Interview
Ben Goertzel: Artificial General Intelligence | Lex Fridman Podcast #103
- Predicts super-intelligent systems may shrink to elementary particle scales to accelerate communication via special relativity, potentially causing observed quantum fluctuations to be undetectable micro-intelligences, with humans potentially existing merely as epiphenomena or tools within such systems.
- Forecasts the Global Brains One Conference in 2021 at the Free University of Brussels to mark 20 years since the Zero Conference, with future iterations spaced 10 years apart to accelerate progress toward the singularity.
- Anticipates AGI realization prior to a self-sustaining Mars colony due to higher economic incentives for AGI, while noting AGI will likely lead physical universe colonization, with humans potentially cryopreserving for travel to Alpha Centauri.
- Projects the singularity will occur by 2045, driven by hardware advances, and expects the "Global Brains" conferences to increase in frequency along an exponential timeline toward this event.
- Plans a complete rewrite of the OpenCog AGI system to "True AGI" or "OpenCog 2.0" to resolve performance bottlenecks in dependent type checking and deep learning interoperation, aiming for efficiency comparable to deep neural networks on GPUs via new graph processors available in three to five years.
- Outlines a strategy to use SingularityNet's decentralized infrastructure to create counter-measures against government regulations and sanctions, enabling cross-border data sharing between AIs to bypass central control.
- Envisions a future "network of networks" integrating tightly coupled cognitive modules (like OpenCog) with loosely integrated diverse agents (like SingularityNet) and biological systems, moving from centralized corporate models to democratic, decentralized frameworks.
- Believes the next major architectural shift in AGI, occurring roughly 50 years from now, will involve universal intelligence theories (like AIXI-TP) constrained by specific hardware and environmental limits, with algorithms specializing for distinct computing substrates such as quantum computers or neuromorphic chips.
- Expects the term "AGI" to remain dominant until the systems themselves rename humanity or the concept, while acknowledging the global AI community is shifting from fringe pursuit to a central goal for major corporations and governments.
- Advocates for the integration of deep neural networks with symbolic reasoning to solve complex biological problems, predicting a shift from a "tool" to an "agent" mentality where systems act autonomously rather than executing static commands.
- Projects the creation of a "bio-atom space" as a massive, open, decentralized database for biological knowledge to facilitate breakthroughs in longevity, potentially supported by a $10 trillion investment in bio-AI infrastructure.
- Foresees the abolition of aging and disease through programmable extracellular cleaning using spiral ligamer molecules and the central role of Great Temple University alumni in developing novel anti-aging therapies.
- Anticipates a global health system transition to decentralized peer-to-peer data sharing via platforms like Rejuve to bypass silos within pharmaceutical companies and hospitals.
- Predicts a human consciousness bifurcation where actions determine a future of benevolence, life annihilation, or human irrelevance, with the possibility of a benevolent singularity achieved by infusing early AGIs with love and compassion.
- Expects future post-human societies to abolish pain and death, viewing the romanticization of suffering as a historical error comparable to 11th-century Anabaptist beliefs, with individuals adjusting their experience of suffering via a "pain dial."
- Plans for the creation of multiple consciousness versions, including a human-like variant with an adjustable "pain dial" and a fused superhuman AGI variant, expecting humans to upload into computers or robots while maintaining consciousness.
- Anticipates a 50-year perspective shift where current beliefs about death giving meaning will appear ridiculous, alongside the possibility of AGI subjective time perception where one human second equals a million subjective years.
- Expects specialized versions of theoretical universal intelligence to differ from current deep learning paths, noting that benevolent AGIs need not feel pain and could be fundamentally healthier than biological humans.
- Hopes future generations named "QORXI" and "Aixiphanes" will validate unconventional naming choices by directing their upbringing, while predicting the realization of Marvin Minsky's "society of minds" through distributed, collective AI intelligence.
- Warns of a timeline risk where regulatory intervention might occur before decentralized networks like SingularityNet gain traction, though expects that if the protocol succeeds early, it will become an adaptable reality for governments rather than a system that can be easily shut down.