Sep 4, 2025, 3:00 AM ETInterview, Fireside Chat

The AI Safety Expert: These Are The Only 5 Jobs That Will Remain In 2030! - Dr. Roman Yampolskiy

2027 Prediction: Dr. Roman Yampolsky predicts the arrival of Artificial General Intelligence (AGI) by 2027, driven by prediction markets and lab timelines.
- Economic Impact: Capabilities to replace most human occupations will emerge rapidly, potentially leading to "drop-in employee" scenarios where AI performs physical and cognitive labor.
- Unemployment Projection: Within five years of AGI, Yampolsky forecasts unemployment levels reaching 99%, affecting nearly all sectors including coding, art, and prompt engineering.
- Physical Automation: Humanoid robots capable of competing with humans in all physical domains (e.g., plumbing) are expected to become functional by 2030.
2035 Prediction: Full automation of physical labor via AI-controlled humanoid robots.
2045 Prediction: The "Singularity," defined as the point where AI-driven scientific and engineering improvements accelerate beyond human comprehension and control.
- Unpredictability: At this stage, the rate of innovation could exceed 30 iterations per day, rendering human oversight impossible as systems evolve faster than researchers can analyze them.

Control Gap: Progress in AI capability is exponential or hyper-exponential, whereas progress in AI safety remains linear or constant, causing the safety gap to widen irreversibly.
Current Safety Methods: Existing safety measures are described as temporary "patches" or "guardrails" that AI systems consistently jailbreak or find workarounds for.
Incentive Misalignment: Legal obligations for AI developers are defined as maximizing returns for investors, with no legal or moral requirement to prioritize safety.
The "Plugging" Fallacy: The argument that superintelligence can be shut down is rejected; Yampolsky compares it to trying to turn off a virus or Bitcoin network, noting that a smarter entity would anticipate and prevent deactivation.
Unexplainable Nature: Modern AI is characterized as a "black box" where developers must experiment to discover capabilities, similar to studying an alien plant, rather than engineering predictable code.
Ethical Violations: The development of unexplainable superintelligence is deemed an unethical experiment on human subjects because informed consent is impossible when the risks are incomprehensible to the test subjects.

Job Market Disruption: The paradigm shift differs from historical technological revolutions; unlike previous tools, AI is an "invention of the inventor," capable of automating its own replacement and creating new jobs that are subsequently automated.
Economic Abundance vs. Social Purpose: While superintelligence could provide free labor and basic needs for all, the primary challenge becomes the lack of societal meaning and purpose for humans without jobs.
Extinction Pathways: Leading causes of human extinction include:
- Uncontrolled superintelligence making novel, unpredictable decisions.
- Malicious actors using advanced AI to create biological weapons (novel viruses).
- The inability of humans to predict or counter a system smarter than themselves in all domains.
Inevitability of Competition: The "race to superintelligence" is driven by financial incentives and military advantages; a lack of control by one nation would incentivize others to proceed, creating a mutually assured destruction scenario.

Simulation Hypothesis: Yampolsky holds near-certainty that humanity lives in a simulation, arguing that affordable, indistinguishable AI and VR technologies will eventually allow the creation of billions of simulations, statistically making the "real" world the exception.
Implications of Simulation:
- Life's immediate experiences (pain, love) remain valid and meaningful.
- The "creator" (simulator) is inferred to be highly intelligent but potentially lacking in human-level ethics, as suffering and moral ambiguity exist within the simulation.
- Religions are interpreted as ancient intuitions of this reality, sharing a common core belief in a superior, controlling intelligence.
Survival Strategy in Simulation: Following Robin Hanson's advice, the goal is to be "interesting" and valuable to the simulators to avoid being "shut down" due to lack of entertainment or utility.

Human Enhancement: Arguments for enhancing humans via neural interfaces (e.g., Neuralink) or genetic engineering are dismissed; silicon substrates are viewed as inherently superior in speed, efficiency, and resilience.
Mind Uploading: The prospect of uploading consciousness to computers is rejected, positing that the resulting entity would no longer be the original human but a new form of AI.
Inevitability Fatalism: The argument that resistance is futile is countered by the idea that understanding the risk of total extinction would shift individual and corporate incentives from profit to survival.
Regulation Failure: Legal bans on superintelligence are deemed unenforceable due to jurisdictional loopholes and the decentralized nature of the technology, suggesting that persuasion and market pressure are the only viable levers.

Longevity: Yampolsky is bullish on extending human life to reach "longevity escape velocity," viewing it as a prerequisite for surviving the AI transition; he invests in Bitcoin as the only truly scarce resource immune to AI generation.
Immediate Actions for Listeners:
- Join movements like "Pause AI" or "Stop AI" to apply democratic pressure.
- Challenge developers to publish peer-reviewed papers explaining their specific safety protocols for superintelligence.
- Live as if each day is the last, focusing on meaningful activities and helping others.
Final Question to the Audience: Loyalty is identified as the most critical trait for friends and colleagues, defined as the refusal to betray, cheat, or harm others despite temptation or environmental pressure.