The challenge with Western intellectual approaches to philosophy is that they rely on some framework for understanding the world, where each such framework precludes some subset of possibilities, and predisposes others, as opposed to more holistic Eastern approaches of considering each philosophy, ideology, or any other cognitive construct to be a function that generates its own outcomes, and just observing what each functions to achieve in every aspect of this existence. The benefit of Western intellectual approaches is that they are more readily understood and communicated, whereas Eastern approaches are notoriously difficult to understand and communicate without years of such observation, usually in the form of meditation or other practices. This makes them largely unavailable to Western science.
A recently developed approach called Human-Centric Functional Modeling (HCFM) however bridges these two approaches by representing the human systems through which we perceive and otherwise interact with the world as moving through graphs called functional state spaces, where these graphs can also hypothetically be used to describe the behavior of any living system, whether as simple as homeostasis or as complex as cognition. In these functional state spaces, all problems are defined as the lack of a path from one functional state to another, resulting in the insight that all adaptive processes through which nature solves problems are potentially generalizations of the same solution in functional state space. Representing the behavior of the cognitive system in terms of a set of paths through this mathematical space means that any property of that system’s behavior potentially can be given a mathematical definition, which would mean that properties ascribed to the human system within existential traditions, where those properties are only discernable through the first person observation generally rejected by as subjective and unscientific by Western science, might for the first time be objectively validated. Since many if not most properties of cognition can’t yet be observed by objective third party methods and tools, this would be a major step forward in our understanding.
An example of the usefulness of this approach is that it allows us to model the collective problem-solving ability of networks of systems in terms of their ability to navigate a given volume and density of the collective functional state space per unit time, which to date might be the only functional model of intelligence that has been hypothesized, and it allows us to see how networks create the potential to exponentially increase that ability. This in turn has led to a model for group decision-making called a General Collective Intelligence or GCI platform that copies this same solution. GCI is important because arguments can be constructed using this functional state space approach to support the prediction that GCI creates the capacity to exponentially increase the general problem-solving ability (intelligence) of groups.
An example of the value of a functional modeling methodology that facilitates merging these intellectual and existential approaches is an article here which concludes by saying “successful AI labs like DeepMind, OpenAI, and Anthropic do not seem to espouse this “pivotal act” philosophy for doing good in the world”. A functional state space based model of individual and group decision-making must be able to accommodate any possible interpretation of any reasoning and must be able to model the fitness of that reasoning at achieving its targeted outcome. This involves defining a “fitness space” for each individual cognition, as well as a collective fitness space for the group. Considering the basic problem that the individual fitness space is hypothesized to solve, as opposed to the problem that the collective fitness space solves, and considering the fact that GCI has not yet been implemented, and considering the prediction that collective fitness cannot reliably be optimized without GCI at group sizes above those characteristic of the ancestral tribes and villages humans evolved in (according to the collective social brain hypothesis), this suggests that the exact opposite is true. Namely, that this “pivotal act” and its accompanying philosophy might be the only reliably achievable outcome of trying to achieve AI safety through the current centralized methods, as opposed to doing so through a network based approach. In fact, this modeling approach suggests that the recent fear-mongering about the danger of AI and the recent petition to slow down the development of AI will most likely function to provide a competitive advantage to the signatories of this petition, and will in a number of ways further centralize control over AI, and therefore will function to make AI more dangerous.
The collective social brain hypothesis posits that individuals tend strongly to fall into either of two groups, one of which that tends to use type I (intuitive) reasoning processes on specific types of topics, and the other that tends to use type II (logical) reasoning processes on those same topics. The word “logical” here does not mean “correct”. It simply means some combination of reasoning that can be free of intuitive processes such as detecting what the consensus is. Both reasoning types are optimal in different problem domains. The challenge is that an understanding of the collective social brain hypothesis is required to understand the need for GCI to switch to the optimal reasoning type as required to achieve collective fitness, and experiments suggest that understanding the collective social brain hypothesis requires a non-consensus based reasoning process than cannot reliably be propagated by the high impact journals, conferences, and other high impact forums that are most widely read. In other words, already having GCI appears to be critical to solving the problem of understanding and communicating why GCI is critical in solving problems like AI alignment. If so, GCI is then currently outside the Overton window. The question is how to let it in. One potential way is an experiment that is currently being planned to explore the validity of the claim that GCI can exponentially increase our ability to solve the problem of achieving any collective outcome. Stay tuned.
Why the Solutions to AI Alignment are Likely Outside the Overton Window
The challenge with Western intellectual approaches to philosophy is that they rely on some framework for understanding the world, where each such framework precludes some subset of possibilities, and predisposes others, as opposed to more holistic Eastern approaches of considering each philosophy, ideology, or any other cognitive construct to be a function that generates its own outcomes, and just observing what each functions to achieve in every aspect of this existence. The benefit of Western intellectual approaches is that they are more readily understood and communicated, whereas Eastern approaches are notoriously difficult to understand and communicate without years of such observation, usually in the form of meditation or other practices. This makes them largely unavailable to Western science.
A recently developed approach called Human-Centric Functional Modeling (HCFM) however bridges these two approaches by representing the human systems through which we perceive and otherwise interact with the world as moving through graphs called functional state spaces, where these graphs can also hypothetically be used to describe the behavior of any living system, whether as simple as homeostasis or as complex as cognition. In these functional state spaces, all problems are defined as the lack of a path from one functional state to another, resulting in the insight that all adaptive processes through which nature solves problems are potentially generalizations of the same solution in functional state space. Representing the behavior of the cognitive system in terms of a set of paths through this mathematical space means that any property of that system’s behavior potentially can be given a mathematical definition, which would mean that properties ascribed to the human system within existential traditions, where those properties are only discernable through the first person observation generally rejected by as subjective and unscientific by Western science, might for the first time be objectively validated. Since many if not most properties of cognition can’t yet be observed by objective third party methods and tools, this would be a major step forward in our understanding.
An example of the usefulness of this approach is that it allows us to model the collective problem-solving ability of networks of systems in terms of their ability to navigate a given volume and density of the collective functional state space per unit time, which to date might be the only functional model of intelligence that has been hypothesized, and it allows us to see how networks create the potential to exponentially increase that ability. This in turn has led to a model for group decision-making called a General Collective Intelligence or GCI platform that copies this same solution. GCI is important because arguments can be constructed using this functional state space approach to support the prediction that GCI creates the capacity to exponentially increase the general problem-solving ability (intelligence) of groups.
An example of the value of a functional modeling methodology that facilitates merging these intellectual and existential approaches is an article here which concludes by saying “successful AI labs like DeepMind, OpenAI, and Anthropic do not seem to espouse this “pivotal act” philosophy for doing good in the world”. A functional state space based model of individual and group decision-making must be able to accommodate any possible interpretation of any reasoning and must be able to model the fitness of that reasoning at achieving its targeted outcome. This involves defining a “fitness space” for each individual cognition, as well as a collective fitness space for the group. Considering the basic problem that the individual fitness space is hypothesized to solve, as opposed to the problem that the collective fitness space solves, and considering the fact that GCI has not yet been implemented, and considering the prediction that collective fitness cannot reliably be optimized without GCI at group sizes above those characteristic of the ancestral tribes and villages humans evolved in (according to the collective social brain hypothesis), this suggests that the exact opposite is true. Namely, that this “pivotal act” and its accompanying philosophy might be the only reliably achievable outcome of trying to achieve AI safety through the current centralized methods, as opposed to doing so through a network based approach. In fact, this modeling approach suggests that the recent fear-mongering about the danger of AI and the recent petition to slow down the development of AI will most likely function to provide a competitive advantage to the signatories of this petition, and will in a number of ways further centralize control over AI, and therefore will function to make AI more dangerous.
The collective social brain hypothesis posits that individuals tend strongly to fall into either of two groups, one of which that tends to use type I (intuitive) reasoning processes on specific types of topics, and the other that tends to use type II (logical) reasoning processes on those same topics. The word “logical” here does not mean “correct”. It simply means some combination of reasoning that can be free of intuitive processes such as detecting what the consensus is. Both reasoning types are optimal in different problem domains. The challenge is that an understanding of the collective social brain hypothesis is required to understand the need for GCI to switch to the optimal reasoning type as required to achieve collective fitness, and experiments suggest that understanding the collective social brain hypothesis requires a non-consensus based reasoning process than cannot reliably be propagated by the high impact journals, conferences, and other high impact forums that are most widely read. In other words, already having GCI appears to be critical to solving the problem of understanding and communicating why GCI is critical in solving problems like AI alignment. If so, GCI is then currently outside the Overton window. The question is how to let it in. One potential way is an experiment that is currently being planned to explore the validity of the claim that GCI can exponentially increase our ability to solve the problem of achieving any collective outcome. Stay tuned.