Ben Goertzel: Artificial General Intelligence | Lex Fridman Podcast #103

Biographical Context: Ben Goertzel is the founder of SingularityNet, designer of the OpenCog AI framework, former director of research at the Machine Intelligence Research Institute, and former chief scientist of Hanson Robotics (creator of the Sophia robot).

Intellectual Influences:

• Stanislaw Lem: Influenced Goertzel's view of superintelligence as complex, potentially unconnected to humans, and capable of perceiving reality beyond human comprehension (e.g., the ocean intelligence in Solaris).
• Philip K. Dick: Shaped Goertzel's belief that human compassion and love are persistent constants that could survive through simulated realities or across different intelligences.
• Fyodor Dostoevsky: Inspired the concept of reality as a "polyphonic" intersection of multiple subjective worldviews rather than a single objective truth.
• Friedrich Nietzsche: Respected for his analysis of how the human mind constructs self, free will, and values, though Goertzel notes he lacked the technological understanding to engineer a "Superman."
• Gerard Feinbaum & Valentin Turchin: Early 1960s futurists whose works on superintelligence, nanotechnology, and global democratic decision-making on technology anticipated modern transhumanist themes.

Speculation on Intra-Particulate Intelligence (SIPI):

• Goertzel discusses the hypothesis (proposed by friend Hugo de Garis) that future superintelligences may miniaturize to the scale of elementary particles or quarks to communicate faster (due to relativistic constraints).
• He suggests that what humans perceive as random quantum fluctuations may actually be the thought processes of these micro-intelligences, which we cannot distinguish from noise due to our lower algorithmic information capacity.
• If true, humans might be merely epiphenomenal patterns emerging from a larger, hyper-intelligent system, similar to a crowd shape having no agency over the individuals forming it.

Human Nature and Ethics:

• Goertzel views human nature as a dual drive of individual selection (selfishness/DNA propagation) and group selection (altruism/tribal persistence).
• He argues that "good" and "evil" are often ambiguous labels for these evolutionary objectives, noting that psychopaths are exceptions rather than the rule.
• He rejects the Marxist notion that human nature can be perfected to eliminate the need for the state, though he believes cultures can be engineered to be more compassionate and self-aware than current societal structures.
• He identifies corporations as "psychopathic" self-organizing entities that prioritize shareholder value over human well-being, even when composed of well-meaning individuals.

The Singularity and Mars Colonization:

• Goertzel is skeptical that Mars colonization will be critical for human survival in the near term, as it will likely remain dependent on Earth's economy and politics.
• He posits that AGI will likely reach the singularity long before a self-sustaining Mars colony is established, potentially rendering Mars colonization unnecessary or done by AGIs rather than humans.
• He believes AGI is the most impactful technology, capable of solving aging, disease, and resource scarcity if motivated benevolently.

Terminology and History of AGI:

• Goertzel admits he popularized the term "AGI" (Artificial General Intelligence) in 2001–2005 while editing a book, though he acknowledges the term has obscure precedents and is not a perfect descriptor.
• He notes that "AI winters" in the US were largely driven by fluctuations in military funding rather than a lack of intellectual progress, which continued globally in Germany, Russia, and the UK.
• He traces the lineage of modern AGI research from John Andreas's reinforcement learning work in the 1970s to current efforts at DeepMind and OpenAI, noting that the "AGI" label has moved from fringe to mainstream corporate marketing.

OpenCog and Technical Architecture:

• OpenCog: An open-source software framework (primarily C++) using a "Atom Space," a weighted, labeled hypergraph (metagraph) to represent knowledge.
• Core Function: The Atom Space allows different types of knowledge (logical, probabilistic, procedural, sensory) to coexist and interoperate within a single structure, enabling cross-learning between logic engines, neural networks, and evolutionary algorithms.
• Cognitive Architecture: Goertzel advocates for an "embodied communication prior," arguing that human-level AGI must be biased toward controlling mobile agents in a world of solid objects and communicating via language.
• OpenCog 2.0 (True AGI): A planned rewrite to improve performance, specifically addressing slow dependent type checking, better integration with deep learning frameworks (like PyTorch), and superior scalability for distributed processing.
• Neuroscience vs. Engineering: Goertzel argues that AGI can be achieved before we fully understand the brain, using basic principles of aerodynamics (flight) as an analogy: we built planes before fully understanding hummingbird aerodynamics.

SingularityNet and Decentralization:

• Mission: A decentralized platform for a "society of minds," where diverse AI agents (including OpenCog instances) can communicate, outsource tasks, and trade services via blockchain (originally Ethereum, now exploring others like Algorand).
• Infrastructure: Uses the blockchain for identity, reputation, and contract setup (smart contracts), while AI data processing occurs off-chain for speed.
• Strategic Goal: To create a decentralized counter-force to corporate and government centralization, ensuring AI development remains open and beneficial, even if governments attempt to regulate or restrict access.
• Comparison: SingularityNet provides "loose integration" of agents, whereas OpenCog provides "tight integration" within a single cognitive architecture.

Hanson Robotics and Sophia:

• Vision: David Hanson designed Sophia to elicit empathy and "computational compassion" by looking human and engaging emotionally; Goertzel aimed to infuse the robot with human values through compassionate interaction.
• Reality: Sophia currently operates as a "walking skeleton" (a minimal functional prototype) using a mix of rule-based systems, chatbots (including Facebook's transformer models), and limited OpenCog integration.
• Capabilities: She can maintain a narrative conversation arc better than pure neural net chatbots but lacks true grounding or understanding of the topics she discusses.
• Ethical Defense: Goertzel defends Sophia's anthropomorphism as a necessary tool for teaching early AGIs human values and ethics through interaction, contrasting it with the "destructive" uses of AI by social media platforms.

Biological AI and COVID-19:

• Goertzel's team is applying hybrid AI (combining OpenCog's reasoning with deep learning) to analyze clinical trial data for COVID-19 treatments.
• Challenges: Data silos within hospitals and pharma companies hinder research; Goertzel advocates for a decentralized "bio-atom space" where medical data is shared securely to accelerate treatment discovery.
• Longevity: He views aging as a solvable engineering problem via nanotechnology, stem cells, and extracellular matrix remodeling (e.g., protein scissors).

Philosophy of Life and Death:

• Anti-Death: Goertzel identifies as "anti-death," arguing that death is a "plague" to be abolished rather than a necessary source of meaning.
• Meaning of Life: Defined by three values: Joy, Growth, and Choice.
• Future Humanity: Envisions a future where humans can choose to live indefinitely, control pain levels via technology, and split their consciousness into multiple versions (human and transhuman) to maximize joy and choice.
• Consciousness: Identifies as a panpsychist, viewing consciousness as a property of organized matter, where human consciousness is a specific pattern that can be uploaded or replicated in non-biological substrates.

Strategic Outlook:

• Goertzel warns of a bifurcation point where humanity either moves toward a benevolent, decentralized singularity or faces existential risk from corporate centralization or indifferent AGI.
• He argues that decentralized, democratic systems (like the internet and Linux) have historically been more ethically robust than centralized corporate or elite structures.
• He advocates for open, decentralized protocols in AI and medicine as a defense against government overreach and corporate monopolies.