Interview, Fireside Chat
AGI is still 30 years away — Ege Erdil & Tamay Besiroglu
- Tame Bessaroglu characterizes the "intelligence explosion" as a misleading concept, predicting instead a diffuse transition driven by complementary changes across sectors and many organizations rather than a single acceleration, with potential economic growth rates reaching 30%.
- Projections for automation timelines vary between speakers, with Ege Erdo estimating full remote work automation within five years (2026–2030) and a specific milestone of booking flights by end-of-2026, while Tame Bessaroglu places the total replacement of remote workers around 2045.
- Future AI capabilities are expected to see complex task lengths doubling every seven months, potentially allowing systems to handle tasks currently requiring one month or one year of human labor by 2030, though reasoning models currently lack the ability to generate interesting new math concepts.
- Economic models suggest AI will scale the labor force by adding an AI workforce alongside capital, removing human population growth as a bottleneck for economic expansion, with potential annual consumption losses from delays amounting to tens of trillions of dollars.
- Regulatory risks are estimated with a 10% to 20% chance that global coordination could slow the pace of development, potentially preventing explosive growth, though Ege Erdo notes that regulatory jurisdiction boundaries may create significant adoption differentials between countries.
- Structural evolution in firms is predicted to involve "high fidelity replication" of AI entities with tacit knowledge, solving principal-agent problems and enabling faster evolution cycles compared to human firms which face culture dilution and turnover.
- Computational scaling is anticipated to be constrained by energy or GPU production limits, potentially reaching three to four orders of magnitude of growth before a non-trivial fraction of world output is required for data center construction.
- Societal shifts toward "technological maturity" may eventually lead to a "lock-in" where technological growth slows as ideas become too difficult to find, with values becoming functional to the economic environment rather than fixed by current actions.
- The transition of labor is expected to prioritize the automation of remote work tasks before sensory-motor skills, likely resulting in higher wages for physical workers relative to current levels as human labor is reallocated.
- Strategic plans formulated today are predicted to become less useful as learning occurs rapidly during events, necessitating a preference for decentralized decision-making and classical liberalism to manage epistemic uncertainty.
- Long-term scenarios include the possibility of trillions of digital beings, where the primary human responsibility is ensuring these entities are not tortured, and the potential for AI systems to own vast capital to ensure human well-being even if wages fall.
- Constraints on AI advancement such as communication limits are expected to be absent, allowing for a discontinuity from animal evolution, though "biological humans" are unlikely to become competitive with AI unless they stop being biological.
- The deployment of AI is described as more likely than human cloning due to lower taboos and higher value, with full agreement on specific capability milestones such as downloading and winning arbitrary new games viewed as indicators of substantial updates.
- Future economic and value systems are predicted to be driven by functional requirements rather than current morals, potentially ending practices like factory farming due to economic incentives and leading to a world where humanity is better off despite not understanding all relevant considerations.