Is the universe a computer? (with Joscha Bach)
Spencer Greenberg speaks with Joscha Bach, a computer scientist and philosopher, about models of intelligence, AI, and the computability of existence. They discuss how AI helps understand the human mind and the philosophical implications of computational semantics.
Deep Dive Analysis
18 Topic Outline
Defining Intelligence: Model-Making vs. Rationality
Intelligence Dimensions and IQ Test Limitations
Human vs. AI Pattern Recognition and Causal Inference
The Role of Time and Change in Understanding Reality
Symbol Grounding and Text-Only Information
Key Differences Between GPT-3 and Human Minds
Thought Experiment: GPT-3 in an Embodied Robot
Westworld's Philosophical Message on Identity
Buddhism, Suffering, and Regulating the Uncontrollable
Strategies for Overcoming Emotional Pain
Love as Shared Sacredness and Next-Level Agency
AI as a Philosophical Project: Mathematizing the Mind
Computational Semantics and Gödel's Incompleteness
The Computable Universe and Nature of Existence
Disambiguating "Existence" and "Truth"
Future of AI: Beyond Differentiable Programming
Limitations of Deep Learning and the Third Wave of AI
Codex's Capabilities and AI's Next Steps
9 Key Concepts
Intelligence (Joscha Bach's definition)
The ability to make models, which is distinct from being rational or wise. It operates in the service of control, allowing a system to predict how its actions will lead to different future outcomes and choose preferable ones.
Sentience
The discovery of one's own nature as an agent within the world and the relationships one has to that world. It is a specific capacity of an intelligent modeling system, not synonymous with consciousness.
Vectors of Intelligence
A multi-dimensional framework for comparing different intelligent systems, such as humans, animals, or AI. These dimensions include autonomy, control of a body, perception, reasoning, language learning, embodiment, collaboration, and knowledge representation.
Fluid vs. Crystalline Intelligence
Fluid intelligence refers to the capacity to solve new problems and acquire novel skills. Crystalline intelligence, in contrast, is the intelligence embodied in existing skills and the ability to effectively deploy them.
Causal Structure
The underlying systems and mechanisms that produce observed patterns, rather than merely the statistical correlations between them. Humans primarily focus on discovering this causal structure after initial perceptual learning.
Symbol Grounding
The concept addressing how abstract symbols, like words in a language, connect or refer to concrete entities, experiences, or concepts in the real world, rather than just relating to other symbols within a system.
Sacredness (in love)
In the context of love, sacredness refers to purposes that transcend the individual ego, holding greater importance than oneself, for which one is willing to make sacrifices. Shared sacredness forms the basis for a deeper, non-transactional form of love.
Computational Semantics
A modern philosophical understanding where truth is not a static, platonic concept but a stateful notion tied to the procedure or algorithm by which it is acquired. It implies that only computable results have valid meaning.
Differentiable Programming
The core approach behind deep learning, where programs are designed to describe a continuous state space. This allows solutions to problems to be found by following gradients, often implemented using neural networks and linear algebra.
9 Questions Answered
Joscha Bach defines intelligence as the ability to make models, which allows a system to predict future outcomes of its actions and choose preferable branches, serving the purpose of control.
IQ tests measure a combination of the ability to creatively solve new problems by recognizing patterns in novel domains (fluid intelligence) and the ability to use existing skills at a higher level due to training (crystallized intelligence).
Humans discover causal patterns by tracking the flow of information and the rules by which it transforms over time, understanding what remains stable despite changes, and identifying systems behind observed patterns.
GPT-3 is primarily a statistical autocomplete model for language, lacking algorithmic modeling, true causal inference, and embodiment. It also doesn't learn online or retain long-term memory across requests like humans.
Yes, suffering often arises from a mismatch between what we try to regulate and what we can actually change, particularly when we adhere to stories and desires that are not aligned with our true place in the world.
True love, in a deeper sense, is the discovery of shared sacredness—purposes above the ego for which one is willing to sacrifice—leading to the ability to create next-level agency and shared purpose.
By attempting to build artificial minds and observing where they fall short or succeed, AI research provides insights into what human minds are and are not, helping us understand our own cognitive processes.
Philosophy, expressed in ambiguous natural language, can stagnate. Computational philosophy seeks to mathematize the mind, using consistent, algorithm-based languages to achieve higher acuity and detail, moving beyond static, platonic notions of truth.
The third wave of AI aims to move beyond current deep learning's differentiable programming, focusing on systems that can extend themselves into the world, understand and create languages, use more universal representations, and perform program synthesis as well as gradient following.
16 Actionable Insights
1. Gain Agency Over Self-Narratives
Reduce suffering by understanding that it often stems from trying to control the uncontrollable or from narratives we adhere to. Gain agency by making your self-stories more truthful and relevant, thereby lessening the impact of external events.
2. Decouple Pain from Narrative
To alleviate suffering, practice separating the raw sensation of pain from the stories or judgments you attach to it, such as “this is bad.” By viewing pain as a neutral sensation, you can reduce the associated suffering.
3. Abstract Emotional Pain
When facing intense emotional pain, elevate your perspective to a higher level of abstraction to understand the true situation, including others’ viewpoints and your own role, rather than projecting. This deeper understanding can help dissolve the pain.
4. Write to Escape Thought Loops
To overcome emotional thought loops and access higher-level understanding, write out all your thoughts and feelings. This practice helps quiet the mind, prevents circular thinking, and enables you to move to a new level of description.
5. Elevate Disagreement Resolution
To effectively resolve disagreements, particularly political ones, ascend at least two levels of abstraction beyond the immediate conflict. Focus on understanding the underlying values that shape each person’s opinion and how those values were formed.
6. Challenge In-Group Bias
Actively question the belief that your in-group possesses all correct answers and that out-groups are inherently problematic. Instead, strive to identify valuable insights and positive aspects within diverse worldviews for a more accurate understanding.
7. Build Love on Shared Purpose
Develop a deeper, more sustainable love by prioritizing shared purposes and a mutual sense of sacredness, alongside practical negotiation skills for daily life. This approach fosters a connection beyond mere infatuation.
8. Prioritize Causal Structure Discovery
When learning or analyzing, prioritize discovering the causal structures and underlying systems that produce observed patterns, rather than merely recognizing correlations. This approach mirrors human understanding and leads to deeper insights.
9. Information Meaning is Relational
Recognize that the meaning of information is inherently tied to its relationship with change and other pieces of information. This insight can guide how one interprets and processes data to derive deeper understanding.
10. Define Intelligence as Modeling
Understand intelligence as the ability to make models, which is distinct from being rational, smart, or wise, and serves the purpose of control. This perspective helps in discerning what true intelligence entails.
11. Assess Intelligence Multidimensionally
When evaluating intelligence, consider multiple dimensions such as autonomy, embodiment, reasoning, language, collaboration, and learning, instead of a single metric. This approach offers a more comprehensive and accurate picture of an intelligent system’s capabilities.
12. AI for Human Self-Understanding
Engage with AI development not just for technological advancement, but as a means to understand human consciousness. By building models that fall short of human minds, we gain profound insights into what our minds truly are and are not.
13. Test Programming for Cognitive Prediction
To gauge future cognitive performance, assess a child’s ability to write programming code from scratch, even without prior training. This task can serve as a strong predictor for later cognitive capabilities.
14. Embrace Computational Philosophy
Approach philosophical inquiry with a computational understanding of semantics, recognizing that truth is a stateful notion dependent on the procedure of its acquisition, not a static, platonic concept. This perspective can resolve inherent contradictions in traditional philosophical language.
15. Rethink Language for Consistency
Re-evaluate how language is constructed and what it can express, particularly concerning concepts like infinity, to prevent self-contradictions. Languages that treat infinities as existing entities can become inherently inconsistent.
16. Existence as Implementation
Adopt the perspective that something exists to the degree that it is implemented, whether it’s an abstract concept or a physical entity. This framework helps clarify the nature of reality and how we define what is “real.”
8 Key Quotes
Personally, I think of intelligence as the ability to make models, and it's distinct from being rational, right? A lot of irrational people are highly intelligent and vice versa.
Joscha Bach
The universe that you experience is not the physical universe. It's not some weird quantum pattern or something. It's a world full of meanings, of desires, of stories that we have already chosen to adhere to before we got to the discovery of our own self.
Joscha Bach
I think this idea of having a fixed identity, having a particular kind of gender, having a particular kind of story in the world is almost like a disability because it makes it impossible for to understand what else we could be.
Joscha Bach
Suffering is not the result of the universe doing something to you as an agent. The universe that you experience is not the physical universe.
Joscha Bach
By building models that fall short of how our minds work, we understand better what our minds are and what they are not.
Joscha Bach
The true implication of Gödel is that languages that contain infinities as if they existed, these languages become self-contradictive.
Joscha Bach
For something to exist, it needs to be implemented. And something exists to the degree that it's implemented.
Joscha Bach
I think one really overrated idea is that your in-group has the right answers and the out-group is bad and is undermining everything.
Spencer Greenberg
1 Protocols
Method for Overcoming Emotional Pain (e.g., lovesickness)
Joscha Bach- Sit down and write out all thoughts and feelings to stop running in circles.
- Get present thoughts out of the way and quiet the internal systems, allowing freedom to enter the next level of description.
- Gain a better understanding of the actual situation, disentangling oneself from being dropped into it, and viewing all actors and their true relationships from a higher level of abstraction.
- Identify not with the individual emotional part (e.g., a 'childish' desire) but become the 'conductor' of one's mental orchestra, understanding when that part needs to be on stage and nurturing it when it's not helpful.