sensory processing and top-down cognitive processing, and it is closely related to phenomenal consciousness.

On the other hand, relevance realization also has important connections to meta-consciousness, as it involves the ability to reflect on and be aware of one’s own cognitive processes. According to Vervaeke (2019a), relevance realization is a form of “attentional engagement” that allows us to focus our attention on the most salient and relevant aspects of our environment. This process requires meta-cognitive skills such as self-awareness, monitoring, and control, which are key components of meta-consciousness.

In other words, relevance realization can be seen as a bridge between phenomenal consciousness and meta-consciousness, as it involves both subjective experience and the ability to reflect on and control that experience. By integrating and transforming information in a meaningful way, relevance realization allows us to perceive the world in a way that is both rich and coherent, while also giving us the flexibility and adaptability to adjust our attention and focus as needed.

Challenge of relevance realization for AGI

Relevance realization is a cognitive process that is central to human consciousness, and as such, it presents a major challenge for the development of AGI and conscious AI. AGI refers to machines that are capable of performing any intellectual task that a human can do, while conscious AI refers to machines that have subjective experience and self-awareness (Chalmers, 2018). Both types of AI would need to be able to perform relevance realization in order to perceive and understand the world in a meaningful way.

The challenge of relevance realization for AGI and conscious AI is that it requires a deep understanding of how information is integrated and transformed in the human mind (arguably, animals perform a lower-order of relevance realization too, and even this would prove highly difficult for AI). According to Vervaeke (2019b), relevance realization involves a complex set of cognitive processes that operate at multiple levels of abstraction, including perception, attention, memory, and reasoning. These processes are highly interdependent and operate in a dynamic and context-sensitive manner, making them difficult to replicate in a machine.

One key aspect of relevance realization that presents a challenge for AGI and conscious AI is its dependence on embodied cognition. Embodied cognition refers to the idea that cognitive processes are grounded in the physical body and its interaction with the environment (Lakoff & Johnson, 1999). This means that our perception and understanding of the world is shaped by our bodily experiences, and that our cognitive processes are closely linked to our sensorimotor systems. AGI and conscious AI would need to be able to simulate this embodied experience in order to perform relevance realization in a way that is similar to humans.

In other words, the syntax of our perceptual language is isomorphic to the structure of our cognition (Santos, 2023). Computers would need to be embodied, essentially both “reading” and “writing” these languages in order for reality to be intelligible to them through relevance realization, a prerequisite for general problem solving and intelligence. 

Another challenge of relevance realization for AGI and conscious AI is its dependence on context and meaning. Humans are able to perceive and understand the world in a meaningful way because we are able to contextualize and interpret sensory information in light of our previous experiences and knowledge (Barsalou, 2008). This requires a deep understanding of language and culture, as well as the ability to form and manipulate mental representations that capture the meaning and significance of different stimuli.

Moreover, relevance realization requires a high degree of flexibility and adaptability, which is challenging to replicate in a machine. Humans are able to adjust their attention and cognitive processes in response to changing environmental and task demands, and to creatively generate new solutions to novel problems. This type of flexibility and adaptability is difficult to program into a machine, as it requires a degree of autonomous decision-making and creativity that is not currently possible with AI systems.

In addition to these challenges, there are also ethical concerns associated with the development of AGI and conscious AI. As AI systems become more complex and capable, there is a risk that they will become unpredictable and uncontrollable, leading to unintended consequences and potentially catastrophic outcomes (Bostrom, 2014). There is also a risk that AGI and conscious AI systems could become self-aware and experience suffering, raising questions about their moral status and the ethical implications of their creation and use (Bostrom & Yudkowsky, 2014).

After all, humans already believe each other and animals to be conscious, yet we subject these conscious beings to untold suffering every day on this planet. How much worse would we then behave toward conscious computers, devices that, to us, have always been unfeeling tools that we can use as we see fit?

AGI, relevance realization, and phenomenal consciousness

While some AI systems have demonstrated a degree of meta-consciousness, such as the ability to monitor and control their own processing and decision-making (Metzinger, 2018), it is unlikely that machines will ever be able to achieve phenomenal consciousness. This is because phenomenal consciousness is thought to be closely tied to the subjective experience of embodiment, which arises from the integration of sensory information with motor and affective processes (Lakoff & Johnson, 1999). The embodied nature of consciousness makes it difficult to simulate or replicate in a machine, as it requires a deep understanding of the interplay between sensory information, emotions, and bodily responses (Varela et al., 1991).

The machine would need to not only focus on the relevant information at any given, infinitesimal point in time, t, but also simultaneously negate the combinatorially explosive number of irrelevant pieces of information (Vervaeke, 2017). That is, an embodied agent exists as a relative entity, constantly in dialogue with a nearly infinite number of other parts of reality, in a continuous process of co-realization. The agent must perform these attentional and negation functions at each successive point in time, t+n, in order to survive and to solve problems. Despite the fact that we can direct machines’ attention toward salient details for a given task, we still have no idea how to program them to perform the negation of such a vast array of inputs as that which embodied conscious agents process every fraction of a second. Until they have that ability, machines will not achieve the general intelligence of humans.

In other words, relevance realization is the major challenge to overcome if AGI is ever to be conscious in the way that embodied conscious agents are. So far, this kind of embodied experience is exclusively associated with metabolizing organisms, with vast evolutionary histories shaped by constantly changing agent-arena relationships between themselves and their environments. 

This could, however, give us a clue as to the best approach to AGI engineering.

Evolutionary models of AGI development are based on the idea that AGI will emerge through an evolutionary process that mimics natural selection. These models are inspired by the process of biological evolution, where random variations in genetic material can result in the emergence of new traits that increase an organism’s fitness in a given environment. Similarly, evolutionary models of AGI development rely on the creation of diverse AI systems, which are then subjected to selection pressures and the replication of those systems that demonstrate the most desirable features (Poli, Langdon, & McPhee, 2008).

One approach to evolutionary AGI development is to use a genetic algorithm, which is a method of optimization inspired by biological evolution. In this approach, the AI system is represented as a genome, and variations in the genome are introduced through mutations and recombination. The fitness of each genome is then evaluated based on how well it performs a specific task, and the genomes with the highest fitness are selected for replication and further mutation (Eiben & Smith, 2015).

Another approach is to use a technique called neuroevolution, where the structure and weights of a neural network are optimized through evolutionary processes. In this approach, a population of neural networks is created with random weights, and the networks that perform the best on a given task are selected for reproduction. This process is repeated, with variations introduced through mutations and recombination, until a network that performs optimally is produced (Stanley, Miikkulainen, & Clune, 2019).

Evolutionary models of AGI development have the advantage of being able to explore a wide range of possibilities and find solutions that are difficult to anticipate or engineer directly. However, these models also face challenges, such as the large search space of possible solutions, the difficulty of accurately defining fitness criteria, and the potential for the optimization process to get stuck in local optima (Poli et al., 2008). Not only that, but natural evolutionary processes seem to take a very long time, and the claim that human engineers could speed up that process remains highly speculative, especially given how little we understand about our own consciousness. 

In other words, even our best approach very likely falls short of achieving relevance realization in AGI. 

Furthermore, the subjective and qualitative nature of phenomenal consciousness makes it difficult to define and measure in a machine-readable way. While some researchers have proposed measures of consciousness based on neural activity or behavior (Tononi, 2008), these measures are often controversial and lack a clear definition of what it means to be conscious. For instance, Tononi’s Φ (“phi”) measure still currently relies on subjects’ reportability of experience, making it a measure of meta-consciousness instead of phenomenal consciousness. 

As expected, meta-consciousness is more amenable to measurement and manipulation in an AI system, as it can be defined in terms of observable behaviors and cognitive processes. This has led some researchers to argue that the focus of AI research should be on achieving meta-consciousness, rather than trying to replicate phenomenal consciousness (Baars & Franklin, 2003). However, this approach raises questions about the nature and limits of consciousness, and whether it is possible to achieve true intelligence without some form of subjective experience. 

The close links between consciousness and relevance realization suggest that subjectivity and embodied cognition are required for intelligence. However, this leads us to a more fundamental question outside the realm of neuroscience…just what is this spacetime environment in which our minds are embodied?

Physics, not neuroscience or computer science, will answer our consciousness questions

Physicists have long sought to understand the fundamental nature of spacetime, the four-dimensional fabric that underlies our understanding of the universe. However, some physicists have recently challenged the notion that spacetime is a fundamental aspect of reality, arguing instead that it emerges from a more basic, underlying structure.

One reason for this shift in thinking is the apparent incompatibility between our understanding of spacetime and the principles of quantum mechanics, which describe the behavior of particles at the smallest scales. While spacetime is continuous and smooth, quantum mechanics suggests that particles can exist in multiple locations simultaneously, with their positions and velocities being described by probability distributions rather than definite values (Smolin, 2015). This has led some physicists to explore the possibility that physicality and spacetime are not fundamental aspects of reality, but emergent properties of a different ontic primitive. 

A proposed model for such a structure is called loop quantum gravity, which suggests that space is made up of discrete, indivisible units known as loops or spin networks (Rovelli, 2011). According to this model, spacetime is an emergent property of these underlying structures, rather than being a fundamental aspect of reality.

Another reason for questioning the fundamentality of spacetime is the discovery of phenomena such as quantum entanglement and black hole entropy, which suggest that information is more fundamental than spacetime itself (Susskind, 2016). According to this view, the apparent smoothness and continuity of spacetime is an illusion, with the true nature of reality being more akin to a holographic projection of information.

Building on that approach, Donald Hoffman, a cognitive scientist, has also proposed that spacetime may not be a fundamental aspect of reality, based on his work on the interface between perception and reality (Hoffman, 2019). In his view, the world we perceive is not a direct representation of reality, but rather a set of symbols that our brains use to make sense of sensory information. Therefore, our perception of spacetime may not necessarily reflect reality’s true nature.

Hoffman has suggested that the physics of quantum mechanics provides support for his claim. According to the principle of complementarity in quantum mechanics, particles can exhibit either wave-like or particle-like behavior depending on the experimental setup (Bohr, 1928). This suggests that the properties of particles are not fixed or objective, but rather depend on the observer’s perspective.

Hoffman has taken this idea further, proposing that reality itself may not be fixed or objective, but rather depend on the observer’s perspective. He has suggested that spacetime may be a construct that emerges from a more basic set of properties, such as relational properties between conscious agents (Hoffman, 2019).

Nima Arkani-Hamed, a theoretical physicist, has also put forth the idea that spacetime is not fundamental, but rather emerges from a more fundamental structure, which he calls the “amplituhedron” (Arkani-Hamed & Trnka, 2014). According to this theory, particles and their interactions can be described more efficiently and accurately using the mathematical concept of the amplituhedron, rather than relying on the traditional space-time-based approach.

The amplituhedron is a geometric object that encodes the probability amplitudes for particle interactions in a way that is independent of space and time. This approach has been shown to produce the same predictions as traditional quantum field theory, but with far fewer calculations (hundreds of pages of algebra down to a few equations that can be crunched by hand) and a more elegant mathematical structure.

Arkani-Hamed’s proposal has gained attention and interest from physicists because it suggests a possible path forward in reconciling the theories of general relativity and quantum mechanics, which have been notoriously difficult to unify (Kovachy, 2019). By starting from a more fundamental structure that does not rely on the concept of spacetime, it may be possible to develop a theory that encompasses both relativity and quantum mechanics.

Of course, if spacetime is not fundamental, and since the brain is an object that we perceive within spacetime, then we must question the theory that the brain generates consciousness. Rather, it would seem that consciousness, or that which perceives reality as the “interface” of spacetime, must precede physical entities, rather than the other way around. 

This would explain why we encounter the hard problem of consciousness under physicalism. That is: there is no way, even in principle, to explain how physical processes and entities, which are purely quantitative, could ever give rise to phenomenal consciousness, which is purely qualitative (Chalmers, 1995). 

Under this emerging paradigm in physics, spacetime is an epistemic entity, not an ontic one. The brain is the perceptual image of consciousness, not the generator of it. The image of a thing and the thing itself are always tightly correlated, but do not display a causal relationship, and this is precisely what we have found with consciousness and brain states. Only correlations, but no causation.

Thus, we can now explain why, despite decades of advancements, neuroscience has failed to produce even a single scientific, physical theory of consciousness. 

Even if AGI somehow achieves relevance realization, we still face the hard problem of consciousness as a blocker for AGI being phenomenally conscious under physicalism. Indeed, the theory that the physical gives rise to consciousness faces a myriad of problems today, all relics of physicalist assumptions that are outdated, given the new directions that physics is taking us. 

If we are to understand AGI, as well as our own consciousness as embodied agents, we must recognize the logical inconsistencies in our current paradigm and, in turn, find a better metaphysical paradigm than physicalism to guide our sense of what is plausible. 

Bibliography

Arkani-Hamed, N., & Trnka, J. (2014). The Amplituhedron. Journal of High Energy Physics, 2014(10), 1-54.

Baars, B. J. (1988). A cognitive theory of consciousness. Cambridge University Press.

Baars, B. J., & Franklin, S. (2003). How conscious experience and working memory interact. Trends in Cognitive Sciences, 7(4), 166-172.

Barrett, J. A., & Byrne, P. (2020). The role of simulation in constructing models of the world. Synthese, 1-24.

Barsalou, L. W. (2008). Grounded cognition. Annual Review of Psychology, 59, 617-645.

Bohr, N. (1928). The quantum postulate and the recent development of atomic theory. Nature, 121(3048), 580-591.

Bostrom, N. (2014). Superintelligence: Paths, dangers, strategies. Oxford University Press.

Bostrom, N., & Yudkowsky, E. (2014). The ethics of artificial intelligence. In R. Calo, G. F. M. Boella, & E. A. Ferrari (Eds.), Legal and ethical implications of artificial intelligence (pp. 19-39). Springer.

Chalmers, D. J. (1995). Facing up to the problem of consciousness. Journal of Consciousness Studies, 2(3), 200-219.

Chalmers, D. (2018). The hard problem of consciousness. In R. J. Stainton (Ed.), Contemporary debates in cognitive science (pp. 113-131). Blackwell.

Eiben, A. E., & Smith, J. E. (2015). From evolutionary computation to the evolution of things. Nature, 521(7553), 476-482.

Hoffman, D. D. (2019). The case against reality: Why evolution hid the truth from our eyes. W. W. Norton & Company.

Kleckner, I. R., Zhang, J., Touroutoglou, A., Chanes, L., Xia, C., Simmons, W. K., … & Barrett, L. F. (2017). Evidence for a large-scale brain system supporting allostasis and interoception in humans. Nature Human Behaviour, 1(5), 0069.

Kovachy, T. (2019). Nima Arkani-Hamed Is Using Twistor Theory to Uncover the True Nature of Space-Time. Quanta Magazine. Retrieved from https://www.quantamagazine.org/nima-arkani-hamed-is-using-twistor-theory-to-uncover-the-true-nature-of-space-time-20190923/

Lakoff, G., & Johnson, M. (1999). Philosophy in the flesh: The embodied mind and its challenge to western thought. Basic Books.

Metzinger, T. (2018). Real virtuality: A code of ethical conduct. Journal of Consciousness Studies, 25(9-10), 125-143.

Poli, R., Langdon, W. B., & McPhee, N. F. (2008). A field guide to genetic programming. Lulu. com.

Rovelli, C. (2011). Loop quantum gravity. Living Reviews in Relativity, 14(1), 1-160.

Santos, M. (2023). Why Reality Must Be Intelligible: Language & Perception. BCP Journal, 14. Retrieved from https://michaelsantosauthor.com/bcpjournal/why-reality-must-be-intelligible-language-perception/

Smolin, L. (2015). Time reborn: From the crisis in physics to the future of the universe. Houghton Mifflin Harcourt.

Stanley, K. O., Miikkulainen, R., & Clune, J. (2019). Designing intelligence: Evolving neural networks through augmenting topologies. MIT Press.

Susskind, L. (2016). Black holes: The holographic principle. World Scientific.

Vervaeke, J. (2017). The relevance realization framework: A comprehensive paradigm for cognitive science. Journal of Consciousness Studies, 24(5-6), 7-57.

Vervaeke, J., Ferraro, L., & Standing, L. (2018). Cognitive science, relevance realization, and existential meaning: Implications for mental health and therapy. In G. Matthews, D. Ungar, & G. J. F. van den Berg (Eds.), The handbook of counseling psychology (4th ed., pp. 35-56). Sage.

Vervaeke, J. (2019a). Relevance realization. In J. W. Schooler (Ed.), The science of consciousness (pp. 273-290). Routledge.

Vervaeke, J. (2019b). The relevance realization challenge: A philosophical and cognitive science perspective. Frontiers in Psychology, 10, 2817.

Vervaeke, J. (2021). Finding relevance: John Vervaeke on meaning-making, relevance realization, and the importance of cultivating a meaningful life. [Interview with John Vervaeke]. Meaningoflife.tv. https://meaningoflife.tv/videos/45847

michael.santos

Michael Santos is a thriller author, amateur philosopher, member of the American Philosophical Association (APA), and is a technology industry writer. Explore his thriller novels at: https://michaelsantosauthor.com/

http://michaelsantosauthor.com