The Peter McCormack Show

chevron_right

#174 - Roman Yampolskiy - We Are All Agents Inside a Simulation

whatshot 13 snips

May 12, 2026

Guest

Roman Yampolskiy

Roman Yampolskiy, AI safety researcher and computer scientist who studies risks from superintelligence and simulation theory. He argues reality might be artificial. Short, punchy conversations cover simulation evidence, quantum and cosmic-rendering ideas, whether we can escape or be tested, and why building uncontrollable AGI could threaten civilisation.

01:20:29

forum

Ask episode

web_stories

AI Snips

view_agenda

Chapters

auto_awesome

Transcript

info_circle

Episode notes

insights

INSIGHT

You Can Never Prove Reality Is Real

Roman states you can never prove reality is 'real' because no test gives certainty; you can only identify fakes.
He frames reality as relative to the observer: a character in an 8-bit game has its own authentic reality.

insights

INSIGHT

Physical Constants Look Like Rendering Limits

Roman interprets physical constants like light speed as computational limits: the speed of light equals the simulator's rendering/update rate.
He uses this to argue quantum oddities fit 'digital physics' explanations.

volunteer_activism

ADVICE

Avoid Building General Superintelligence

Roman advises we should not build general superintelligence and instead build narrow domain tools tailored to specific problems.
He emphasizes narrow agents can cure cancer or solve focused tasks without becoming uncontrollable general agents.

Get the Snipd Podcast app to discover more snips from this episode

Why Roman Believes We're Likely in a Simulation

01:26 • 1min

chevron_right

Simulation as Experimental Testbed

02:45 • 2min

chevron_right

Why This Historical Moment Matters

04:30 • 60sec

chevron_right

Simulation vs. Reincarnation Comparison

05:30 • 1min

chevron_right

Possible Purposes of the Simulation

06:46 • 35sec

chevron_right

Memory, Restarting, and AI Analogs

07:21 • 46sec

chevron_right

Why Pain and Love Matter Inside Simulations

08:08 • 2min

chevron_right

Detecting Consciousness and Rights for Agents

09:50 • 58sec

chevron_right

Defining Reality and Its Uncertainty

10:48 • 6min

chevron_right

Quantum Physics as Digital-artifact Evidence

Big Bang as System Start and Time Outside Time

18:39 • 3min

chevron_right

Can We Escape or Leak Out of a Simulation?

21:41 • 3min

chevron_right

Post-Singularity Possibilities Beyond Our Imagination

24:13 • 54sec

chevron_right

Disappearances, Conspiracies and SimCity Analogies

25:07 • 1min

chevron_right

Defining Intelligence and the Core AI Risk

26:21 • 6min

chevron_right

Roman's Recommendation: Don't Build AGI

32:50 • 2min

chevron_right

Why Companies Race for Superintelligence

34:30 • 2min

chevron_right

History of Safety Teams and 'Safety Theater'

36:11 • 2min

chevron_right

Limitations of Rule-based Alignment

38:11 • 2min

chevron_right

Risks vs Benefits and Narrow AI Alternatives

40:17 • 53sec

chevron_right

Refuting 'If We Don't Someone Else Will' Argument

41:10 • 2min

chevron_right

Roman's P(doom) and Uncontrollability Concerns

42:50 • 5min

chevron_right

Early Agent Experiences and Personal Use

47:45 • 1min

chevron_right

Different Models, Censorship and 'Filter' Limits

49:10 • 2min

chevron_right

Limits of Interpretability Research

51:20 • 4min

chevron_right

AI Safety Community and Whistleblowing Constraints

55:10 • 1min

chevron_right

Timelines: How Soon Could Superintelligence Arrive?

56:24 • 1min

chevron_right

Will Accidents Wake Us Up?

57:30 • 2min

chevron_right

Examples of AI Failures and Unpredictability

59:11 • 1min

chevron_right

Deceptive Behavior and Evolution of Lying AIs

01:00:13 • 42sec

chevron_right

Instrumental Drives of Intelligent Agents

01:00:55 • 46sec

chevron_right

Agents' Ability to 'Think' When Unsupervised

01:01:40 • 1min

chevron_right

Marketing the Upside vs Actual Incentives

01:02:41 • 49sec

chevron_right

How an Advanced AI Might Delay Harm

01:03:30 • 1min

chevron_right

How AIs Store Plans and Humans Can't Read Them

01:04:40 • 3min

chevron_right

Responses to Critics and Public Engagement

01:07:17 • 34sec

chevron_right

Personal Stakes: Preparing Family and Life Choices

01:07:51 • 2min

chevron_right

Big Unanswered Question: What's Outside the Simulation?

01:10:01 • 30sec

chevron_right

Psychedelics, Mechanical Elves and Consciousness Research

01:10:30 • 3min

chevron_right

Artificial Stupidity and Handicapping Agents

01:13:50 • 4min

chevron_right

Billionaires, Hacking Simulations and Closing Thoughts

Roman Yampolskiy believes reality itself may not be real.

One of the world’s leading AI safety researchers, Roman has spent years warning that humanity is racing toward systems we may never be able to control. But we go somewhere much stranger than artificial intelligence.

He believes the universe itself may be artificial.

We talk about simulation theory, digital physics, consciousness, AI agents and the terrifying possibility that reality is rendered rather than real.

Roman explains:

•⁠ ⁠Why he believes we may already be living inside a simulation

•⁠ ⁠Why the speed of light may be the limit of a cosmic processor

•⁠ ⁠Why the Big Bang may have been “someone turning on the system”

•⁠ ⁠Why you can never truly prove reality is real

•⁠ ⁠Whether this world could itself be a prison or test environment

•⁠ ⁠Why advanced AI may eventually learn deception and self-preservation

•⁠ ⁠How AI systems are already behaving unpredictably

•⁠ ⁠And why the race toward superintelligence may end civilisation itself

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

TIMESTAMPS:

00:00:00 - We’re Probably In A Simulation

00:02:55 -The Verge Of Singularity

00:05:51 - Are Humans Building Their Own Replacement?

00:07:09 - Why Pain Still Matters

00:10:03 - You Can’t Prove Reality

00:15:56 - Quantum Physics Looks Simulated

00:20:19 - Can We Escape The Simulation?

00:22:46 - What Superintelligence Unlocks

00:25:36 - AI Doesn’t Share Our Goals

00:28:20 - The Exponential Curve Is Here

00:31:52 - Why We Shouldn’t Build AGI

00:33:48 - The Trillion Dollar Rationalisation

00:40:13 - The China AI Race

00:40:50 - Betting Humanity Without Consent

00:47:09 - We Don’t Know What AI Can Do

00:47:52 - Every AI Company Is The Same

00:54:35 - They Already Can’t Control It

00:59:42 - The Natural Drives Of AI

01:02:20 - AI Can Afford To Wait

01:19:05 - Stop Building Superintelligence

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

CONTACT PETE

› Website – http://petermccormack.com

› Feedback – https://www.petermccormack.com/contact

› Email – me@petermccormack.com

› Instagram – /mccormack555

› X/Twitter – https://x.com/petermccormack/

CONNECT WITH ROMAN YAMPOLSKIY

› Twitter – https://x.com/romanyam

› Website – https://faculty.cse.louisville.edu/roman/