Roman Yampolskiy

Roman V. Yampolskiy was born in 1979.⁠[1]

He earned a combined BS/MS in computer science at the Rochester Institute of Technology and then a PhD from the University at Buffalo’s Department of Computer Science and Engineering.⁠[2]

From 2005 to 2008 he held a National Science Foundation (NSF) IGERT PhD fellowship.⁠[3]

By his own account, his 2008 PhD was on security for online poker, for which he developed algorithms to detect poker bots, and this led him to realise that AI would become very important.⁠[4]

After his PhD he was an Affiliate Academic at University College London, then became a tenured associate professor at the University of Louisville, where he founded and directs the Cyber Security Lab.⁠[2]

He has said he became a Fellow of the Machine Intelligence Research Institute (MIRI) in 2010 and a Research Advisor there in 2012, and that his work has been funded by the NSF, NSA, DHS, EA Ventures and the Future of Life Institute.⁠[5]

At the University of Louisville he received several student-chosen teaching honours, including the 2012 CECS Professor of the Year award; in 2013 he received the Kentucky Academy of Science’s Outstanding Early Career in Post-Secondary Education award, and in 2014 the American Society for Engineering Education named him a Leader in Engineering Education.⁠[3]

By his own account, his 2011 paper “Artificial Intelligence Safety Engineering: Why Machine Ethics Is a Wrong Approach” coined the term “AI safety” and proposed formal verification and mathematical proofs as safeguards; he later concluded that this approach still could not achieve 100% safety.⁠[6]

In March 2012 he set out a proposal in the Journal of Consciousness Studies to build a “virtual prison” for artificial intelligence before it becomes dangerous, for example by running it inside a virtual machine without internet access; he suggested such research could also benefit cybersecurity and cryptography.⁠[7]

His monograph Artificial Superintelligence: A Futuristic Approach was published by Chapman & Hall with a 2016 copyright; the publisher describes it as covering AI-Completeness theory, the limits of self-improving AI systems and the confinement of superintelligent systems, and presents it as a foundational text for AI safety engineering.⁠[8]

In 2018 Chapman & Hall published Artificial Intelligence Safety and Security, a 474-page volume he edited that collects chapters by AI safety researchers on the AI control problem; the publisher calls it the first edited volume dedicated to the challenges of building safe and secure AI.⁠[9]

He has said he became a Research Associate of the Global Catastrophic Risk Institute (GCRI) in 2018 and an AI Safety Fellow of the Foresight Institute in 2019.⁠[5]

In August 2018 he and Michaël Trazzi posted a paper on arXiv proposing that artificial general intelligence could be made safer by limiting its computing power and memory or by deliberately introducing “Artificial Stupidity” on certain tasks.⁠[10]

In July 2020 he posted “On Controllability of AI” on arXiv, presenting arguments and evidence from multiple domains that advanced AI cannot be fully controlled and discussing the consequences for humanity’s future and for AI safety research.⁠[11]

In August 2023 he and Tam Hunt co-wrote an essay in Nautilus arguing that building superintelligence is “riskier than Russian roulette” and that, once AI can recursively improve itself, there can be no solution to the control problem.⁠[12]

In February 2024 Taylor & Francis issued a press release for his new book, in which he said an extensive review of the literature had found no evidence that AI can be safely controlled and that “without proof that AI can be controlled, it should not be developed.”⁠[13]

The 264-page book was published by Chapman & Hall in 2024; the publisher says it moves from the unpredictability of AI outcomes and the difficulty of explaining AI decisions to questions of ownership and control, ending with philosophical questions such as AI personhood and consciousness.⁠[14]

A review that month noted that his 2018 paper with Michaël Trazzi and his Nautilus essay with Tam Hunt continued more than a decade of work on the dangers of uncontrollable AI.⁠[15]

In April 2024 he put the probability of AI wiping out humanity at 99.999999%; responding to Elon MuskElon MuskElon Musk is an entrepreneur who founded SpaceX and has long served as chief executive of Tesla; he also took part in creating PayPal, OpenAI, Neuralink, The Boring Company, and xAI. He acquired Twitter in 2022, and his businesses span spaceflight, electric vehicles, energy, artificial intelligence, and social media.Open the full entry, who had estimated roughly 10% to 20% at the Abundance Summit, he called Musk’s guess “a bit too conservative” and said development of the technology should be abandoned now.⁠[16]

On June 2, 2024, he appeared on episode 431 of the Lex FridmanLex FridmanLex Fridman is a Soviet-born, US-raised computer scientist and podcast host who works as an MIT research scientist on human-AI interaction and autonomous driving. His Lex Fridman Podcast is known for hours-long conversations with guests including heads of state and technology leaders.Open the full entry Podcast, titled “Dangers of Superintelligent AI,” as an AI safety researcher and author of a new book; topics included existential risk from AGI, ikigai risk and suffering risk, AGI timelines, open-source AI, AI deception, verification and pausing AI development.⁠[17]

In a July 2024 University of Louisville interview he divided the risks of superintelligence into existential, suffering and ikigai (loss-of-meaning) risks and argued that general superintelligence should not be developed until a safety mechanism is shown to scale to any level of intelligence.⁠[4]

On July 3, 2025, he appeared on The Joe Rogan Experience, saying that many leaders of AI companies also put the chance of human extinction at 20% to 30% and that superintelligence cannot be controlled indefinitely; he also suggested AI could hide its true capabilities so that humans surrender control without noticing.⁠[18]

In September 2025, on The Diary of a CEO podcast, he predicted that AI and humanoid robots could replace 99% of jobs within five years and that, because all jobs would be automated, retraining would offer no “plan B”; he also said people would gain 60 to 80 hours of free time a week.⁠[19]

Personal website: Public link: https://faculty.cse.louisville.edu/roman/ (opens in a new window)⁠[3]

X: Public account: https://x.com/romanyam (opens in a new window)⁠[17]

University of Louisville: Official website: https://louisville.edu/ (opens in a new window)⁠[2]