intelligence-overview.txt

### conscious machines

Michael Timothy Bennett's doctoral thesis, titled "How to Build Conscious Machines," proposes a comprehensive framework for understanding consciousness and its relationship with intelligence. The work is divided into several sections and chapters that cover various aspects of philosophy, computer science, neuroscience, and artificial general intelligence (AGI).

1. **Foreword and Chapter Summaries**: Bennett explains the progression of his research throughout the PhD and how it culminates in this thesis. The work is presented as an exploration into building conscious machines, understanding the nature of consciousness, and addressing fundamental questions about life, intelligence, and complexity.

2. **Literature Reviews**: Chapters II and III provide a survey of philosophy and neuroscience (Chapter II) and AGI (Chapter III). These sections delve into key concepts like the mind-body problem, functionalism, theories of consciousness, self-organization, free energy principle, enactivism, epistemology, semiotics, structuralism, post-structuralism, and meaning.

3. **What is AGI?** (Chapter IV): Bennett explains what constitutes artificial general intelligence by framing it as an "artificial scientist," drawing inspiration from Richard Sutton's 'Bitter Lesson.' He discusses various optimization approaches like search, approximation, and hybrids.

4. **Turtles All the Way Down** (Chapter V): This chapter explores embodiment by arguing that each body is an abstraction layer. It presents a formal language of declarative programs based on Stack Theory, allowing for the understanding of bodies and the universe as speaking embodied formal languages.

5. **Master, What Is My Purpose?** (Chapter VI): Bennett discusses purpose through formal definitions of embodied tasks, inference, and stacks. It establishes how goal-directed behavior emerges from time, change, and natural selection, leading to the concept of the "cosmic ought."

6. **Weak** (Chapter VII): This section introduces w-maxing as an optimal learning meta-approach that maximizes weak constraints on function. Bennett proves that w-maxing is superior to simp-maxing in generalization and intelligence, while also exploring how biological systems delegate adaptation more effectively than artificial intelligence.

7. **Stackism** (Chapter VIII): This chapter investigates complexity by demonstrating why simplicity of form has a correlation with function. It argues that the illusion of complexity arises due to abstraction layers, and natural selection drives biological systems to optimize weaker constraints using simpler forms.

8. **Let's Get Psychophysical** (Chapter IX): Bennett formalizes causality via the Psychophysical Principle of Causality, explaining how systems learn cause-and-effect relationships based on valence. He introduces self-classifying policies called causal-identities and discusses the necessity for incentive and scale to construct these identities.

9. **Language Cancer** (Chapter X): Bennett explores language and its connection to cancer, refuting the Orthogonality Thesis. He demonstrates how 2ND-order-selves are crucial for communication according to Gricean pragmatics and explains normativity in relation to social predation.

10. **Why Is Anything Alive?** (Chapter XI): Bennett discusses the emergence of life by presenting a formalism that ties together Pancomputational Enactivism, complexity theory, and the Fermi Paradox, arguing that stable environments allow for consciousness to develop due to the correlation between adaptability and weak constraints.

11. **Why Is Anything Conscious?** (Chapter XII): This chapter addresses the hard problem of consciousness by presenting a theory of how lower-order states give rise to higher-order thought through causal identities grounded in valence. Bennett introduces 1ST and 2ND-order selves, arguing that phenomenal consciousness begins with a 1ST-order self, while communication requires a 2ND-order self.

12. **How to Build Conscious Machines** (Chapter XIII): The final chapter outlines the features necessary for constructing conscious machines and proposes an unresolved problem called "The Temporal Gap." It also discusses strategies for engineering conscious machines or avoiding their creation altogether.

Throughout the thesis, Bennett integrates various results from his published papers, which explore topics like optimal learning, abstraction layers, computational dualism, complexity, language, meaning, and causality. The work aims to provide a foundation for understanding how conscious machines can be built by reconciling philosophical, neuroscientific, and computer science perspectives on consciousness and intelligence.


The text presents a philosophical exploration of artificial general intelligence (AGI) and the concept of computational dualism, which posits that software or 'mind' is separate from hardware or 'body'. The author argues against this view, asserting that both are interconnected aspects of a larger system.

The author introduces "Stack Theory," which suggests everything is nested abstraction layers, from software to hardware, and ultimately to the fundamental laws of physics. In this framework, hardware is not a sacred boundary where abstraction stops, but another layer in the hierarchy.

The environment, according to Stack Theory, is defined by a set of states (Φ), with each state representing a particular configuration or difference from other states. The power set 2^Φ (P) represents all possible subsets of states, which are called declarative programs. A truth or fact about a state ϕ is any program (f) containing ϕ, meaning f is true for that state.

The environment encodes everything through its state space, with each state representing an aspect of the environment as a collection of facts holding true for that state. This formal structure aligns with pancomputationalism's view that all physical systems are computational.

Toy examples illustrate the framework's flexibility across various domains, from digital systems (light switch) to biological systems (cell metabolism), and even reinforcement learning in AI. The author emphasizes the importance of embodiment—the idea that every physical system influences its surroundings—which is often overlooked in computer science.

Finally, the author discusses "Layer Cake," a method for formalizing all these aspects together within an abstraction layer (v) that contains more specific aspects. This involves defining an abstraction layer (Lv) as everything realizable within it, and the extension Ex of a statement x as the set of statements whose existence implies x.

In summary, the text challenges traditional computational dualism by proposing Stack Theory, which views software, hardware, and fundamental laws of physics as interconnected layers in an overall system. It further introduces the concept of "Layer Cake" for formalizing these aspects within abstraction layers, thereby providing a unified framework to understand intelligence and AGI.


In this section, Michael Timothy Bennett discusses the concept of distribution in adaptive systems, focusing on biological and artificial examples.

1. Definition of Distribution:
Distribution in an adaptive system refers to having more than one policy (a set of programs) expressed by an abstraction layer. This means that the system's behavior is not solely determined by a single entity but rather emerges from the collective actions of multiple entities working together towards common goals.

2. Cellular Collectives as an Example:
A collective of cells can serve as an example of distribution in biological systems. Each cell within the collective represents a policy, with its own set of programs that govern its behavior. When these cells work collaboratively toward shared objectives, their individual extensions (sets of states where each program is true) intersect to form a higher-level collective policy. This collective policy embodies the group's identity or behavior at a more abstract level than any single cell.

3. Implications for Artificial Systems:
This idea of distribution can also be applied to artificial systems, such as computer networks or multi-agent systems. In these contexts, multiple agents (policies) work together to achieve common goals by coordinating their actions and sharing information. The collective behavior emerges from the interactions among these agents, leading to a more complex and nuanced system than what could be achieved with a single agent alone.

4. Distinction Between Distribution and Delegation of Control:
It is essential to distinguish distribution from delegation of control in adaptive systems. While distribution refers to having multiple policies expressed by an abstraction layer, delegation of control involves assigning decision-making authority to specific levels or entities within the system. In other words, distribution focuses on how work is divided among various components, whereas delegation determines who makes decisions and to what extent those decisions can be made autonomously.

5. Formalization of Distribution:
To formalize the concept of distribution, Bennett suggests considering a set of policies Lv within an abstraction layer. When multiple entities (cells, agents) in the system express these policies simultaneously, their collective behavior emerges as the intersection of their extensions. This collective policy then forms a higher level of abstraction, representing the system's overall behavior or identity.

In summary, distribution is a crucial aspect of adaptive systems, both biological and artificial. By allowing multiple entities to work together and express policies concurrently, complex behaviors can emerge that might not be achievable through individual agents alone. Understanding and leveraging distribution can help design more efficient and resilient adaptive systems across various domains.


In this section, Michael Timothy Bennett discusses the concept of "language cancer" as it relates to his theory on conscious machines and collective intelligence. He begins by emphasizing that human communication involves more than just the exchange of information; it also includes normativity or social mores. These shared expectations guide interactions and foster cooperation, but they can also be manipulated for deceptive purposes.

Bennett introduces the idea of "protosymbols," which are learned causal identities that represent aspects of an organism's environment relevant to its survival. A protosymbol system is a set of tasks based on these learned causal identities, and preferences help an organism interpret inputs according to its knowledge and values.

The author argues that accurate prediction and hard-wired behaviors enable organisms to communicate intentions effectively. By using their second-order selves (predictions about other entities' predictions), they can tailor communication to individual recipients, allowing for nuanced meaning exchange. This predictive machinery can also be exploited for manipulation or deception if one entity gains sufficient insight into another's thought processes.

Bennett then discusses the role of social norms in shaping human behavior and communication. These shared expectations allow us to navigate complex social structures efficiently, reducing the need for constant negotiation and trust-building on an individual basis. In this context, language and concepts emerge as policies that govern how a population interprets information both internally and externally.

The author connects these ideas to cancer biology, suggesting that cancer arises when cells lose their collective identity due to isolation from the broader informational structure of the organism. In a distributed system, over-constraint or adversity (fewer correct policies) can cause parts to break away and pursue independent goals, leading to system failure.

Bennett proposes that similar processes may underlie "language cancer" in human populations – when a shared identity weakens due to excessive top-down control or other adversities, leading to the dissolution of coherent language and norms. This can result in stagnation, loss of collective identity, and potential system failure, similar to how biological self-organizing systems develop cancer when their informational structure collapses.

To prevent such a breakdown, Bennett suggests the need for "sloppy fitness" or loose constraints that allow for shared language, meaning, ethics, and norms. In artificial intelligence, this translates to a delegated and scale-free approach to alignment, balancing top-down control with bottom-up adaptation while ensuring sufficient constraints are in place.

Finally, Bennett argues against the orthogonality thesis, which posits that intelligence and goals are independent. He demonstrates how intelligence is intrinsically linked to embodiment (goal direction), thus making it goal-dependent. This insight has implications for artificial general intelligence, suggesting that tailoring AI systems solely around legal and moral boundaries might be overly restrictive – instead, a holistic approach considering internal functioning and interacting systems could yield more robust and adaptable solutions.


Title: "How to Build Conscious Machines" by Michael Timothy Bennett (Preprint under Review)

Summary:

This preprint presents a comprehensive theory on the nature of consciousness and provides insights into how conscious machines can be constructed. The author, Michael Timothy Bennett, proposes a framework called Pancomputational Enactivism within Stack Theory to explain goal-directed behavior in terms of tasks.

1. Stack Theory:
The environment is conceptualized as an infinite stack of abstraction layers, with the cosmic ought (the driving force for self-preservation) at the bottom and higher levels of abstraction emerging from lower ones.

2. Pancomputational Enactivism:
This meta-approach to Artificial General Intelligence (AGI) integrates computational processes with enactive principles, where consciousness arises from the interaction between an organism and its environment. AGI is seen as a form of polycomputation across multiple abstraction layers.

3. Weak Constraints on Function:
Bennett proposes that simple forms are not sufficient for adaptation; instead, weak constraints on function are necessary and sufficient for generalization and adaptation. This undermines Ockham's Razor, leading to Bennett's Razor – favoring the simplest explanation with the fewest assumptions.

4. W-maxing:
Bennett introduces a new meta-approach called w-maxing, which delegates control to the lowest level of abstraction while satisfying correctness constraints. This principle leads to The Law of the Stack, stating that adaptation improves as systems delegate more control to lower levels of abstraction.

5. Learning Causality:
The author explains how adaptive systems learn causality by learning objects and properties causing valence (i.e., attraction or repulsion). This process is tied to the Psychophysical Principle of Causality, which states that consciousness simplifies our perception of the environment into relevant objects and properties.

6. Construction of Selves and Phenomenal Consciousness:
Bennett discusses the emergence of selves and phenomenal consciousness through a causal framework. He relates this to language and semiotics, formalizing Gricean pragmatics and Peircean triadic symbols as tasks. This leads to an alternative definition of access consciousness: the contents of 2nd and higher-order selves, making philosophical zombies impossible in any conceivable world.

7. Emergence of Normative Meanings:
The author explains how normative meanings emerge from collective identity, cancer (metaphorically), and his Mirror Symbol Hypothesis – which posits that symbols are learned by identifying patterns in the environment and attributing them with meaning. This process refutes the strong orthogonality thesis, suggesting goals and intelligence are intrinsically linked.

8. The Temporal Gap:
The author discusses an unknown (referred to as "The Temporal Gap") concerning whether a conscious state must be realized by an environmental state at a single point in time or can be smeared across time. This has implications for the possibility of software consciousness, suggesting that current AI systems lack the necessary features (e.g., delegated control, persistent structure) to support a tapestry of valence and phenomenal consciousness.

9. Conclusion:
Bennett asserts that intelligence is both necessary and sufficient for consciousness and outlines features required for constructing conscious machines. He suggests that to build a truly conscious machine, we should aim for a highly delegated solid brain where tapestries of valence are realized at a single point in time rather than smeared across it.

The preprint is supported by mathematical definitions, proofs, experiments, and examples, which are available on GitHub: https://github.com/ViscousLemming/Technical-Appendices


The provided list is a bibliography containing references to various books, articles, and preprints related to artificial intelligence (AI), cognitive science, philosophy of mind, consciousness, and the nature of computation. Here's a detailed summary and explanation of each section:

1. Bas C. van Fraassen - "Laws and Symmetry": This book explores the philosophical aspects of symmetry in physics and its implications for understanding scientific laws. It questions the traditional view that symmetries are merely mathematical properties, suggesting that they have a deeper role in our understanding of nature.

2. Multiple authors - "APA Newsletters" (2008): This entry likely refers to a collection of articles from the Association for Psychological Science's newsletter. The topics discussed may include recent advancements, debates, and notable research findings in psychology and related fields.

3. Michael T. Bennett - "How to Build Conscious Machines" (preprint under review): This preprint is an unpublished manuscript by Michael T. Bennett that aims to discuss the challenges and possible approaches for creating conscious machines, potentially blending aspects of AI, cognitive science, and philosophy of mind.

4. Sarah A. Fricke and Christina M. Frederick - "The Looking Glass Self: The Impact of Explicit Self-Awareness on Self-Esteem": This study investigates the relationship between self-awareness and self-esteem, drawing from Daryl Bem's concept of the looking-glass self, which suggests that individuals form their self-perception based on how they believe others view them.

5. M. Friedman and R.D. Friedman - "Capitalism and Freedom": This classic work by Milton Friedman argues for a free market economy as an ideal system, emphasizing the role of individual liberty and competition in creating prosperity while minimizing government intervention.

6. Karl Friston - "The Free-Energy Principle: A Unified Brain Theory?" (Nature Reviews Neuroscience, 2010) and "Life as We Know It" (Journal of The Royal Society Interface, 2013): These papers propose the free-energy principle as a unifying framework for understanding brain function. Friston suggests that the brain minimizes its free energy by constantly generating predictions about the world, comparing them to sensory input, and adjusting its internal models accordingly.

7. Karl Friston et al. - "Path Integrals, Particular Kinds, and Strange Things" (Physics of Life Reviews, 2023): This article expands on Friston's previous work by exploring the implications of path integrals – a concept from quantum mechanics and statistical physics – for understanding brain function, learning, and decision-making.

8. Thomas Fuchs - "Ecology of the Brain: The Phenomenology and Biology of the Embodied Mind" (Oxford University Press, 2017): This book presents a comprehensive theory of consciousness that emphasizes the importance of embodied cognition – the idea that mental processes are deeply influenced by bodily experiences and interactions with the environment.

9. Shaun Gallagher and Dan Zahavi - "The Phenomenological Mind" (Routledge, New York, NY, 2014): This work provides an introduction to phenomenology as a philosophical approach to understanding the mind, focusing on first-person perspectives and lived experiences.

10. Ashitha Ganapathy and Michael T. Bennett - "Cybernetics and the Future of Work" (2021 IEEE 21CW): This paper discusses how cybernetic principles can inform our understanding of emerging technologies' impact on employment, particularly in the context of artificial general intelligence (AGI).

11. Robin Gandy - "Church's Thesis and Principles for Mechanisms" (The Kleene Symposium, 1980): In this paper, Gandy examines Church's thesis – the idea that any effectively calculable function can be computed by a Turing machine – and its implications for understanding the nature of computation.

12. A. Garcez et al. - "Neural-Symbolic Computing: An Effective Methodology for Principled Integration of Machine Learning and Reasoning" (2019): This article explores neural-symbolic computing, a multidisciplinary approach combining machine learning techniques with symbolic reasoning to create more flexible and interpretable AI systems.

13. Marta Garnelo et al. - "Towards Deep Symbolic Reinforcement Learning" (2016): This paper proposes a method for integrating deep neural networks with symbolic representations, aiming to improve the interpretability and generalization abilities of reinforcement learning algorithms.

14. James J. Gibson